fix ci

add to release flow darwin/amd64
test: expand dsl/runner coverage and add go-test CI workflow
2026-04-18 09:58:03 +03:00 · 2026-04-18 09:55:18 +03:00 · 2026-04-16 00:59:09 +03:00 · 2026-04-16 00:28:30 +03:00 · 2026-04-15 21:24:11 +03:00 · 2026-04-15 11:02:05 +03:00
64 changed files with 16197 additions and 328 deletions
--- a/.gitea/workflows/build-dsl-smoke.yml
+++ b/.gitea/workflows/build-dsl-smoke.yml
@@ -0,0 +1,144 @@
+name: build-dsl-smoke
+run-name: "Structured build DSL smoke test"
+
+on:
+  workflow_dispatch:
+
+env:
+  SUITE_FILE:  sum.jdg
+  EXAMPLE_DIR: example/c-sum-v2
+
+jobs:
+  build_judge:
+    name: Build judge
+    runs-on: Linux-Runner
+    timeout-minutes: 10
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: actions/setup-go@v5
+        with:
+          go-version: '1.22'
+
+      - name: Cross-compile judge
+        shell: bash
+        run: |
+          mkdir -p dist
+          CGO_ENABLED=0 GOOS=linux   GOARCH=amd64 go build -o dist/judge-linux-amd64       ./cmd/cli
+          CGO_ENABLED=0 GOOS=windows GOARCH=amd64 go build -o dist/judge-windows-amd64.exe ./cmd/cli
+          ls -la dist/
+
+      - name: Upload judge binaries
+        uses: https://github.com/christopherHX/gitea-upload-artifact@v4
+        with:
+          name: judge-bin-csum2
+          path: dist/
+          retention-days: 1
+
+  test:
+    needs: build_judge
+    name: "${{ matrix.build }} / ${{ matrix.toolchain }} / ${{ matrix.platform }}"
+
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - { build: release,        toolchain: gcc,   platform: linux   }
+          - { build: release,        toolchain: clang, platform: linux   }
+          - { build: release,        toolchain: clang, platform: windows }
+          - { build: release,        toolchain: msvc,  platform: windows }
+          - { build: debug,          toolchain: gcc,   platform: linux   }
+          - { build: debug,          toolchain: clang, platform: linux   }
+          - { build: debug,          toolchain: clang, platform: windows }
+          - { build: debug,          toolchain: msvc,  platform: windows }
+          - { build: sanitized,      toolchain: gcc,   platform: linux   }
+          - { build: sanitized,      toolchain: clang, platform: linux   }
+          - { build: debug-valgrind, toolchain: gcc,   platform: linux   }
+
+    runs-on: ${{ matrix.platform == 'windows' && 'Windows-Runner' || 'Linux-Runner' }}
+    timeout-minutes: 10
+
+    env:
+      REPORT_NAME: "report_${{ matrix.build }}_${{ matrix.toolchain }}_${{ matrix.platform }}"
+      JUDGE_TOOLCHAIN: ${{ matrix.toolchain }}
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Set up MSVC environment
+        if: matrix.toolchain == 'msvc'
+        uses: ilammy/msvc-dev-cmd@v1
+
+      - name: Download judge binary
+        uses: https://github.com/christopherHX/gitea-download-artifact@v4
+        with:
+          name: judge-bin-csum2
+          path: judge-bin
+
+      - name: Install judge on PATH
+        shell: bash
+        run: |
+          mkdir -p bin
+          if [ "${{ matrix.platform }}" = "windows" ]; then
+            cp judge-bin/judge-windows-amd64.exe bin/judge.exe
+          else
+            cp judge-bin/judge-linux-amd64 bin/judge
+            chmod +x bin/judge
+          fi
+          echo "$PWD/bin" >> "$GITHUB_PATH"
+
+      - name: Run judge
+        shell: bash
+        working-directory: ${{ env.EXAMPLE_DIR }}
+        run: |
+          judge "$SUITE_FILE" . --build=${{ matrix.build }} --json > "$GITHUB_WORKSPACE/${REPORT_NAME}.json" || true
+          cat "$GITHUB_WORKSPACE/${REPORT_NAME}.json"
+          judge "$SUITE_FILE" . --build=${{ matrix.build }}
+
+      - name: Upload report
+        if: ${{ always() }}
+        uses: https://github.com/christopherHX/gitea-upload-artifact@v4
+        with:
+          name: ${{ env.REPORT_NAME }}
+          path: ${{ env.REPORT_NAME }}.json
+          retention-days: 7
+          compression-level: 9
+
+  summary:
+    needs: [build_judge, test]
+    if: ${{ always() }}
+    name: SUMMARY
+    runs-on: Linux-Runner
+    timeout-minutes: 5
+
+    steps:
+      - name: Download judge binary
+        uses: https://github.com/christopherHX/gitea-download-artifact@v4
+        with:
+          name: judge-bin-csum2
+          path: judge-bin
+
+      - name: Install judge on PATH
+        shell: bash
+        run: |
+          mkdir -p bin
+          cp judge-bin/judge-linux-amd64 bin/judge
+          chmod +x bin/judge
+          echo "$PWD/bin" >> "$GITHUB_PATH"
+
+      - name: Download all reports
+        uses: https://github.com/christopherHX/gitea-download-artifact@v4
+        with:
+          path: reports
+          pattern: report_*
+
+      - name: Aggregate
+        shell: bash
+        run: judge aggregate reports | tee SUMMARY.md
+
+      - name: Upload summary
+        uses: https://github.com/christopherHX/gitea-upload-artifact@v4
+        with:
+          name: SUMMARY
+          path: SUMMARY.md
+          retention-days: 7
--- a/.gitea/workflows/go-test.yml
+++ b/.gitea/workflows/go-test.yml
@@ -0,0 +1,36 @@
+name: go-test
+run-name: "Go unit tests"
+
+on:
+  push:
+  pull_request:
+  workflow_dispatch:
+
+jobs:
+  test:
+    name: go test
+    runs-on: Linux-Runner
+    timeout-minutes: 10
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: actions/setup-go@v5
+        with:
+          go-version: '1.26'
+
+      - name: go vet
+        run: go vet ./...
+
+      - name: go test
+        run: go test -race -coverprofile=coverage.out ./...
+
+      - name: Coverage summary
+        run: go tool cover -func=coverage.out | tail -20
+
+      - name: Upload coverage
+        if: ${{ always() }}
+        uses: https://github.com/christopherHX/gitea-upload-artifact@v4
+        with:
+          name: coverage
+          path: coverage.out
+          retention-days: 7
--- a/.gitea/workflows/memory.yml
+++ b/.gitea/workflows/memory.yml
@@ -2,9 +2,6 @@ name: memory-limit
 run-name: "memory_limit enforcement check"

 on:
-  push:
-    branches: [main]
-  pull_request:
  workflow_dispatch:

 env:
--- a/.gitea/workflows/release.yml
+++ b/.gitea/workflows/release.yml
@@ -16,12 +16,12 @@ jobs:
      - name: Set up Go
        uses: actions/setup-go@v5
        with:
-          go-version: '1.22'
+          go-version: "1.22"

      - name: Set up Node.js
        uses: actions/setup-node@v4
        with:
-          node-version: '20'
+          node-version: "20"

      - name: Cross-compile
        shell: bash
@@ -30,6 +30,7 @@ jobs:
          CGO_ENABLED=0 GOOS=linux   GOARCH=amd64 go build -ldflags "-s -w" -o "dist/judge-linux-amd64"       ./cmd/cli
          CGO_ENABLED=0 GOOS=windows GOARCH=amd64 go build -ldflags "-s -w" -o "dist/judge-windows-amd64.exe" ./cmd/cli
          CGO_ENABLED=0 GOOS=darwin  GOARCH=arm64 go build -ldflags "-s -w" -o "dist/judge-darwin-arm64"      ./cmd/cli
+          CGO_ENABLED=0 GOOS=darwin  GOARCH=amd64 go build -ldflags "-s -w" -o "dist/judge-darwin-amd64"      ./cmd/cli

      - name: Build VS Code extension
        shell: bash
--- a/.gitignore
+++ b/.gitignore
@@ -1,2 +1,5 @@
 example/c-sum/solution
 example/solution/solution
+.DS_Store
+*/.DS_Store
+.claude
--- a/cmd/cli/main.go
+++ b/cmd/cli/main.go
@@ -1,6 +1,7 @@
 package main

 import (
+	"encoding/json"
 	"fmt"
 	"os"
 	"strings"
@@ -20,11 +21,16 @@ Flags:
  --json             output as JSON instead of text
  --wrapper <cmd>    exec wrapper (e.g. "valgrind --error-exitcode=99")
  --binary <name>    name of executable produced by build (overrides .jdg)
+  --build <name>     run only the named structured build (use with matrix CI)
+  --list-builds      print the names of structured builds in the suite as JSON
+  --list-matrix      print the full (build, toolchain, platform) matrix as JSON
  --help             show help

 Example:
  judge lab1.jdg ./student-solution
  judge lab1.jdg ./student-solution --json
+  judge lab1.jdg ./student-solution --build=sanitized
+  judge --list-builds lab1.jdg
  judge aggregate reports/
 `

@@ -41,9 +47,20 @@ func main() {
 		return
 	}

+	if hasFlag(args, "--list-builds") {
+		runListBuilds(args)
+		return
+	}
+
+	if hasFlag(args, "--list-matrix") {
+		runListMatrix(args)
+		return
+	}
+
 	jsonOutput := hasFlag(args, "--json")
 	wrapper := flagValue(args, "--wrapper")
 	binary := flagValue(args, "--binary")
+	targetBuild := flagValue(args, "--build")
 	positional := positionalArgs(args)

 	if len(positional) < 2 {
@@ -54,27 +71,17 @@ func main() {
 	testFile := positional[0]
 	solutionDir := positional[1]

-	src, err := os.ReadFile(testFile)
-	if err != nil {
-		fatalf("cannot read %q: %v", testFile, err)
-	}
-
-	f, warns, err := dsl.Parse(string(src))
-	if err != nil {
-		fatalf("parse error in %q:\n  %v", testFile, err)
-	}
-	for _, w := range warns {
-		fmt.Fprintf(os.Stderr, "warning: %s\n", w)
-	}
+	f := parseSuite(testFile)

 	if _, err := os.Stat(solutionDir); err != nil {
 		fatalf("solution dir %q not found: %v", solutionDir, err)
 	}

 	r := runner.New(f, runner.Config{
-		WorkDir:    solutionDir,
-		BinaryName: binary,
-		Wrapper:    wrapper,
+		WorkDir:     solutionDir,
+		BinaryName:  binary,
+		Wrapper:     wrapper,
+		TargetBuild: targetBuild,
 	})
 	result := r.Run()

@@ -91,6 +98,91 @@ func main() {
 	}
 }

+func parseSuite(path string) *dsl.File {
+	f, warns, err := dsl.ParseFile(path)
+	if err != nil {
+		fatalf("parse error in %q:\n  %v", path, err)
+	}
+	for _, w := range warns {
+		fmt.Fprintf(os.Stderr, "warning: %s\n", w)
+	}
+	return f
+}
+
+func runListBuilds(args []string) {
+	positional := positionalArgs(args)
+	if len(positional) < 1 {
+		fatalf("--list-builds requires the path to a .jdg file")
+	}
+	f := parseSuite(positional[0])
+
+	var names []string
+	if len(f.Builds) == 0 {
+		names = []string{"default"}
+	} else {
+		for _, b := range f.Builds {
+			names = append(names, b.Name)
+		}
+	}
+	enc := json.NewEncoder(os.Stdout)
+	if err := enc.Encode(names); err != nil {
+		fatalf("encode list-builds: %v", err)
+	}
+}
+
+type matrixEntry struct {
+	Build     string `json:"build"`
+	Toolchain string `json:"toolchain"`
+	Platform  string `json:"platform"`
+	Wrapper   string `json:"wrapper,omitempty"`
+}
+
+func runListMatrix(args []string) {
+	positional := positionalArgs(args)
+	if len(positional) < 1 {
+		fatalf("--list-matrix requires the path to a .jdg file")
+	}
+	f := parseSuite(positional[0])
+
+	var entries []matrixEntry
+
+	if len(f.Builds) == 0 {
+		if len(f.Toolchains) == 0 {
+			entries = append(entries, matrixEntry{Build: "default", Toolchain: "default", Platform: "linux"})
+		} else {
+			for _, tc := range f.Toolchains {
+				for _, platform := range tc.Platforms {
+					entries = append(entries, matrixEntry{Build: "default", Toolchain: tc.Name, Platform: platform})
+				}
+			}
+		}
+	} else if len(f.Toolchains) == 0 {
+		fatalf("suite has structured builds but no `toolchains { ... }` block; add one to use --list-matrix")
+	} else {
+		for _, b := range f.Builds {
+			for _, tc := range f.Toolchains {
+				for _, platform := range tc.Platforms {
+					eff := b.Resolve(f.BuildDefaults, platform)
+					if !eff.AppliesTo(platform, tc.Name) {
+						continue
+					}
+					entries = append(entries, matrixEntry{
+						Build:     b.Name,
+						Toolchain: tc.Name,
+						Platform:  platform,
+						Wrapper:   eff.Wrapper,
+					})
+				}
+			}
+		}
+	}
+
+	enc := json.NewEncoder(os.Stdout)
+	if err := enc.Encode(entries); err != nil {
+		fatalf("encode list-matrix: %v", err)
+	}
+}
+
 func runAggregate(args []string) {
 	if len(args) < 1 {
 		fatalf("usage: judge aggregate <reports-dir>")
@@ -128,8 +220,18 @@ func flagValue(args []string, name string) string {
 }

 func positionalArgs(args []string) []string {
-	known := map[string]bool{"--json": true, "--help": true, "-h": true}
-	withValue := map[string]bool{"--wrapper": true, "--binary": true}
+	known := map[string]bool{
+		"--json":        true,
+		"--help":        true,
+		"-h":            true,
+		"--list-builds": true,
+		"--list-matrix": true,
+	}
+	withValue := map[string]bool{
+		"--wrapper": true,
+		"--binary":  true,
+		"--build":   true,
+	}

 	var out []string
 	skip := false
@@ -145,7 +247,7 @@ func positionalArgs(args []string) []string {
 			skip = true
 			continue
 		}
-		if strings.HasPrefix(a, "--wrapper=") || strings.HasPrefix(a, "--binary=") {
+		if strings.HasPrefix(a, "--wrapper=") || strings.HasPrefix(a, "--binary=") || strings.HasPrefix(a, "--build=") {
 			continue
 		}
 		out = append(out, a)
--- a/dsl/ast.go
+++ b/dsl/ast.go
@@ -1,19 +1,26 @@
 package dsl

-import "time"
+import (
+	"time"
+)

 type File struct {
 	Build        string
 	BuildLinux   string
 	BuildWindows string
 	BuildDarwin  string
-	Timeout      time.Duration
-	MemoryLimit  int64  // bytes; 0 means no limit
-	Binary       string // executable name produced by build (default: solution)
-	Sources      string // glob pattern for source files, expanded as $SOURCES in build

-	NormalizeCRLF  bool // strip \r before matching stdout/stderr/outFiles
-	TrimTrailingWS bool // trim trailing whitespace on each line before matching
+	BuildDefaults *BuildConfig
+	Builds        []*BuildConfig
+	Toolchains    []*ToolchainSpec
+
+	Timeout     time.Duration
+	MemoryLimit int64
+	Binary      string
+	Sources     string
+
+	NormalizeCRLF  bool
+	TrimTrailingWS bool

 	Groups []*Group
 }
@@ -25,7 +32,7 @@ type Group struct {
 	MemoryLimit int64
 	Env         map[string]string
 	Scoring     ScoringMode
-	Wrapper     string // exec wrapper command (e.g., "valgrind --error-exitcode=1")
+	Wrapper     string

 	Tests   []*Test
 	Pattern *Pattern
@@ -34,8 +41,8 @@ type Group struct {
 type ScoringMode int

 const (
-	ScoringPartial   ScoringMode = iota // weight * passed/total (default)
-	ScoringAllOrNone                    // weight or 0
+	ScoringPartial ScoringMode = iota
+	ScoringAllOrNone
 )

 type Pattern struct {
@@ -45,6 +52,8 @@ type Pattern struct {
 	DirsGlob   string
 	InputFile  string
 	OutputFile string
+
+	Args []string
 }

 func (p *Pattern) IsDirMode() bool {
@@ -68,41 +77,20 @@ type Test struct {
 	OutFiles map[string]string
 }

-type Matcher interface {
-	matcherNode()
+func (t *Test) SetInputFile(inputName string, inputContent []byte) {
+	t.InFiles[inputName] = string(inputContent)
 }

-type ExactMatcher struct {
-	Value string
+func (t *Test) SetStdin(inputContent []byte) {
+	s := string(inputContent)
+	t.Stdin = &s
 }

-func (ExactMatcher) matcherNode() {}
-
-type ContainsMatcher struct {
-	Substr string
+func (t *Test) SetOutputFile(outputName string, outputContent []byte) {
+	t.OutFiles[outputName] = string(outputContent)
+	t.Stdout = NoMatcher{}
 }

-func (ContainsMatcher) matcherNode() {}
-
-type RegexMatcher struct {
-	Pattern string
+func (t *Test) SetStdout(outputContent []byte) {
+	t.Stdout = ExactMatcher{Value: string(outputContent)}
 }
-
-func (RegexMatcher) matcherNode() {}
-
-type NumericEpsMatcher struct {
-	Epsilon float64
-	Value   string
-}
-
-func (NumericEpsMatcher) matcherNode() {}
-
-type AnyOrderMatcher struct {
-	Lines []string
-}
-
-func (AnyOrderMatcher) matcherNode() {}
-
-type NoMatcher struct{}
-
-func (NoMatcher) matcherNode() {}
--- a/dsl/ast_test.go
+++ b/dsl/ast_test.go
@@ -0,0 +1,75 @@
+package dsl
+
+import "testing"
+
+func newTest() *Test {
+	return &Test{
+		InFiles:  map[string]string{},
+		OutFiles: map[string]string{},
+	}
+}
+
+func TestSetInputFile(t *testing.T) {
+	tst := newTest()
+	tst.SetInputFile("input.txt", []byte("hello"))
+	if got := tst.InFiles["input.txt"]; got != "hello" {
+		t.Errorf("InFiles[input.txt] = %q, want %q", got, "hello")
+	}
+	if tst.Stdin != nil {
+		t.Errorf("Stdin should remain nil, got %v", *tst.Stdin)
+	}
+}
+
+func TestSetStdin(t *testing.T) {
+	tst := newTest()
+	tst.SetStdin([]byte("data\n"))
+	if tst.Stdin == nil {
+		t.Fatal("Stdin is nil")
+	}
+	if *tst.Stdin != "data\n" {
+		t.Errorf("Stdin = %q, want %q", *tst.Stdin, "data\n")
+	}
+	if len(tst.InFiles) != 0 {
+		t.Errorf("InFiles should be empty, got %v", tst.InFiles)
+	}
+}
+
+func TestSetOutputFileSetsNoMatcher(t *testing.T) {
+	tst := newTest()
+	tst.SetOutputFile("out.txt", []byte("result"))
+	if got := tst.OutFiles["out.txt"]; got != "result" {
+		t.Errorf("OutFiles[out.txt] = %q, want %q", got, "result")
+	}
+	if _, ok := tst.Stdout.(NoMatcher); !ok {
+		t.Errorf("Stdout = %T, want NoMatcher", tst.Stdout)
+	}
+}
+
+func TestSetStdoutSetsExactMatcher(t *testing.T) {
+	tst := newTest()
+	tst.SetStdout([]byte("expected\n"))
+	m, ok := tst.Stdout.(ExactMatcher)
+	if !ok {
+		t.Fatalf("Stdout = %T, want ExactMatcher", tst.Stdout)
+	}
+	if m.Value != "expected\n" {
+		t.Errorf("ExactMatcher.Value = %q, want %q", m.Value, "expected\n")
+	}
+	if len(tst.OutFiles) != 0 {
+		t.Errorf("OutFiles should be empty, got %v", tst.OutFiles)
+	}
+}
+
+func TestIsDirModeTrue(t *testing.T) {
+	p := &Pattern{DirsGlob: "tests/*"}
+	if !p.IsDirMode() {
+		t.Error("IsDirMode() = false, want true")
+	}
+}
+
+func TestIsDirModeFalse(t *testing.T) {
+	p := &Pattern{InputGlob: "tests/*.in"}
+	if p.IsDirMode() {
+		t.Error("IsDirMode() = true, want false")
+	}
+}
--- a/dsl/build.go
+++ b/dsl/build.go
@@ -0,0 +1,152 @@
+package dsl
+
+import (
+	"maps"
+	"slices"
+)
+
+type BuildProfile int
+
+const (
+	ProfileUnset BuildProfile = iota
+	ProfileRelease
+	ProfileDebug
+	ProfileSanitized
+)
+
+func (p BuildProfile) String() string {
+	switch p {
+	case ProfileRelease:
+		return "release"
+	case ProfileDebug:
+		return "debug"
+	case ProfileSanitized:
+		return "sanitized"
+	default:
+		return "unset"
+	}
+}
+
+type WarningLevel int
+
+const (
+	WarningsUnset WarningLevel = iota
+	WarningsDefault
+	WarningsStrict
+	WarningsPedantic
+)
+
+func (w WarningLevel) String() string {
+	switch w {
+	case WarningsDefault:
+		return "default"
+	case WarningsStrict:
+		return "strict"
+	case WarningsPedantic:
+		return "pedantic"
+	default:
+		return "unset"
+	}
+}
+
+type BuildConfig struct {
+	Name string
+
+	Language string
+	Standard string
+	Sources  []string
+	Includes []string
+	Output   string
+
+	Profile  BuildProfile
+	Warnings WarningLevel
+	Sanitize []string
+	Wrapper  string
+
+	Defines map[string]string
+	Link    []string
+	Extra   []string
+
+	Platforms []string
+	Compilers []string
+
+	Linux   *BuildConfig
+	Windows *BuildConfig
+	Darwin  *BuildConfig
+}
+
+func (dst *BuildConfig) MergeFrom(src *BuildConfig) {
+	if src == nil {
+		return
+	}
+	if src.Language != "" {
+		dst.Language = src.Language
+	}
+	if src.Standard != "" {
+		dst.Standard = src.Standard
+	}
+	if src.Output != "" {
+		dst.Output = src.Output
+	}
+	if src.Profile != ProfileUnset {
+		dst.Profile = src.Profile
+	}
+	if src.Warnings != WarningsUnset {
+		dst.Warnings = src.Warnings
+	}
+	if src.Wrapper != "" {
+		dst.Wrapper = src.Wrapper
+	}
+
+	dst.Sources = append(dst.Sources, src.Sources...)
+	dst.Includes = append(dst.Includes, src.Includes...)
+	dst.Sanitize = append(dst.Sanitize, src.Sanitize...)
+	dst.Link = append(dst.Link, src.Link...)
+	dst.Extra = append(dst.Extra, src.Extra...)
+	dst.Platforms = append(dst.Platforms, src.Platforms...)
+	dst.Compilers = append(dst.Compilers, src.Compilers...)
+
+	if len(src.Defines) > 0 {
+		if dst.Defines == nil {
+			dst.Defines = map[string]string{}
+		}
+		maps.Copy(dst.Defines, src.Defines)
+	}
+}
+
+func (b *BuildConfig) Resolve(defaults *BuildConfig, os string) BuildConfig {
+	var out BuildConfig
+	out.MergeFrom(defaults)
+	out.MergeFrom(b)
+	out.Name = b.Name
+
+	var osOverride *BuildConfig
+	switch os {
+	case "linux":
+		osOverride = b.Linux
+	case "windows":
+		osOverride = b.Windows
+	case "darwin":
+		osOverride = b.Darwin
+	}
+	out.MergeFrom(osOverride)
+
+	return out
+}
+
+func (b *BuildConfig) AppliesTo(os, compiler string) bool {
+	if len(b.Platforms) > 0 && !slices.Contains(b.Platforms, os) {
+		return false
+	}
+	if len(b.Compilers) > 0 && !slices.Contains(b.Compilers, compiler) {
+		return false
+	}
+	return true
+}
+
+type ToolchainSpec struct {
+	Name      string
+	Platforms []string
+	Binary    string
+	Class     string
+}
--- a/dsl/build_parser.go
+++ b/dsl/build_parser.go
@@ -0,0 +1,355 @@
+package dsl
+
+import "fmt"
+
+func (p *Parser) parseBuildBlock(name string) (*BuildConfig, error) {
+	return p.parseBuildBlockInner(name, false)
+}
+
+func (p *Parser) parseBuildBlockInner(name string, inOSOverride bool) (*BuildConfig, error) {
+	if _, err := p.expect(TOKEN_LBRACE); err != nil {
+		return nil, err
+	}
+
+	bc := &BuildConfig{Name: name}
+
+	for !p.isRBrace() {
+		t := p.peek()
+		if t.Type != TOKEN_IDENT {
+			return nil, fmt.Errorf("%d:%d: unexpected token %q in build block", t.Line, t.Col, t.Value)
+		}
+
+		switch t.Value {
+		case "language":
+			p.advance()
+			s, err := p.parseAssignString()
+			if err != nil {
+				return nil, err
+			}
+			bc.Language = s
+
+		case "standard":
+			p.advance()
+			s, err := p.parseAssignString()
+			if err != nil {
+				return nil, err
+			}
+			bc.Standard = s
+
+		case "output":
+			p.advance()
+			s, err := p.parseAssignString()
+			if err != nil {
+				return nil, err
+			}
+			bc.Output = s
+
+		case "wrapper":
+			p.advance()
+			s, err := p.parseAssignString()
+			if err != nil {
+				return nil, err
+			}
+			bc.Wrapper = s
+
+		case "sources":
+			p.advance()
+			xs, err := p.parseAssignStringList()
+			if err != nil {
+				return nil, err
+			}
+			bc.Sources = xs
+
+		case "includes":
+			p.advance()
+			xs, err := p.parseAssignStringList()
+			if err != nil {
+				return nil, err
+			}
+			bc.Includes = xs
+
+		case "sanitize":
+			p.advance()
+			xs, err := p.parseAssignStringList()
+			if err != nil {
+				return nil, err
+			}
+			bc.Sanitize = xs
+
+		case "link":
+			p.advance()
+			xs, err := p.parseAssignStringList()
+			if err != nil {
+				return nil, err
+			}
+			bc.Link = xs
+
+		case "extra":
+			p.advance()
+			xs, err := p.parseAssignStringList()
+			if err != nil {
+				return nil, err
+			}
+			bc.Extra = xs
+
+		case "platforms":
+			p.advance()
+			xs, err := p.parseAssignStringList()
+			if err != nil {
+				return nil, err
+			}
+			if err := validatePlatformList(xs, t.Line, t.Col); err != nil {
+				return nil, err
+			}
+			bc.Platforms = xs
+
+		case "compilers":
+			p.advance()
+			xs, err := p.parseAssignStringList()
+			if err != nil {
+				return nil, err
+			}
+			bc.Compilers = xs
+
+		case "profile":
+			p.advance()
+			if _, err := p.expect(TOKEN_ASSIGN); err != nil {
+				return nil, err
+			}
+			id, err := p.expect(TOKEN_IDENT)
+			if err != nil {
+				return nil, err
+			}
+			prof, err := parseProfileIdent(id.Value, id.Line, id.Col)
+			if err != nil {
+				return nil, err
+			}
+			bc.Profile = prof
+
+		case "warnings":
+			p.advance()
+			if _, err := p.expect(TOKEN_ASSIGN); err != nil {
+				return nil, err
+			}
+			id, err := p.expect(TOKEN_IDENT)
+			if err != nil {
+				return nil, err
+			}
+			w, err := parseWarningsIdent(id.Value, id.Line, id.Col)
+			if err != nil {
+				return nil, err
+			}
+			bc.Warnings = w
+
+		case "define":
+			p.advance()
+			if _, err := p.expect(TOKEN_LPAREN); err != nil {
+				return nil, err
+			}
+			key, err := p.expect(TOKEN_STRING)
+			if err != nil {
+				return nil, err
+			}
+			if _, err := p.expect(TOKEN_RPAREN); err != nil {
+				return nil, err
+			}
+			if _, err := p.expect(TOKEN_ASSIGN); err != nil {
+				return nil, err
+			}
+			val, err := p.expect(TOKEN_STRING)
+			if err != nil {
+				return nil, err
+			}
+			if bc.Defines == nil {
+				bc.Defines = map[string]string{}
+			}
+			bc.Defines[key.Value] = val.Value
+
+		case "linux", "windows", "darwin":
+			if inOSOverride {
+				return nil, fmt.Errorf("%d:%d: OS override %q cannot be nested inside another OS override", t.Line, t.Col, t.Value)
+			}
+			osName := t.Value
+			p.advance()
+			sub, err := p.parseBuildBlockInner("", true)
+			if err != nil {
+				return nil, err
+			}
+			switch osName {
+			case "linux":
+				if bc.Linux != nil {
+					return nil, fmt.Errorf("%d:%d: duplicate linux override", t.Line, t.Col)
+				}
+				bc.Linux = sub
+			case "windows":
+				if bc.Windows != nil {
+					return nil, fmt.Errorf("%d:%d: duplicate windows override", t.Line, t.Col)
+				}
+				bc.Windows = sub
+			case "darwin":
+				if bc.Darwin != nil {
+					return nil, fmt.Errorf("%d:%d: duplicate darwin override", t.Line, t.Col)
+				}
+				bc.Darwin = sub
+			}
+
+		default:
+			return nil, fmt.Errorf("%d:%d: unknown field %q in build block", t.Line, t.Col, t.Value)
+		}
+	}
+
+	if _, err := p.expect(TOKEN_RBRACE); err != nil {
+		return nil, err
+	}
+	return bc, nil
+}
+
+func (p *Parser) parseAssignString() (string, error) {
+	if _, err := p.expect(TOKEN_ASSIGN); err != nil {
+		return "", err
+	}
+	s, err := p.expect(TOKEN_STRING)
+	if err != nil {
+		return "", err
+	}
+	return s.Value, nil
+}
+
+func (p *Parser) parseAssignStringList() ([]string, error) {
+	if _, err := p.expect(TOKEN_ASSIGN); err != nil {
+		return nil, err
+	}
+	return p.parseStringList()
+}
+
+func parseProfileIdent(v string, line, col int) (BuildProfile, error) {
+	switch v {
+	case "release":
+		return ProfileRelease, nil
+	case "debug":
+		return ProfileDebug, nil
+	case "sanitized":
+		return ProfileSanitized, nil
+	default:
+		return ProfileUnset, fmt.Errorf("%d:%d: unknown profile %q (expected release/debug/sanitized)", line, col, v)
+	}
+}
+
+func parseWarningsIdent(v string, line, col int) (WarningLevel, error) {
+	switch v {
+	case "default":
+		return WarningsDefault, nil
+	case "strict":
+		return WarningsStrict, nil
+	case "pedantic":
+		return WarningsPedantic, nil
+	default:
+		return WarningsUnset, fmt.Errorf("%d:%d: unknown warnings level %q (expected default/strict/pedantic)", line, col, v)
+	}
+}
+
+func validatePlatformList(xs []string, line, col int) error {
+	for _, x := range xs {
+		switch x {
+		case "linux", "windows", "darwin":
+		default:
+			return fmt.Errorf("%d:%d: unknown platform %q (expected linux/windows/darwin)", line, col, x)
+		}
+	}
+	return nil
+}
+
+func (p *Parser) parseToolchainsBlock() ([]*ToolchainSpec, error) {
+	if _, err := p.expect(TOKEN_LBRACE); err != nil {
+		return nil, err
+	}
+
+	var specs []*ToolchainSpec
+	seen := map[string]bool{}
+
+	for !p.isRBrace() {
+		nameTok := p.peek()
+		if nameTok.Type != TOKEN_IDENT && nameTok.Type != TOKEN_STRING {
+			return nil, fmt.Errorf("%d:%d: expected toolchain name, got %q", nameTok.Line, nameTok.Col, nameTok.Value)
+		}
+		p.advance()
+		name := nameTok.Value
+		if seen[name] {
+			return nil, fmt.Errorf("%d:%d: duplicate toolchain %q", nameTok.Line, nameTok.Col, name)
+		}
+		seen[name] = true
+
+		spec, err := p.parseToolchainEntry(name)
+		if err != nil {
+			return nil, err
+		}
+		specs = append(specs, spec)
+	}
+
+	if _, err := p.expect(TOKEN_RBRACE); err != nil {
+		return nil, err
+	}
+	return specs, nil
+}
+
+func (p *Parser) parseToolchainEntry(name string) (*ToolchainSpec, error) {
+	if _, err := p.expect(TOKEN_LBRACE); err != nil {
+		return nil, err
+	}
+
+	spec := &ToolchainSpec{Name: name}
+	for !p.isRBrace() {
+		t := p.peek()
+		if t.Type != TOKEN_IDENT {
+			return nil, fmt.Errorf("%d:%d: unexpected token %q in toolchain block", t.Line, t.Col, t.Value)
+		}
+		switch t.Value {
+		case "platforms":
+			p.advance()
+			xs, err := p.parseAssignStringList()
+			if err != nil {
+				return nil, err
+			}
+			if err := validatePlatformList(xs, t.Line, t.Col); err != nil {
+				return nil, err
+			}
+			spec.Platforms = xs
+
+		case "binary":
+			p.advance()
+			s, err := p.parseAssignString()
+			if err != nil {
+				return nil, err
+			}
+			spec.Binary = s
+
+		case "class":
+			p.advance()
+			if _, err := p.expect(TOKEN_ASSIGN); err != nil {
+				return nil, err
+			}
+			id, err := p.expect(TOKEN_IDENT)
+			if err != nil {
+				return nil, err
+			}
+			switch id.Value {
+			case "gnu", "msvc":
+			default:
+				return nil, fmt.Errorf("%d:%d: unknown compiler class %q (expected gnu/msvc)", id.Line, id.Col, id.Value)
+			}
+			spec.Class = id.Value
+
+		default:
+			return nil, fmt.Errorf("%d:%d: unknown field %q in toolchain block", t.Line, t.Col, t.Value)
+		}
+	}
+
+	if _, err := p.expect(TOKEN_RBRACE); err != nil {
+		return nil, err
+	}
+
+	if len(spec.Platforms) == 0 {
+		return nil, fmt.Errorf("toolchain %q: platforms is required", name)
+	}
+	return spec, nil
+}
--- a/dsl/build_parser_test.go
+++ b/dsl/build_parser_test.go
@@ -0,0 +1,379 @@
+package dsl
+
+import (
+	"strings"
+	"testing"
+)
+
+const buildDefaultsPrefix = `
+build_defaults {
+    language = "c"
+    standard = "c11"
+    sources  = "*.c"
+    output   = "sol"
+    warnings = strict
+}
+`
+
+func TestBuildAllFieldsParsed(t *testing.T) {
+	src := `
+build_defaults {
+    language = "cpp"
+    standard = "c++17"
+    sources  = "a.cpp" "b.cpp"
+    includes = "inc" "inc2"
+    output   = "sol"
+    warnings = pedantic
+    wrapper  = "timeout"
+    sanitize = "address" "undefined"
+    link     = "-lm"
+    extra    = "-g"
+    platforms = "linux" "darwin"
+    compilers = "gcc" "clang"
+    define("DEBUG") = "1"
+    define("VERSION") = "2"
+}
+
+build "release" {
+    profile = release
+}
+
+group("g") { weight = 1.0 test("t") { stdout = "" } }
+`
+	f, _, err := Parse(src)
+	if err != nil {
+		t.Fatalf("parse: %v", err)
+	}
+	bd := f.BuildDefaults
+	if bd == nil {
+		t.Fatal("no build_defaults")
+	}
+	if bd.Language != "cpp" || bd.Standard != "c++17" || bd.Output != "sol" {
+		t.Errorf("basic fields wrong: %+v", bd)
+	}
+	if bd.Warnings != WarningsPedantic {
+		t.Errorf("Warnings = %v", bd.Warnings)
+	}
+	if bd.Wrapper != "timeout" {
+		t.Errorf("Wrapper = %q", bd.Wrapper)
+	}
+	if len(bd.Sources) != 2 || len(bd.Includes) != 2 || len(bd.Sanitize) != 2 {
+		t.Errorf("lists: sources=%v includes=%v sanitize=%v", bd.Sources, bd.Includes, bd.Sanitize)
+	}
+	if len(bd.Link) != 1 || len(bd.Extra) != 1 {
+		t.Errorf("link/extra: %v / %v", bd.Link, bd.Extra)
+	}
+	if len(bd.Platforms) != 2 || len(bd.Compilers) != 2 {
+		t.Errorf("platforms/compilers: %v / %v", bd.Platforms, bd.Compilers)
+	}
+	if bd.Defines["DEBUG"] != "1" || bd.Defines["VERSION"] != "2" {
+		t.Errorf("Defines = %v", bd.Defines)
+	}
+}
+
+func TestBuildOSOverrides(t *testing.T) {
+	src := `
+build_defaults {
+    language = "c"
+    standard = "c11"
+    sources  = "main.c"
+    output   = "sol"
+    warnings = strict
+}
+
+build "release" {
+    profile = release
+    linux   { extra = "-pthread" }
+    windows { extra = "/MT" }
+    darwin  { extra = "-framework CoreFoundation" }
+}
+
+group("g") { weight = 1.0 test("t") { stdout = "" } }
+`
+	f, _, err := Parse(src)
+	if err != nil {
+		t.Fatalf("parse: %v", err)
+	}
+	b := f.Builds[0]
+	if b.Linux == nil || b.Windows == nil || b.Darwin == nil {
+		t.Errorf("OS overrides missing: %+v", b)
+	}
+}
+
+func TestBuildOSNestedForbidden(t *testing.T) {
+	src := buildDefaultsPrefix + `
+build "r" {
+    profile = release
+    linux {
+        darwin { extra = "-x" }
+    }
+}
+group("g") { weight = 1.0 test("t") { stdout = "" } }
+`
+	_, _, err := Parse(src)
+	if err == nil || !strings.Contains(err.Error(), "nested inside another OS override") {
+		t.Errorf("want nested OS override error, got %v", err)
+	}
+}
+
+func TestBuildDuplicateOSOverride(t *testing.T) {
+	src := buildDefaultsPrefix + `
+build "r" {
+    profile = release
+    linux { extra = "-a" }
+    linux { extra = "-b" }
+}
+group("g") { weight = 1.0 test("t") { stdout = "" } }
+`
+	_, _, err := Parse(src)
+	if err == nil || !strings.Contains(err.Error(), "duplicate linux override") {
+		t.Errorf("want duplicate linux override, got %v", err)
+	}
+}
+
+func TestBuildUnknownField(t *testing.T) {
+	src := `
+build_defaults {
+    bogus = "x"
+    language = "c"
+    standard = "c11"
+    sources  = "main.c"
+    output   = "sol"
+    warnings = strict
+}
+group("g") { weight = 1.0 test("t") { stdout = "" } }
+`
+	_, _, err := Parse(src)
+	if err == nil || !strings.Contains(err.Error(), "unknown field") {
+		t.Errorf("want unknown field error, got %v", err)
+	}
+}
+
+func TestBuildUnknownProfile(t *testing.T) {
+	src := `
+build_defaults {
+    language = "c"
+    standard = "c11"
+    sources  = "main.c"
+    output   = "sol"
+    warnings = strict
+}
+build "weird" { profile = bogus }
+group("g") { weight = 1.0 test("t") { stdout = "" } }
+`
+	_, _, err := Parse(src)
+	if err == nil || !strings.Contains(err.Error(), "unknown profile") {
+		t.Errorf("want unknown profile error, got %v", err)
+	}
+}
+
+func TestBuildUnknownWarnings(t *testing.T) {
+	src := `
+build_defaults {
+    language = "c"
+    standard = "c11"
+    sources  = "main.c"
+    output   = "sol"
+    warnings = bogus
+}
+group("g") { weight = 1.0 test("t") { stdout = "" } }
+`
+	_, _, err := Parse(src)
+	if err == nil || !strings.Contains(err.Error(), "unknown warnings level") {
+		t.Errorf("want unknown warnings error, got %v", err)
+	}
+}
+
+func TestBuildWarningsDefaultAndStrict(t *testing.T) {
+	src := `
+build_defaults {
+    language = "c"
+    standard = "c11"
+    sources  = "main.c"
+    output   = "sol"
+    warnings = default
+}
+build "r" {
+    warnings = strict
+}
+group("g") { weight = 1.0 test("t") { stdout = "" } }
+`
+	f, _, err := Parse(src)
+	if err != nil {
+		t.Fatalf("parse: %v", err)
+	}
+	if f.BuildDefaults.Warnings != WarningsDefault {
+		t.Errorf("defaults warnings = %v", f.BuildDefaults.Warnings)
+	}
+	if f.Builds[0].Warnings != WarningsStrict {
+		t.Errorf("build warnings = %v", f.Builds[0].Warnings)
+	}
+}
+
+func TestBuildInvalidPlatform(t *testing.T) {
+	src := buildDefaultsPrefix + `
+build "r" {
+    profile = release
+    platforms = "bsd"
+}
+group("g") { weight = 1.0 test("t") { stdout = "" } }
+`
+	_, _, err := Parse(src)
+	if err == nil || !strings.Contains(err.Error(), "unknown platform") {
+		t.Errorf("want unknown platform error, got %v", err)
+	}
+}
+
+func TestBuildAssignStringMissingAssign(t *testing.T) {
+	src := `
+build_defaults {
+    language "c"
+    standard = "c11"
+    sources  = "main.c"
+    output   = "sol"
+    warnings = strict
+}
+group("g") { weight = 1.0 test("t") { stdout = "" } }
+`
+	_, _, err := Parse(src)
+	if err == nil || !strings.Contains(err.Error(), "expected") {
+		t.Errorf("want expected = error, got %v", err)
+	}
+}
+
+func TestBuildAssignStringListMissingAssign(t *testing.T) {
+	src := `
+build_defaults {
+    language = "c"
+    standard = "c11"
+    sources  "main.c"
+    output   = "sol"
+    warnings = strict
+}
+group("g") { weight = 1.0 test("t") { stdout = "" } }
+`
+	_, _, err := Parse(src)
+	if err == nil || !strings.Contains(err.Error(), "expected") {
+		t.Errorf("want expected = error, got %v", err)
+	}
+}
+
+func TestBuildDefineErrors(t *testing.T) {
+	cases := []string{
+		`build_defaults { language="c" standard="c11" sources="x.c" output="s" warnings=strict define "K" = "v" }`,
+		`build_defaults { language="c" standard="c11" sources="x.c" output="s" warnings=strict define(K) = "v" }`,
+		`build_defaults { language="c" standard="c11" sources="x.c" output="s" warnings=strict define("K" = "v" }`,
+		`build_defaults { language="c" standard="c11" sources="x.c" output="s" warnings=strict define("K") "v" }`,
+		`build_defaults { language="c" standard="c11" sources="x.c" output="s" warnings=strict define("K") = K }`,
+	}
+	for i, src := range cases {
+		full := src + "\ngroup(\"g\") { weight = 1.0 test(\"t\") { stdout = \"\" } }"
+		if _, _, err := Parse(full); err == nil {
+			t.Errorf("case %d: expected error", i)
+		}
+	}
+}
+
+func TestBuildNonIdentInBlock(t *testing.T) {
+	src := `
+build_defaults { "not-an-ident" = "x" }
+group("g") { weight = 1.0 test("t") { stdout = "" } }
+`
+	_, _, err := Parse(src)
+	if err == nil || !strings.Contains(err.Error(), "unexpected token") {
+		t.Errorf("want unexpected token error, got %v", err)
+	}
+}
+
+func TestToolchainsDuplicateInSameBlock(t *testing.T) {
+	src := `
+toolchains {
+    gcc { platforms = "linux" }
+    gcc { platforms = "darwin" }
+}
+build_defaults { language="c" standard="c11" sources="x.c" output="s" warnings=strict }
+group("g") { weight = 1.0 test("t") { stdout = "" } }
+`
+	_, _, err := Parse(src)
+	if err == nil || !strings.Contains(err.Error(), "duplicate toolchain") {
+		t.Errorf("want duplicate toolchain error, got %v", err)
+	}
+}
+
+func TestToolchainsMissingPlatforms(t *testing.T) {
+	src := `
+toolchains {
+    gcc { binary = "gcc-13" }
+}
+build_defaults { language="c" standard="c11" sources="x.c" output="s" warnings=strict }
+group("g") { weight = 1.0 test("t") { stdout = "" } }
+`
+	_, _, err := Parse(src)
+	if err == nil || !strings.Contains(err.Error(), "platforms is required") {
+		t.Errorf("want platforms-required error, got %v", err)
+	}
+}
+
+func TestToolchainsBadName(t *testing.T) {
+	src := `
+toolchains {
+    42 { platforms = "linux" }
+}
+`
+	_, _, err := Parse(src)
+	if err == nil || !strings.Contains(err.Error(), "expected toolchain name") {
+		t.Errorf("want toolchain name error, got %v", err)
+	}
+}
+
+func TestToolchainsUnknownField(t *testing.T) {
+	src := `
+toolchains {
+    gcc { platforms = "linux" bogus = "x" }
+}
+build_defaults { language="c" standard="c11" sources="x.c" output="s" warnings=strict }
+group("g") { weight = 1.0 test("t") { stdout = "" } }
+`
+	_, _, err := Parse(src)
+	if err == nil || !strings.Contains(err.Error(), "unknown field") {
+		t.Errorf("want unknown field error, got %v", err)
+	}
+}
+
+func TestToolchainsUnknownClass(t *testing.T) {
+	src := `
+toolchains {
+    gcc { platforms = "linux" class = bogus }
+}
+build_defaults { language="c" standard="c11" sources="x.c" output="s" warnings=strict }
+group("g") { weight = 1.0 test("t") { stdout = "" } }
+`
+	_, _, err := Parse(src)
+	if err == nil || !strings.Contains(err.Error(), "unknown compiler class") {
+		t.Errorf("want unknown class error, got %v", err)
+	}
+}
+
+func TestToolchainsClassAndBinary(t *testing.T) {
+	src := `
+toolchains {
+    gcc13 { platforms = "linux" binary = "gcc-13" class = gnu }
+    msvc { platforms = "windows" class = msvc }
+}
+build_defaults { language="c" standard="c11" sources="x.c" output="s" warnings=strict }
+group("g") { weight = 1.0 test("t") { stdout = "" } }
+`
+	f, _, err := Parse(src)
+	if err != nil {
+		t.Fatalf("parse: %v", err)
+	}
+	if len(f.Toolchains) != 2 {
+		t.Fatalf("toolchains = %d", len(f.Toolchains))
+	}
+	if f.Toolchains[0].Binary != "gcc-13" || f.Toolchains[0].Class != "gnu" {
+		t.Errorf("gcc13 = %+v", f.Toolchains[0])
+	}
+	if f.Toolchains[1].Class != "msvc" {
+		t.Errorf("msvc class = %q", f.Toolchains[1].Class)
+	}
+}
--- a/dsl/build_string_test.go
+++ b/dsl/build_string_test.go
@@ -0,0 +1,74 @@
+package dsl
+
+import "testing"
+
+func TestBuildProfileString(t *testing.T) {
+	cases := []struct {
+		p    BuildProfile
+		want string
+	}{
+		{ProfileRelease, "release"},
+		{ProfileDebug, "debug"},
+		{ProfileSanitized, "sanitized"},
+		{ProfileUnset, "unset"},
+		{BuildProfile(999), "unset"},
+	}
+	for _, c := range cases {
+		if got := c.p.String(); got != c.want {
+			t.Errorf("BuildProfile(%d).String() = %q, want %q", c.p, got, c.want)
+		}
+	}
+}
+
+func TestWarningLevelString(t *testing.T) {
+	cases := []struct {
+		w    WarningLevel
+		want string
+	}{
+		{WarningsDefault, "default"},
+		{WarningsStrict, "strict"},
+		{WarningsPedantic, "pedantic"},
+		{WarningsUnset, "unset"},
+		{WarningLevel(999), "unset"},
+	}
+	for _, c := range cases {
+		if got := c.w.String(); got != c.want {
+			t.Errorf("WarningLevel(%d).String() = %q, want %q", c.w, got, c.want)
+		}
+	}
+}
+
+func TestMergeFromWrapperAndDefines(t *testing.T) {
+	dst := &BuildConfig{}
+	src := &BuildConfig{
+		Wrapper: "valgrind",
+		Defines: map[string]string{"DEBUG": "1", "VERSION": "2"},
+	}
+	dst.MergeFrom(src)
+	if dst.Wrapper != "valgrind" {
+		t.Errorf("Wrapper = %q, want valgrind", dst.Wrapper)
+	}
+	if dst.Defines["DEBUG"] != "1" || dst.Defines["VERSION"] != "2" {
+		t.Errorf("Defines = %v", dst.Defines)
+	}
+}
+
+func TestMergeFromDefinesIntoExisting(t *testing.T) {
+	dst := &BuildConfig{Defines: map[string]string{"KEEP": "old"}}
+	src := &BuildConfig{Defines: map[string]string{"NEW": "1", "KEEP": "new"}}
+	dst.MergeFrom(src)
+	if dst.Defines["KEEP"] != "new" {
+		t.Errorf("KEEP = %q, want new (overridden)", dst.Defines["KEEP"])
+	}
+	if dst.Defines["NEW"] != "1" {
+		t.Errorf("NEW = %q, want 1", dst.Defines["NEW"])
+	}
+}
+
+func TestMergeFromNilSrc(t *testing.T) {
+	dst := &BuildConfig{Language: "c"}
+	dst.MergeFrom(nil)
+	if dst.Language != "c" {
+		t.Errorf("dst mutated on nil merge: %+v", dst)
+	}
+}
--- a/dsl/build_test.go
+++ b/dsl/build_test.go
@@ -0,0 +1,455 @@
+package dsl
+
+import (
+	"strings"
+	"testing"
+)
+
+func TestParseBuildBlockMinimal(t *testing.T) {
+	src := `
+build_defaults {
+    language = "c"
+    standard = "c11"
+    sources  = "*.c"
+    output   = "solution"
+    warnings = strict
+}
+
+build "release" {
+    profile = release
+}
+
+build "debug" {
+    profile = debug
+}
+
+group("g1") {
+    weight = 1.0
+    test("t1") {
+        stdout = "ok\n"
+    }
+}
+`
+	f, _, err := Parse(src)
+	if err != nil {
+		t.Fatalf("parse: %v", err)
+	}
+
+	if f.BuildDefaults == nil {
+		t.Fatal("BuildDefaults not populated")
+	}
+	if f.BuildDefaults.Language != "c" {
+		t.Errorf("language = %q", f.BuildDefaults.Language)
+	}
+	if f.BuildDefaults.Standard != "c11" {
+		t.Errorf("standard = %q", f.BuildDefaults.Standard)
+	}
+	if len(f.BuildDefaults.Sources) != 1 || f.BuildDefaults.Sources[0] != "*.c" {
+		t.Errorf("sources = %v", f.BuildDefaults.Sources)
+	}
+	if f.BuildDefaults.Output != "solution" {
+		t.Errorf("output = %q", f.BuildDefaults.Output)
+	}
+	if f.BuildDefaults.Warnings != WarningsStrict {
+		t.Errorf("warnings = %v", f.BuildDefaults.Warnings)
+	}
+
+	if len(f.Builds) != 2 {
+		t.Fatalf("expected 2 builds, got %d", len(f.Builds))
+	}
+	if f.Builds[0].Name != "release" || f.Builds[0].Profile != ProfileRelease {
+		t.Errorf("builds[0] = %+v", f.Builds[0])
+	}
+	if f.Builds[1].Name != "debug" || f.Builds[1].Profile != ProfileDebug {
+		t.Errorf("builds[1] = %+v", f.Builds[1])
+	}
+}
+
+func TestParseBuildLegacyStillWorks(t *testing.T) {
+	src := `
+build "cc -O2 solution.c -o solution"
+timeout 5s
+
+group("g1") {
+    weight = 1.0
+    test("t1") { stdout = "ok\n" }
+}
+`
+	f, _, err := Parse(src)
+	if err != nil {
+		t.Fatalf("parse: %v", err)
+	}
+	if f.Build != "cc -O2 solution.c -o solution" {
+		t.Errorf("legacy build = %q", f.Build)
+	}
+	if f.BuildDefaults != nil || len(f.Builds) != 0 {
+		t.Errorf("structured fields should be empty for legacy form")
+	}
+}
+
+func TestParseBuildOSOverride(t *testing.T) {
+	src := `
+build_defaults {
+    language = "c"
+    standard = "c11"
+    sources  = "*.c"
+    output   = "solution"
+}
+
+build "release" {
+    profile = release
+    linux   { extra = "-fPIC" }
+    windows { extra = "/bigobj" }
+}
+
+group("g1") {
+    weight = 1.0
+    test("t1") { stdout = "ok\n" }
+}
+`
+	f, _, err := Parse(src)
+	if err != nil {
+		t.Fatalf("parse: %v", err)
+	}
+	b := f.Builds[0]
+	if b.Linux == nil || len(b.Linux.Extra) != 1 || b.Linux.Extra[0] != "-fPIC" {
+		t.Errorf("linux override = %+v", b.Linux)
+	}
+	if b.Windows == nil || len(b.Windows.Extra) != 1 || b.Windows.Extra[0] != "/bigobj" {
+		t.Errorf("windows override = %+v", b.Windows)
+	}
+	if b.Darwin != nil {
+		t.Errorf("darwin override should be nil")
+	}
+}
+
+func TestParseBuildPlatformsFilter(t *testing.T) {
+	src := `
+build "sanitized" {
+    profile   = sanitized
+    sanitize  = "address" "undefined"
+    platforms = "linux"
+    compilers = "gcc" "clang"
+}
+
+group("g1") {
+    weight = 1.0
+    test("t1") { stdout = "ok\n" }
+}
+`
+	f, _, err := Parse(src)
+	if err != nil {
+		t.Fatalf("parse: %v", err)
+	}
+	b := f.Builds[0]
+	if b.Profile != ProfileSanitized {
+		t.Errorf("profile = %v", b.Profile)
+	}
+	if len(b.Sanitize) != 2 || b.Sanitize[0] != "address" || b.Sanitize[1] != "undefined" {
+		t.Errorf("sanitize = %v", b.Sanitize)
+	}
+	if len(b.Platforms) != 1 || b.Platforms[0] != "linux" {
+		t.Errorf("platforms = %v", b.Platforms)
+	}
+	if len(b.Compilers) != 2 {
+		t.Errorf("compilers = %v", b.Compilers)
+	}
+}
+
+func TestParseBuildDefineField(t *testing.T) {
+	src := `
+build "rel" {
+    profile = release
+    define("NDEBUG") = "1"
+    define("VERSION") = "42"
+}
+
+group("g1") {
+    weight = 1.0
+    test("t1") { stdout = "ok\n" }
+}
+`
+	f, _, err := Parse(src)
+	if err != nil {
+		t.Fatalf("parse: %v", err)
+	}
+	d := f.Builds[0].Defines
+	if d["NDEBUG"] != "1" || d["VERSION"] != "42" {
+		t.Errorf("defines = %v", d)
+	}
+}
+
+func TestParseBuildErrors(t *testing.T) {
+	cases := []struct {
+		name string
+		src  string
+		want string
+	}{
+		{
+			name: "unknown profile",
+			src: `
+build "x" { profile = ultra }
+group("g") { weight=1.0 test("t"){ stdout="ok\n" } }
+`,
+			want: "unknown profile",
+		},
+		{
+			name: "unknown warnings",
+			src: `
+build "x" { warnings = insane }
+group("g") { weight=1.0 test("t"){ stdout="ok\n" } }
+`,
+			want: "unknown warnings",
+		},
+		{
+			name: "unknown platform",
+			src: `
+build "x" { platforms = "bsd" }
+group("g") { weight=1.0 test("t"){ stdout="ok\n" } }
+`,
+			want: "unknown platform",
+		},
+		{
+			name: "nested OS override",
+			src: `
+build "x" {
+    linux {
+        windows { extra = "nope" }
+    }
+}
+group("g") { weight=1.0 test("t"){ stdout="ok\n" } }
+`,
+			want: "cannot be nested",
+		},
+		{
+			name: "duplicate OS override",
+			src: `
+build "x" {
+    linux { extra = "-a" }
+    linux { extra = "-b" }
+}
+group("g") { weight=1.0 test("t"){ stdout="ok\n" } }
+`,
+			want: "duplicate linux override",
+		},
+		{
+			name: "unknown field",
+			src: `
+build "x" { magic = "yes" }
+group("g") { weight=1.0 test("t"){ stdout="ok\n" } }
+`,
+			want: "unknown field",
+		},
+	}
+
+	for _, c := range cases {
+		t.Run(c.name, func(t *testing.T) {
+			_, _, err := Parse(c.src)
+			if err == nil {
+				t.Fatal("expected error, got nil")
+			}
+			if !strings.Contains(err.Error(), c.want) {
+				t.Errorf("error %q does not contain %q", err.Error(), c.want)
+			}
+		})
+	}
+}
+
+func TestParseBuildMixedLegacyAndStructuredRejected(t *testing.T) {
+	src := `
+build "cc -O2 solution.c -o solution"
+build "release" { profile = release }
+group("g") { weight = 1.0 test("t") { stdout = "ok\n" } }
+`
+	_, _, err := Parse(src)
+	if err == nil {
+		t.Fatal("expected error mixing legacy and structured builds")
+	}
+	if !strings.Contains(err.Error(), "cannot mix") {
+		t.Errorf("error %q does not mention mixing", err.Error())
+	}
+}
+
+func TestParseBuildDuplicateNameRejected(t *testing.T) {
+	src := `
+build "release" { profile = release }
+build "release" { profile = debug }
+group("g") { weight = 1.0 test("t") { stdout = "ok\n" } }
+`
+	_, _, err := Parse(src)
+	if err == nil {
+		t.Fatal("expected error on duplicate build name")
+	}
+	if !strings.Contains(err.Error(), "duplicate") {
+		t.Errorf("error %q does not mention duplicate", err.Error())
+	}
+}
+
+func TestBuildConfigResolveMerge(t *testing.T) {
+	defaults := &BuildConfig{
+		Language: "c",
+		Standard: "c11",
+		Sources:  []string{"*.c"},
+		Output:   "solution",
+		Warnings: WarningsStrict,
+	}
+	b := &BuildConfig{
+		Name:    "release",
+		Profile: ProfileRelease,
+		Extra:   []string{"-DFOO"},
+		Linux: &BuildConfig{
+			Extra: []string{"-fPIC"},
+		},
+		Windows: &BuildConfig{
+			Extra:  []string{"/bigobj"},
+			Output: "solution",
+		},
+	}
+
+	linux := b.Resolve(defaults, "linux")
+	if linux.Language != "c" || linux.Standard != "c11" {
+		t.Errorf("linux defaults not merged: %+v", linux)
+	}
+	if linux.Profile != ProfileRelease {
+		t.Errorf("linux profile = %v", linux.Profile)
+	}
+	if linux.Warnings != WarningsStrict {
+		t.Errorf("linux warnings = %v", linux.Warnings)
+	}
+	if len(linux.Sources) != 1 || linux.Sources[0] != "*.c" {
+		t.Errorf("linux sources = %v", linux.Sources)
+	}
+	if len(linux.Extra) != 2 || linux.Extra[0] != "-DFOO" || linux.Extra[1] != "-fPIC" {
+		t.Errorf("linux extra = %v", linux.Extra)
+	}
+
+	windows := b.Resolve(defaults, "windows")
+	if len(windows.Extra) != 2 || windows.Extra[0] != "-DFOO" || windows.Extra[1] != "/bigobj" {
+		t.Errorf("windows extra = %v", windows.Extra)
+	}
+
+	darwin := b.Resolve(defaults, "darwin")
+	if len(darwin.Extra) != 1 || darwin.Extra[0] != "-DFOO" {
+		t.Errorf("darwin extra (no override) = %v", darwin.Extra)
+	}
+
+	if len(b.Extra) != 1 {
+		t.Errorf("receiver mutated: %v", b.Extra)
+	}
+}
+
+func TestParseToolchainsBlock(t *testing.T) {
+	src := `
+toolchains {
+    gcc   { platforms = "linux" }
+    clang {
+        platforms = "linux" "windows"
+        binary    = "clang-17"
+        class     = gnu
+    }
+    msvc  { platforms = "windows" }
+    nvcc  {
+        platforms = "linux"
+        class     = gnu
+    }
+}
+
+build "release" { profile = release }
+
+group("g") { weight = 1.0 test("t") { stdout = "ok\n" } }
+`
+	f, _, err := Parse(src)
+	if err != nil {
+		t.Fatalf("parse: %v", err)
+	}
+	if len(f.Toolchains) != 4 {
+		t.Fatalf("expected 4 toolchains, got %d", len(f.Toolchains))
+	}
+	gcc := f.Toolchains[0]
+	if gcc.Name != "gcc" || len(gcc.Platforms) != 1 || gcc.Platforms[0] != "linux" {
+		t.Errorf("gcc = %+v", gcc)
+	}
+	clang := f.Toolchains[1]
+	if clang.Name != "clang" || len(clang.Platforms) != 2 || clang.Binary != "clang-17" || clang.Class != "gnu" {
+		t.Errorf("clang = %+v", clang)
+	}
+	if clang.Platforms[0] != "linux" || clang.Platforms[1] != "windows" {
+		t.Errorf("clang platforms = %v", clang.Platforms)
+	}
+	nvcc := f.Toolchains[3]
+	if nvcc.Name != "nvcc" || len(nvcc.Platforms) != 1 || nvcc.Platforms[0] != "linux" || nvcc.Class != "gnu" {
+		t.Errorf("nvcc = %+v", nvcc)
+	}
+}
+
+func TestParseToolchainsErrors(t *testing.T) {
+	cases := []struct {
+		name string
+		src  string
+		want string
+	}{
+		{
+			"missing platforms",
+			`toolchains { gcc { } }
+			 group("g") { weight=1.0 test("t") { stdout="ok\n" } }`,
+			"platforms is required",
+		},
+		{
+			"unknown platform",
+			`toolchains { gcc { platforms = "bsd" } }
+			 group("g") { weight=1.0 test("t") { stdout="ok\n" } }`,
+			"unknown platform",
+		},
+		{
+			"unknown class",
+			`toolchains { gcc { platforms = "linux" class = llvm } }
+			 group("g") { weight=1.0 test("t") { stdout="ok\n" } }`,
+			"unknown compiler class",
+		},
+		{
+			"duplicate toolchain",
+			`toolchains {
+                gcc { platforms = "linux" }
+                gcc { platforms = "linux" }
+            }
+            group("g") { weight=1.0 test("t") { stdout="ok\n" } }`,
+			"duplicate toolchain",
+		},
+	}
+	for _, c := range cases {
+		t.Run(c.name, func(t *testing.T) {
+			_, _, err := Parse(c.src)
+			if err == nil {
+				t.Fatal("expected error, got nil")
+			}
+			if !strings.Contains(err.Error(), c.want) {
+				t.Errorf("error %q does not contain %q", err.Error(), c.want)
+			}
+		})
+	}
+}
+
+func TestBuildConfigAppliesTo(t *testing.T) {
+	cases := []struct {
+		name     string
+		b        BuildConfig
+		os, cc   string
+		expected bool
+	}{
+		{"no filters", BuildConfig{}, "linux", "gcc", true},
+		{"os match", BuildConfig{Platforms: []string{"linux"}}, "linux", "gcc", true},
+		{"os mismatch", BuildConfig{Platforms: []string{"linux"}}, "windows", "msvc", false},
+		{"cc match", BuildConfig{Compilers: []string{"gcc", "clang"}}, "linux", "clang", true},
+		{"cc mismatch", BuildConfig{Compilers: []string{"gcc"}}, "linux", "msvc", false},
+		{"both match", BuildConfig{Platforms: []string{"linux"}, Compilers: []string{"gcc"}}, "linux", "gcc", true},
+		{"os ok cc bad", BuildConfig{Platforms: []string{"linux"}, Compilers: []string{"gcc"}}, "linux", "clang", false},
+	}
+	for _, c := range cases {
+		t.Run(c.name, func(t *testing.T) {
+			got := c.b.AppliesTo(c.os, c.cc)
+			if got != c.expected {
+				t.Errorf("AppliesTo(%q, %q) = %v, want %v", c.os, c.cc, got, c.expected)
+			}
+		})
+	}
+}
--- a/dsl/include_test.go
+++ b/dsl/include_test.go
@@ -0,0 +1,210 @@
+package dsl
+
+import (
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+)
+
+func writeTempJdg(t *testing.T, dir, name, content string) string {
+	t.Helper()
+	path := filepath.Join(dir, name)
+	if err := os.MkdirAll(filepath.Dir(path), 0755); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(path, []byte(content), 0644); err != nil {
+		t.Fatal(err)
+	}
+	return path
+}
+
+func TestIncludeBasicMerge(t *testing.T) {
+	dir := t.TempDir()
+
+	writeTempJdg(t, dir, "common.jdg", `
+toolchains {
+    gcc { platforms = "linux" }
+}
+
+build_defaults {
+    language = "c"
+    standard = "c11"
+    sources  = "*.c"
+    output   = "solution"
+    warnings = strict
+}
+`)
+
+	mainPath := writeTempJdg(t, dir, "main.jdg", `
+include "common.jdg"
+
+build "release" { profile = release }
+
+group("g") {
+    weight = 1.0
+    test("t") { stdout = "ok\n" }
+}
+`)
+
+	f, _, err := ParseFile(mainPath)
+	if err != nil {
+		t.Fatalf("parse: %v", err)
+	}
+
+	if len(f.Toolchains) != 1 || f.Toolchains[0].Name != "gcc" {
+		t.Errorf("toolchains not merged: %+v", f.Toolchains)
+	}
+	if f.BuildDefaults == nil || f.BuildDefaults.Language != "c" {
+		t.Errorf("build_defaults not merged: %+v", f.BuildDefaults)
+	}
+	if len(f.Builds) != 1 || f.Builds[0].Name != "release" {
+		t.Errorf("local build lost: %+v", f.Builds)
+	}
+}
+
+func TestIncludeLocalOverridesIncluded(t *testing.T) {
+	dir := t.TempDir()
+
+	writeTempJdg(t, dir, "common.jdg", `
+build_defaults {
+    language = "c"
+    standard = "c11"
+    warnings = strict
+}
+`)
+
+	mainPath := writeTempJdg(t, dir, "main.jdg", `
+include "common.jdg"
+
+build_defaults {
+    standard = "c17"
+}
+
+group("g") { weight = 1.0 test("t") { stdout = "ok\n" } }
+`)
+
+	f, _, err := ParseFile(mainPath)
+	if err != nil {
+		t.Fatalf("parse: %v", err)
+	}
+	if f.BuildDefaults.Standard != "c17" {
+		t.Errorf("local should override included standard: got %q", f.BuildDefaults.Standard)
+	}
+	if f.BuildDefaults.Language != "c" {
+		t.Errorf("language should survive from include: got %q", f.BuildDefaults.Language)
+	}
+	if f.BuildDefaults.Warnings != WarningsStrict {
+		t.Errorf("warnings should survive from include: got %v", f.BuildDefaults.Warnings)
+	}
+}
+
+func TestIncludeRelativePathResolution(t *testing.T) {
+	dir := t.TempDir()
+
+	writeTempJdg(t, dir, "shared/tools.jdg", `
+toolchains {
+    gcc { platforms = "linux" }
+}
+`)
+
+	mainPath := writeTempJdg(t, dir, "suite/main.jdg", `
+include "../shared/tools.jdg"
+
+build "release" {
+    language = "c"
+    sources  = "solution.c"
+    output   = "solution"
+    profile  = release
+}
+
+group("g") { weight = 1.0 test("t") { stdout = "ok\n" } }
+`)
+
+	f, _, err := ParseFile(mainPath)
+	if err != nil {
+		t.Fatalf("parse: %v", err)
+	}
+	if len(f.Toolchains) != 1 {
+		t.Errorf("relative include failed to resolve: %+v", f.Toolchains)
+	}
+}
+
+func TestIncludeDuplicateToolchainErrors(t *testing.T) {
+	dir := t.TempDir()
+
+	writeTempJdg(t, dir, "common.jdg", `
+toolchains {
+    gcc { platforms = "linux" }
+}
+`)
+
+	mainPath := writeTempJdg(t, dir, "main.jdg", `
+include "common.jdg"
+
+toolchains {
+    gcc { platforms = "linux" }
+}
+
+group("g") { weight = 1.0 test("t") { stdout = "ok\n" } }
+`)
+
+	_, _, err := ParseFile(mainPath)
+	if err == nil {
+		t.Fatal("expected duplicate toolchain error")
+	}
+	if !strings.Contains(err.Error(), "duplicate toolchain") {
+		t.Errorf("error %q does not mention duplicate toolchain", err.Error())
+	}
+}
+
+func TestIncludeCircularDetection(t *testing.T) {
+	dir := t.TempDir()
+
+	aPath := writeTempJdg(t, dir, "a.jdg", `
+include "b.jdg"
+group("ga") { weight = 1.0 test("t") { stdout = "ok\n" } }
+`)
+	writeTempJdg(t, dir, "b.jdg", `
+include "a.jdg"
+`)
+
+	_, _, err := ParseFile(aPath)
+	if err == nil {
+		t.Fatal("expected circular include error")
+	}
+	if !strings.Contains(err.Error(), "circular include") {
+		t.Errorf("error %q does not mention circular include", err.Error())
+	}
+}
+
+func TestIncludeMissingFileErrors(t *testing.T) {
+	dir := t.TempDir()
+
+	mainPath := writeTempJdg(t, dir, "main.jdg", `
+include "nonexistent.jdg"
+group("g") { weight = 1.0 test("t") { stdout = "ok\n" } }
+`)
+
+	_, _, err := ParseFile(mainPath)
+	if err == nil {
+		t.Fatal("expected missing include error")
+	}
+	if !strings.Contains(err.Error(), "nonexistent.jdg") {
+		t.Errorf("error %q does not reference the missing path", err.Error())
+	}
+}
+
+func TestParseRejectsIncludeWithoutFileContext(t *testing.T) {
+	src := `
+include "common.jdg"
+group("g") { weight = 1.0 test("t") { stdout = "ok\n" } }
+`
+	_, _, err := Parse(src)
+	if err == nil {
+		t.Fatal("expected error: include without file context")
+	}
+	if !strings.Contains(err.Error(), "file context") {
+		t.Errorf("error %q does not mention file context", err.Error())
+	}
+}
--- a/dsl/lexer.go
+++ b/dsl/lexer.go
@@ -371,8 +371,6 @@ func (l *Lexer) readNumberOrDuration(line, col int) (Token, error) {
 	return Token{TOKEN_INT, buf.String(), line, col}, nil
 }

-// tryReadSizeSuffix reads memory size suffixes: B, K, KB, KiB, M, MB, MiB, G, GB, GiB.
-// Units are case-sensitive uppercase to avoid collision with duration "m" (minutes).
 func (l *Lexer) tryReadSizeSuffix() string {
 	ch, ok := l.peek()
 	if !ok {
--- a/dsl/lexer_test.go
+++ b/dsl/lexer_test.go
@@ -0,0 +1,175 @@
+package dsl
+
+import (
+	"strings"
+	"testing"
+)
+
+func TestTokenStringAndUnknownType(t *testing.T) {
+	tok := Token{Type: TOKEN_IDENT, Value: "foo", Line: 2, Col: 5}
+	s := tok.String()
+	if !strings.Contains(s, "IDENT") || !strings.Contains(s, "foo") {
+		t.Errorf("Token.String() = %q", s)
+	}
+	if got := TokenType(999).String(); got != "UNKNOWN" {
+		t.Errorf("TokenType(999).String() = %q, want UNKNOWN", got)
+	}
+	for tt, want := range map[TokenType]string{
+		TOKEN_STRING:   "STRING",
+		TOKEN_FLOAT:    "FLOAT",
+		TOKEN_INT:      "INT",
+		TOKEN_DURATION: "DURATION",
+		TOKEN_SIZE:     "SIZE",
+		TOKEN_LBRACE:   "{",
+		TOKEN_RBRACE:   "}",
+		TOKEN_LPAREN:   "(",
+		TOKEN_RPAREN:   ")",
+		TOKEN_ASSIGN:   "=",
+		TOKEN_TILDE:    "~",
+		TOKEN_EOF:      "EOF",
+	} {
+		if got := tt.String(); got != want {
+			t.Errorf("TokenType(%d) = %q, want %q", tt, got, want)
+		}
+	}
+}
+
+func TestLexerLineComment(t *testing.T) {
+	src := `
+// leading comment
+build "make" // trailing
+group("g") { // inside
+    weight = 1.0
+    test("t") { stdout = "" }
+}
+`
+	if _, _, err := Parse(src); err != nil {
+		t.Errorf("parse with comments: %v", err)
+	}
+}
+
+func TestLexerUnterminatedString(t *testing.T) {
+	_, _, err := Parse(`build "unterminated`)
+	if err == nil || !strings.Contains(err.Error(), "unterminated") {
+		t.Errorf("want unterminated string error, got %v", err)
+	}
+}
+
+func TestLexerUnknownEscape(t *testing.T) {
+	_, _, err := Parse(`build "bad\zescape"`)
+	if err == nil || !strings.Contains(err.Error(), "unknown escape") {
+		t.Errorf("want unknown escape error, got %v", err)
+	}
+}
+
+func TestLexerEscapeSequences(t *testing.T) {
+	src := `build "a\nb\tc\\d\"e"`
+	f, _, err := Parse(src + "\ngroup(\"g\") { weight = 1.0 test(\"t\") { stdout = \"\" } }")
+	if err != nil {
+		t.Fatalf("parse: %v", err)
+	}
+	if f.Build != "a\nb\tc\\d\"e" {
+		t.Errorf("escape sequences = %q", f.Build)
+	}
+}
+
+func TestLexerUnexpectedCharacter(t *testing.T) {
+	_, _, err := Parse("build @invalid")
+	if err == nil || !strings.Contains(err.Error(), "unexpected character") {
+		t.Errorf("want unexpected character error, got %v", err)
+	}
+}
+
+func TestLexerUnterminatedHeredoc(t *testing.T) {
+	src := "build \"x\"\ngroup(\"g\") { weight = 1.0 test(\"t\") { stdin = \"\"\"\nnever closed\n } }"
+	_, _, err := Parse(src)
+	if err == nil || !strings.Contains(err.Error(), "unterminated heredoc") {
+		t.Errorf("want unterminated heredoc, got %v", err)
+	}
+}
+
+func TestLexerSizeSuffixes(t *testing.T) {
+	cases := map[string]int64{
+		"1024":    1024,
+		"1B":      1,
+		"1K":      1024,
+		"1KB":     1024,
+		"1KiB":    1024,
+		"1M":      1024 * 1024,
+		"1MB":     1024 * 1024,
+		"1MiB":    1024 * 1024,
+		"1G":      1024 * 1024 * 1024,
+		"1GB":     1024 * 1024 * 1024,
+		"1GiB":    1024 * 1024 * 1024,
+	}
+	for literal, want := range cases {
+		src := "build \"x\"\nmemory_limit = " + literal + "\ngroup(\"g\") { weight = 1.0 test(\"t\") { stdout = \"\" } }"
+		f, _, err := Parse(src)
+		if err != nil {
+			t.Errorf("parse %q: %v", literal, err)
+			continue
+		}
+		if f.MemoryLimit != want {
+			t.Errorf("memory_limit %s = %d, want %d", literal, f.MemoryLimit, want)
+		}
+	}
+}
+
+func TestLexerDurationSuffixes(t *testing.T) {
+	cases := []string{"5ms", "10s", "2m"}
+	for _, d := range cases {
+		src := "build \"x\"\ntimeout " + d + "\ngroup(\"g\") { weight = 1.0 test(\"t\") { stdout = \"\" } }"
+		if _, _, err := Parse(src); err != nil {
+			t.Errorf("parse timeout %s: %v", d, err)
+		}
+	}
+}
+
+func TestLexerNegativeInt(t *testing.T) {
+	src := `
+build "x"
+group("g") {
+    weight = 1.0
+    test("t") {
+        exitCode = -1
+        stdout = ""
+    }
+}
+`
+	f, _, err := Parse(src)
+	if err != nil {
+		t.Fatalf("parse: %v", err)
+	}
+	tst := f.Groups[0].Tests[0]
+	if tst.ExitCode == nil || *tst.ExitCode != -1 {
+		t.Errorf("ExitCode = %v, want -1", tst.ExitCode)
+	}
+}
+
+func TestLexerFloatNumber(t *testing.T) {
+	src := `
+build "x"
+group("g") {
+    weight = 0.75
+    test("t") { stdout = "" }
+}
+group("g2") {
+    weight = 0.25
+    test("t") { stdout = "" }
+}
+`
+	f, _, err := Parse(src)
+	if err != nil {
+		t.Fatalf("parse: %v", err)
+	}
+	if f.Groups[0].Weight != 0.75 {
+		t.Errorf("weight = %v", f.Groups[0].Weight)
+	}
+}
+
+func TestLexerInvalidSizeUnit(t *testing.T) {
+	_, _, err := Parse("build \"x\"\nmemory_limit = 10Z\n")
+	if err == nil {
+		t.Error("expected error for invalid size unit")
+	}
+}
--- a/runner/matcher.go
+++ b/runner/matcher.go
@@ -1,4 +1,4 @@
-package runner
+package dsl

 import (
 	"fmt"
@@ -7,57 +7,90 @@ import (
 	"sort"
 	"strconv"
 	"strings"
-
-	"github.com/Mond1c/judge/dsl"
 )

-func applyMatcher(label string, m dsl.Matcher, actual string) []string {
-	switch m := m.(type) {
-	case dsl.NoMatcher:
-		return nil
-	case dsl.ExactMatcher:
-		if actual != m.Value {
-			return []string{fmt.Sprintf(
-				"%s mismatch:\n  expected: %q\n  actual:   %q",
-				label, m.Value, actual,
-			)}
-		}
-		return nil
-	case dsl.ContainsMatcher:
-		if !strings.Contains(actual, m.Substr) {
-			return []string{fmt.Sprintf(
-				"%s: expected to contain %q, got %q",
-				label, m.Substr, actual,
-			)}
-		}
-		return nil
-	case dsl.RegexMatcher:
-		re, err := regexp.Compile(m.Pattern)
-		if err != nil {
-			return []string{fmt.Sprintf("%s: invalid regex %q: %v", label, m.Pattern, err)}
-		}
-		if !re.MatchString(actual) {
-			return []string{fmt.Sprintf(
-				"%s: %q does not match regex %q",
-				label, actual, m.Pattern,
-			)}
-		}
-		return nil
+// TODO: maybe move to ast.go
+type Matcher interface {
+	matcherNode()

-	case dsl.NumericEpsMatcher:
-		errs := matchNumericEps(label, m, actual)
-		return errs
-
-	case dsl.AnyOrderMatcher:
-		return matchAnyOrder(label, m, actual)
-
-	default:
-		return []string{fmt.Sprintf("unknown matcher type %T", m)}
-
-	}
+	Match(label, actual string) []string
 }

-func matchNumericEps(label string, m dsl.NumericEpsMatcher, actual string) []string {
+type ExactMatcher struct {
+	Value string
+}
+
+func (ExactMatcher) matcherNode() {}
+
+// TODO: think about pointer receivers
+func (m ExactMatcher) Match(label, actual string) []string {
+	if actual != m.Value {
+		return []string{fmt.Sprintf(
+			"%s mismatch:\n  expected: %q\n  actual:   %q",
+			label, m.Value, actual,
+		)}
+	}
+	return nil
+}
+
+type ContainsMatcher struct {
+	Substr string
+}
+
+func (ContainsMatcher) matcherNode() {}
+
+func (m ContainsMatcher) Match(label, actual string) []string {
+	if !strings.Contains(actual, m.Substr) {
+		return []string{fmt.Sprintf(
+			"%s: expected to contain %q, got %q",
+			label, m.Substr, actual,
+		)}
+	}
+	return nil
+}
+
+type RegexMatcher struct {
+	Pattern string
+}
+
+func (RegexMatcher) matcherNode() {}
+
+func (m RegexMatcher) Match(label, actual string) []string {
+	re, err := regexp.Compile(m.Pattern)
+	if err != nil {
+		return []string{fmt.Sprintf("%s: invalid regex %q: %v", label, m.Pattern, err)}
+	}
+
+	if !re.MatchString(actual) {
+		return []string{fmt.Sprintf(
+			"%s: %q does not match regex %q",
+			label, actual, m.Pattern,
+		)}
+	}
+	return nil
+}
+
+type NumericEpsMatcher struct {
+	Epsilon float64
+	Value   string
+}
+
+func (NumericEpsMatcher) matcherNode() {}
+
+func parseNumbers(s string) ([]float64, error) {
+	fields := strings.Fields(s)
+	nums := make([]float64, 0, len(fields))
+	for _, f := range fields {
+		n, err := strconv.ParseFloat(f, 64)
+		if err != nil {
+			return nil, fmt.Errorf("not a number: %q", f)
+		}
+		nums = append(nums, n)
+	}
+	return nums, nil
+}
+
+func (m NumericEpsMatcher) Match(label, actual string) []string {
 	expectedNums, err := parseNumbers(m.Value)
 	if err != nil {
 		return []string{fmt.Sprintf("%s: cannot parse expected numbers %q: %v", label, m.Value, err)}
@@ -85,20 +118,21 @@ func matchNumericEps(label string, m dsl.NumericEpsMatcher, actual string) []str
 	return errs
 }

-func parseNumbers(s string) ([]float64, error) {
-	fields := strings.Fields(s)
-	nums := make([]float64, 0, len(fields))
-	for _, f := range fields {
-		n, err := strconv.ParseFloat(f, 64)
-		if err != nil {
-			return nil, fmt.Errorf("not a number: %q", f)
-		}
-		nums = append(nums, n)
-	}
-	return nums, nil
+type AnyOrderMatcher struct {
+	Lines []string
 }

-func matchAnyOrder(label string, m dsl.AnyOrderMatcher, actual string) []string {
+func (AnyOrderMatcher) matcherNode() {}
+
+func splitLines(s string) []string {
+	s = strings.TrimRight(s, "\n")
+	if s == "" {
+		return []string{}
+	}
+	return strings.Split(s, "\n")
+}
+
+func (m AnyOrderMatcher) Match(label, actual string) []string {
 	actualLines := splitLines(actual)
 	expectedLines := make([]string, len(m.Lines))
 	copy(expectedLines, m.Lines)
@@ -125,10 +159,10 @@ func matchAnyOrder(label string, m dsl.AnyOrderMatcher, actual string) []string
 	return errs
 }

-func splitLines(s string) []string {
-	s = strings.TrimRight(s, "\n")
-	if s == "" {
-		return []string{}
-	}
-	return strings.Split(s, "\n")
+type NoMatcher struct{}
+
+func (NoMatcher) matcherNode() {}
+
+func (NoMatcher) Match(label, actual string) []string {
+	return nil
 }
--- a/dsl/matcher_test.go
+++ b/dsl/matcher_test.go
@@ -0,0 +1,187 @@
+package dsl
+
+import (
+	"strings"
+	"testing"
+)
+
+func TestExactMatcherPass(t *testing.T) {
+	m := ExactMatcher{Value: "hello\n"}
+	if errs := m.Match("stdout", "hello\n"); errs != nil {
+		t.Errorf("expected pass, got %v", errs)
+	}
+}
+
+func TestExactMatcherMismatch(t *testing.T) {
+	m := ExactMatcher{Value: "hello\n"}
+	errs := m.Match("stdout", "world\n")
+	if len(errs) != 1 {
+		t.Fatalf("want 1 error, got %d: %v", len(errs), errs)
+	}
+	if !strings.Contains(errs[0], "stdout mismatch") {
+		t.Errorf("error missing label: %q", errs[0])
+	}
+	if !strings.Contains(errs[0], "hello") || !strings.Contains(errs[0], "world") {
+		t.Errorf("error missing values: %q", errs[0])
+	}
+}
+
+func TestContainsMatcherPass(t *testing.T) {
+	m := ContainsMatcher{Substr: "needle"}
+	if errs := m.Match("stdout", "haystack with a needle inside"); errs != nil {
+		t.Errorf("expected pass, got %v", errs)
+	}
+}
+
+func TestContainsMatcherMissing(t *testing.T) {
+	m := ContainsMatcher{Substr: "needle"}
+	errs := m.Match("stdout", "only hay here")
+	if len(errs) != 1 {
+		t.Fatalf("want 1 error, got %d: %v", len(errs), errs)
+	}
+	if !strings.Contains(errs[0], "needle") {
+		t.Errorf("error missing substring: %q", errs[0])
+	}
+}
+
+func TestRegexMatcherPass(t *testing.T) {
+	m := RegexMatcher{Pattern: `^hello .*!$`}
+	if errs := m.Match("stdout", "hello world!"); errs != nil {
+		t.Errorf("expected pass, got %v", errs)
+	}
+}
+
+func TestRegexMatcherMismatch(t *testing.T) {
+	m := RegexMatcher{Pattern: `^\d+$`}
+	errs := m.Match("stdout", "not-a-number")
+	if len(errs) != 1 {
+		t.Fatalf("want 1 error, got %d: %v", len(errs), errs)
+	}
+	if !strings.Contains(errs[0], "does not match") {
+		t.Errorf("error unexpected: %q", errs[0])
+	}
+}
+
+func TestRegexMatcherInvalidPattern(t *testing.T) {
+	m := RegexMatcher{Pattern: `[`}
+	errs := m.Match("stdout", "anything")
+	if len(errs) != 1 {
+		t.Fatalf("want 1 error, got %d: %v", len(errs), errs)
+	}
+	if !strings.Contains(errs[0], "invalid regex") {
+		t.Errorf("error unexpected: %q", errs[0])
+	}
+}
+
+func TestNumericEpsMatcherPassWithinEps(t *testing.T) {
+	m := NumericEpsMatcher{Epsilon: 0.01, Value: "1.0 2.0 3.0"}
+	if errs := m.Match("stdout", "1.005 1.999 3.0"); errs != nil {
+		t.Errorf("expected pass, got %v", errs)
+	}
+}
+
+func TestNumericEpsMatcherExceedsEps(t *testing.T) {
+	m := NumericEpsMatcher{Epsilon: 0.01, Value: "1.0 2.0"}
+	errs := m.Match("stdout", "1.0 2.5")
+	if len(errs) != 1 {
+		t.Fatalf("want 1 error, got %d: %v", len(errs), errs)
+	}
+	if !strings.Contains(errs[0], "number[1]") {
+		t.Errorf("error missing index: %q", errs[0])
+	}
+}
+
+func TestNumericEpsMatcherCountMismatch(t *testing.T) {
+	m := NumericEpsMatcher{Epsilon: 0.01, Value: "1 2 3"}
+	errs := m.Match("stdout", "1 2")
+	if len(errs) != 1 {
+		t.Fatalf("want 1 error, got %d: %v", len(errs), errs)
+	}
+	if !strings.Contains(errs[0], "expected 3 numbers, got 2") {
+		t.Errorf("error unexpected: %q", errs[0])
+	}
+}
+
+func TestNumericEpsMatcherBadExpected(t *testing.T) {
+	m := NumericEpsMatcher{Epsilon: 0.01, Value: "1 foo"}
+	errs := m.Match("stdout", "1 2")
+	if len(errs) != 1 || !strings.Contains(errs[0], "cannot parse expected") {
+		t.Errorf("want expected-parse error, got %v", errs)
+	}
+}
+
+func TestNumericEpsMatcherBadActual(t *testing.T) {
+	m := NumericEpsMatcher{Epsilon: 0.01, Value: "1 2"}
+	errs := m.Match("stdout", "1 bar")
+	if len(errs) != 1 || !strings.Contains(errs[0], "cannot parse actual") {
+		t.Errorf("want actual-parse error, got %v", errs)
+	}
+}
+
+func TestNumericEpsMatcherMultipleErrors(t *testing.T) {
+	m := NumericEpsMatcher{Epsilon: 0.01, Value: "1 2 3"}
+	errs := m.Match("stdout", "9 8 7")
+	if len(errs) != 3 {
+		t.Errorf("want 3 errors, got %d: %v", len(errs), errs)
+	}
+}
+
+func TestAnyOrderMatcherPassReordered(t *testing.T) {
+	m := AnyOrderMatcher{Lines: []string{"a", "b", "c"}}
+	if errs := m.Match("stdout", "c\nb\na\n"); errs != nil {
+		t.Errorf("expected pass, got %v", errs)
+	}
+}
+
+func TestAnyOrderMatcherPassNoTrailingNewline(t *testing.T) {
+	m := AnyOrderMatcher{Lines: []string{"x", "y"}}
+	if errs := m.Match("stdout", "y\nx"); errs != nil {
+		t.Errorf("expected pass, got %v", errs)
+	}
+}
+
+func TestAnyOrderMatcherCountMismatch(t *testing.T) {
+	m := AnyOrderMatcher{Lines: []string{"a", "b"}}
+	errs := m.Match("stdout", "a\nb\nc\n")
+	if len(errs) != 1 || !strings.Contains(errs[0], "expected 2 lines, got 3") {
+		t.Errorf("want count error, got %v", errs)
+	}
+}
+
+func TestAnyOrderMatcherLineMismatch(t *testing.T) {
+	m := AnyOrderMatcher{Lines: []string{"a", "b"}}
+	errs := m.Match("stdout", "a\nz\n")
+	if len(errs) != 1 || !strings.Contains(errs[0], "line mismatch") {
+		t.Errorf("want line mismatch, got %v", errs)
+	}
+}
+
+func TestAnyOrderMatcherEmptyBoth(t *testing.T) {
+	m := AnyOrderMatcher{Lines: []string{}}
+	if errs := m.Match("stdout", ""); errs != nil {
+		t.Errorf("expected pass on empty, got %v", errs)
+	}
+}
+
+func TestNoMatcherAlwaysPasses(t *testing.T) {
+	m := NoMatcher{}
+	if errs := m.Match("stdout", "anything at all\n"); errs != nil {
+		t.Errorf("NoMatcher should never fail, got %v", errs)
+	}
+	if errs := m.Match("stderr", ""); errs != nil {
+		t.Errorf("NoMatcher should never fail on empty, got %v", errs)
+	}
+}
+
+func TestSplitLinesEmpty(t *testing.T) {
+	if got := splitLines(""); len(got) != 0 {
+		t.Errorf("splitLines(\"\") = %v, want empty", got)
+	}
+}
+
+func TestSplitLinesTrailingNewline(t *testing.T) {
+	got := splitLines("a\nb\n")
+	if len(got) != 2 || got[0] != "a" || got[1] != "b" {
+		t.Errorf("splitLines trailing = %v, want [a b]", got)
+	}
+}
--- a/dsl/merge.go
+++ b/dsl/merge.go
@@ -0,0 +1,81 @@
+package dsl
+
+import "fmt"
+
+func mergeFiles(dst, src *File) error {
+	if src.Build != "" {
+		dst.Build = src.Build
+	}
+	if src.BuildLinux != "" {
+		dst.BuildLinux = src.BuildLinux
+	}
+	if src.BuildWindows != "" {
+		dst.BuildWindows = src.BuildWindows
+	}
+	if src.BuildDarwin != "" {
+		dst.BuildDarwin = src.BuildDarwin
+	}
+
+	if src.BuildDefaults != nil {
+		if dst.BuildDefaults == nil {
+			dst.BuildDefaults = &BuildConfig{}
+		}
+		dst.BuildDefaults.MergeFrom(src.BuildDefaults)
+	}
+
+	seenTC := map[string]bool{}
+	for _, t := range dst.Toolchains {
+		seenTC[t.Name] = true
+	}
+	for _, t := range src.Toolchains {
+		if seenTC[t.Name] {
+			return fmt.Errorf("duplicate toolchain %q", t.Name)
+		}
+		seenTC[t.Name] = true
+		dst.Toolchains = append(dst.Toolchains, t)
+	}
+
+	seenB := map[string]bool{}
+	for _, b := range dst.Builds {
+		seenB[b.Name] = true
+	}
+	for _, b := range src.Builds {
+		if seenB[b.Name] {
+			return fmt.Errorf("duplicate build %q", b.Name)
+		}
+		seenB[b.Name] = true
+		dst.Builds = append(dst.Builds, b)
+	}
+
+	seenG := map[string]bool{}
+	for _, g := range dst.Groups {
+		seenG[g.Name] = true
+	}
+	for _, g := range src.Groups {
+		if seenG[g.Name] {
+			return fmt.Errorf("duplicate group %q", g.Name)
+		}
+		seenG[g.Name] = true
+		dst.Groups = append(dst.Groups, g)
+	}
+
+	if src.Timeout != 0 {
+		dst.Timeout = src.Timeout
+	}
+	if src.MemoryLimit != 0 {
+		dst.MemoryLimit = src.MemoryLimit
+	}
+	if src.Binary != "" {
+		dst.Binary = src.Binary
+	}
+	if src.Sources != "" {
+		dst.Sources = src.Sources
+	}
+	if src.NormalizeCRLF {
+		dst.NormalizeCRLF = true
+	}
+	if src.TrimTrailingWS {
+		dst.TrimTrailingWS = true
+	}
+	return nil
+}
--- a/dsl/merge_test.go
+++ b/dsl/merge_test.go
@@ -0,0 +1,220 @@
+package dsl
+
+import (
+	"strings"
+	"testing"
+	"time"
+)
+
+func TestMergeLegacyBuildFields(t *testing.T) {
+	dir := t.TempDir()
+	writeTempJdg(t, dir, "common.jdg", `
+build         "make"
+build_linux   "make linux"
+build_windows "make windows"
+build_darwin  "make darwin"
+timeout 7s
+memory_limit = 256MB
+`)
+	mainPath := writeTempJdg(t, dir, "main.jdg", `
+include "common.jdg"
+group("g") { weight = 1.0 test("t") { stdout = "ok\n" } }
+`)
+	f, _, err := ParseFile(mainPath)
+	if err != nil {
+		t.Fatalf("parse: %v", err)
+	}
+	if f.Build != "make" || f.BuildLinux != "make linux" ||
+		f.BuildWindows != "make windows" || f.BuildDarwin != "make darwin" {
+		t.Errorf("legacy build fields not merged: %+v", f)
+	}
+	if f.Timeout != 7*time.Second {
+		t.Errorf("Timeout = %v, want 7s", f.Timeout)
+	}
+	if f.MemoryLimit != 256*1024*1024 {
+		t.Errorf("MemoryLimit = %d", f.MemoryLimit)
+	}
+}
+
+func TestMergeDuplicateToolchainFromInclude(t *testing.T) {
+	dir := t.TempDir()
+	writeTempJdg(t, dir, "common.jdg", `
+toolchains {
+    gcc { platforms = "linux" }
+}
+`)
+	mainPath := writeTempJdg(t, dir, "main.jdg", `
+toolchains {
+    gcc { platforms = "linux" }
+}
+include "common.jdg"
+build "make"
+group("g") { weight = 1.0 test("t") { stdout = "ok\n" } }
+`)
+	_, _, err := ParseFile(mainPath)
+	if err == nil {
+		t.Fatal("expected duplicate toolchain error from merge")
+	}
+	if !strings.Contains(err.Error(), "duplicate toolchain") {
+		t.Errorf("error %q does not mention duplicate toolchain", err.Error())
+	}
+}
+
+func TestMergeDuplicateBuildErrors(t *testing.T) {
+	dir := t.TempDir()
+	writeTempJdg(t, dir, "common.jdg", `
+build "release" { profile = release }
+`)
+	mainPath := writeTempJdg(t, dir, "main.jdg", `
+build_defaults {
+    language = "c"
+    standard = "c11"
+    sources  = "*.c"
+    output   = "solution"
+    warnings = strict
+}
+build "release" { profile = debug }
+include "common.jdg"
+group("g") { weight = 1.0 test("t") { stdout = "ok\n" } }
+`)
+	_, _, err := ParseFile(mainPath)
+	if err == nil {
+		t.Fatal("expected duplicate build error")
+	}
+	if !strings.Contains(err.Error(), "duplicate build") {
+		t.Errorf("error %q does not mention duplicate build", err.Error())
+	}
+}
+
+func TestMergeDuplicateGroupErrors(t *testing.T) {
+	dir := t.TempDir()
+	writeTempJdg(t, dir, "common.jdg", `
+group("shared") { weight = 0.5 test("t") { stdout = "ok\n" } }
+`)
+	mainPath := writeTempJdg(t, dir, "main.jdg", `
+build "make"
+group("shared") { weight = 0.5 test("t2") { stdout = "ok\n" } }
+include "common.jdg"
+`)
+	_, _, err := ParseFile(mainPath)
+	if err == nil {
+		t.Fatal("expected duplicate group error")
+	}
+	if !strings.Contains(err.Error(), "duplicate group") {
+		t.Errorf("error %q does not mention duplicate group", err.Error())
+	}
+}
+
+func TestMergeGroupsAppended(t *testing.T) {
+	dir := t.TempDir()
+	writeTempJdg(t, dir, "common.jdg", `
+build "make"
+group("included") { weight = 0.5 test("t") { stdout = "ok\n" } }
+`)
+	mainPath := writeTempJdg(t, dir, "main.jdg", `
+include "common.jdg"
+group("local") { weight = 0.5 test("t") { stdout = "ok\n" } }
+`)
+	f, _, err := ParseFile(mainPath)
+	if err != nil {
+		t.Fatalf("parse: %v", err)
+	}
+	if len(f.Groups) != 2 {
+		t.Fatalf("want 2 groups, got %d", len(f.Groups))
+	}
+	names := []string{f.Groups[0].Name, f.Groups[1].Name}
+	if !(contains(names, "included") && contains(names, "local")) {
+		t.Errorf("groups = %v", names)
+	}
+}
+
+func TestMergeBuildsAppended(t *testing.T) {
+	dir := t.TempDir()
+	writeTempJdg(t, dir, "common.jdg", `
+build_defaults {
+    language = "c"
+    standard = "c11"
+    sources  = "*.c"
+    output   = "solution"
+    warnings = strict
+}
+build "release" { profile = release }
+`)
+	mainPath := writeTempJdg(t, dir, "main.jdg", `
+include "common.jdg"
+build "debug" { profile = debug }
+group("g") { weight = 1.0 test("t") { stdout = "ok\n" } }
+`)
+	f, _, err := ParseFile(mainPath)
+	if err != nil {
+		t.Fatalf("parse: %v", err)
+	}
+	if len(f.Builds) != 2 {
+		t.Fatalf("want 2 builds, got %d", len(f.Builds))
+	}
+}
+
+func TestMergeBinarySourcesAndFlags(t *testing.T) {
+	dir := t.TempDir()
+	writeTempJdg(t, dir, "common.jdg", `
+build "make"
+binary = "solution"
+sources = "*.c"
+normalize_crlf = true
+trim_trailing_ws = true
+`)
+	mainPath := writeTempJdg(t, dir, "main.jdg", `
+include "common.jdg"
+group("g") { weight = 1.0 test("t") { stdout = "ok\n" } }
+`)
+	f, _, err := ParseFile(mainPath)
+	if err != nil {
+		t.Fatalf("parse: %v", err)
+	}
+	if f.Binary != "solution" {
+		t.Errorf("Binary = %q, want solution", f.Binary)
+	}
+	if f.Sources != "*.c" {
+		t.Errorf("Sources = %q, want *.c", f.Sources)
+	}
+	if !f.NormalizeCRLF {
+		t.Error("NormalizeCRLF not merged")
+	}
+	if !f.TrimTrailingWS {
+		t.Error("TrimTrailingWS not merged")
+	}
+}
+
+func TestMergeLocalOverridesTimeout(t *testing.T) {
+	dir := t.TempDir()
+	writeTempJdg(t, dir, "common.jdg", `
+build "make"
+timeout 10s
+memory_limit = 128MB
+`)
+	mainPath := writeTempJdg(t, dir, "main.jdg", `
+include "common.jdg"
+timeout 3s
+memory_limit = 64MB
+group("g") { weight = 1.0 test("t") { stdout = "ok\n" } }
+`)
+	f, _, err := ParseFile(mainPath)
+	if err != nil {
+		t.Fatalf("parse: %v", err)
+	}
+	if f.Timeout != 3*time.Second {
+		t.Errorf("Timeout = %v, want local 3s", f.Timeout)
+	}
+	if f.MemoryLimit != 64*1024*1024 {
+		t.Errorf("MemoryLimit = %d, want local 64MB", f.MemoryLimit)
+	}
+}
+
+func contains(xs []string, s string) bool {
+	for _, x := range xs {
+		if x == s {
+			return true
+		}
+	}
+	return false
+}
--- a/dsl/parser.go
+++ b/dsl/parser.go
@@ -3,14 +3,18 @@ package dsl
 import (
 	"fmt"
 	"math"
+	"os"
+	"path/filepath"
 	"strconv"
 	"time"
 )

 type Parser struct {
-	tokens []Token
-	pos    int
-	warns  []string
+	tokens         []Token
+	pos            int
+	warns          []string
+	includeBaseDir string
+	visited        map[string]bool
 }

 func NewParser(tokens []Token) *Parser {
@@ -72,9 +76,67 @@ func Parse(src string) (*File, []string, error) {
 	if err != nil {
 		return nil, parser.Warnings(), err
 	}
+	if err := parser.finalize(file); err != nil {
+		return nil, parser.Warnings(), err
+	}
 	return file, parser.Warnings(), nil
 }

+func ParseFile(path string) (*File, []string, error) {
+	abs, err := filepath.Abs(path)
+	if err != nil {
+		return nil, nil, fmt.Errorf("resolve %q: %w", path, err)
+	}
+	visited := map[string]bool{}
+	parser, file, err := parseFileOnDisk(abs, visited)
+	if err != nil {
+		var warns []string
+		if parser != nil {
+			warns = parser.Warnings()
+		}
+		return nil, warns, err
+	}
+	if err := parser.finalize(file); err != nil {
+		return nil, parser.Warnings(), err
+	}
+	return file, parser.Warnings(), nil
+}
+
+func parseFileOnDisk(absPath string, visited map[string]bool) (*Parser, *File, error) {
+	if visited[absPath] {
+		return nil, nil, fmt.Errorf("circular include: %s", absPath)
+	}
+	visited[absPath] = true
+
+	src, err := os.ReadFile(absPath)
+	if err != nil {
+		return nil, nil, fmt.Errorf("read %s: %w", absPath, err)
+	}
+	tokens, err := NewLexer(string(src)).Tokenize()
+	if err != nil {
+		return nil, nil, fmt.Errorf("%s: %w", absPath, err)
+	}
+	p := NewParser(tokens)
+	p.includeBaseDir = filepath.Dir(absPath)
+	p.visited = visited
+
+	file, err := p.parseFile()
+	if err != nil {
+		return p, nil, fmt.Errorf("%s: %w", absPath, err)
+	}
+	return p, file, nil
+}
+
+func (p *Parser) finalize(f *File) error {
+	if err := p.validateWeights(f); err != nil {
+		return err
+	}
+	if err := validateBuilds(f); err != nil {
+		return err
+	}
+	return nil
+}
+
 func (p *Parser) parseFile() (*File, error) {
 	f := &File{}

@@ -91,7 +153,72 @@ func (p *Parser) parseFile() (*File, error) {
 			if err != nil {
 				return nil, err
 			}
-			f.Build = s.Value
+			if p.peek().Type == TOKEN_LBRACE {
+				bc, err := p.parseBuildBlock(s.Value)
+				if err != nil {
+					return nil, err
+				}
+				f.Builds = append(f.Builds, bc)
+			} else {
+				f.Build = s.Value
+			}
+
+		case "build_defaults":
+			p.advance()
+			bc, err := p.parseBuildBlock("")
+			if err != nil {
+				return nil, err
+			}
+			if f.BuildDefaults == nil {
+				f.BuildDefaults = &BuildConfig{}
+			}
+			f.BuildDefaults.MergeFrom(bc)
+
+		case "toolchains":
+			p.advance()
+			specs, err := p.parseToolchainsBlock()
+			if err != nil {
+				return nil, err
+			}
+			existing := map[string]bool{}
+			for _, tc := range f.Toolchains {
+				existing[tc.Name] = true
+			}
+			for _, tc := range specs {
+				if existing[tc.Name] {
+					return nil, fmt.Errorf("duplicate toolchain %q", tc.Name)
+				}
+				existing[tc.Name] = true
+				f.Toolchains = append(f.Toolchains, tc)
+			}
+
+		case "include":
+			p.advance()
+			s, err := p.expect(TOKEN_STRING)
+			if err != nil {
+				return nil, err
+			}
+			if p.includeBaseDir == "" {
+				return nil, fmt.Errorf("%d:%d: `include` requires file context (use ParseFile, not Parse)", s.Line, s.Col)
+			}
+			target := s.Value
+			if !filepath.IsAbs(target) {
+				target = filepath.Join(p.includeBaseDir, target)
+			}
+			abs, err := filepath.Abs(target)
+			if err != nil {
+				return nil, fmt.Errorf("%d:%d: resolve include %q: %w", s.Line, s.Col, s.Value, err)
+			}
+			childParser, child, err := parseFileOnDisk(abs, p.visited)
+			if err != nil {
+				return nil, fmt.Errorf("%d:%d: include %q: %w", s.Line, s.Col, s.Value, err)
+			}
+			for _, w := range childParser.Warnings() {
+				p.warn(w)
+			}
+			if err := mergeFiles(f, child); err != nil {
+				return nil, fmt.Errorf("%d:%d: include %q: %w", s.Line, s.Col, s.Value, err)
+			}

 		case "build_linux":
 			p.advance()
@@ -192,11 +319,28 @@ func (p *Parser) parseFile() (*File, error) {
 		}
 	}

-	if err := p.validateWeights(f); err != nil {
-		return nil, err
+	return f, nil
+}
+
+func validateBuilds(f *File) error {
+	hasLegacy := f.Build != "" || f.BuildLinux != "" || f.BuildWindows != "" || f.BuildDarwin != ""
+	hasStructured := f.BuildDefaults != nil || len(f.Builds) > 0
+
+	if hasLegacy && hasStructured {
+		return fmt.Errorf("cannot mix legacy `build \"shell\"` with structured `build \"name\" { ... }` in the same suite")
 	}

-	return f, nil
+	seen := map[string]bool{}
+	for _, b := range f.Builds {
+		if b.Name == "" {
+			return fmt.Errorf("structured build must have a name")
+		}
+		if seen[b.Name] {
+			return fmt.Errorf("duplicate build name %q", b.Name)
+		}
+		seen[b.Name] = true
+	}
+	return nil
 }

 func (p *Parser) validateWeights(f *File) error {
@@ -635,25 +779,40 @@ func (p *Parser) parsePattern() (*Pattern, error) {
 		if _, err := p.expect(TOKEN_ASSIGN); err != nil {
 			return nil, err
 		}
-		val, err := p.expect(TOKEN_STRING)
-		if err != nil {
-			return nil, err
-		}
+
 		switch t.Value {
 		case "input":
+			val, err := p.expect(TOKEN_STRING)
+			if err != nil {
+				return nil, err
+			}
 			if pat.DirsGlob != "" {
 				pat.InputFile = val.Value
 			} else {
 				pat.InputGlob = val.Value
 			}
 		case "output":
+			val, err := p.expect(TOKEN_STRING)
+			if err != nil {
+				return nil, err
+			}
 			if pat.DirsGlob != "" {
 				pat.OutputFile = val.Value
 			} else {
 				pat.OutputGlob = val.Value
 			}
 		case "dirs":
+			val, err := p.expect(TOKEN_STRING)
+			if err != nil {
+				return nil, err
+			}
 			pat.DirsGlob = val.Value
+		case "args":
+			xs, err := p.parseStringList()
+			if err != nil {
+				return nil, err
+			}
+			pat.Args = xs
 		default:
 			return nil, fmt.Errorf("%d:%d: unknown pattern field %q", t.Line, t.Col, t.Value)
 		}
@@ -715,8 +874,6 @@ func (p *Parser) parseInt() (int, error) {
 	return n, nil
 }

-// parseSize accepts either a TOKEN_SIZE (e.g. "256MB", "1GiB", "512K") or a bare
-// TOKEN_INT interpreted as bytes. MiB/MB are both 1024² — we use IEC semantics.
 func (p *Parser) parseSize() (int64, error) {
 	t := p.peek()
 	switch t.Type {
--- a/dsl/parser_errors_test.go
+++ b/dsl/parser_errors_test.go
@@ -0,0 +1,299 @@
+package dsl
+
+import (
+	"strings"
+	"testing"
+)
+
+func TestParserErrorCases(t *testing.T) {
+	const prefix = `build "go build ."` + "\n"
+	cases := []struct {
+		name string
+		body string
+		want string
+	}{
+		{
+			name: "group missing lparen",
+			body: `group "g" { weight = 1.0 test("t") { stdout = "" } }`,
+			want: "expected",
+		},
+		{
+			name: "group name not string",
+			body: `group(foo) { weight = 1.0 test("t") { stdout = "" } }`,
+			want: "expected",
+		},
+		{
+			name: "group missing rparen",
+			body: `group("g" { weight = 1.0 test("t") { stdout = "" } }`,
+			want: "expected",
+		},
+		{
+			name: "group missing lbrace",
+			body: `group("g") weight = 1.0`,
+			want: "expected",
+		},
+		{
+			name: "group non-ident in body",
+			body: `group("g") { 42 }`,
+			want: "unexpected token",
+		},
+		{
+			name: "group weight missing assign",
+			body: `group("g") { weight 1.0 test("t") { stdout = "" } }`,
+			want: "expected",
+		},
+		{
+			name: "group weight non-number",
+			body: `group("g") { weight = "bad" test("t") { stdout = "" } }`,
+			want: "",
+		},
+		{
+			name: "group timeout missing assign",
+			body: `group("g") { weight = 1.0 timeout 5s test("t") { stdout = "" } }`,
+			want: "expected",
+		},
+		{
+			name: "group timeout non-duration",
+			body: `group("g") { weight = 1.0 timeout = "5s" test("t") { stdout = "" } }`,
+			want: "",
+		},
+		{
+			name: "group memory_limit missing assign",
+			body: `group("g") { weight = 1.0 memory_limit 64MB test("t") { stdout = "" } }`,
+			want: "expected",
+		},
+		{
+			name: "group memory_limit non-size",
+			body: `group("g") { weight = 1.0 memory_limit = "64MB" test("t") { stdout = "" } }`,
+			want: "",
+		},
+		{
+			name: "group scoring missing assign",
+			body: `group("g") { weight = 1.0 scoring partial test("t") { stdout = "" } }`,
+			want: "expected",
+		},
+		{
+			name: "group scoring non-ident",
+			body: `group("g") { weight = 1.0 scoring = "partial" test("t") { stdout = "" } }`,
+			want: "expected",
+		},
+		{
+			name: "group env missing lparen",
+			body: `group("g") { weight = 1.0 env "K" = "v" test("t") { stdout = "" } }`,
+			want: "expected",
+		},
+		{
+			name: "group env key not string",
+			body: `group("g") { weight = 1.0 env(K) = "v" test("t") { stdout = "" } }`,
+			want: "expected",
+		},
+		{
+			name: "group env missing rparen",
+			body: `group("g") { weight = 1.0 env("K" = "v" test("t") { stdout = "" } }`,
+			want: "expected",
+		},
+		{
+			name: "group env missing assign",
+			body: `group("g") { weight = 1.0 env("K") "v" test("t") { stdout = "" } }`,
+			want: "expected",
+		},
+		{
+			name: "group env value not string",
+			body: `group("g") { weight = 1.0 env("K") = K test("t") { stdout = "" } }`,
+			want: "expected",
+		},
+		{
+			name: "group wrapper missing assign",
+			body: `group("g") { weight = 1.0 wrapper "x" test("t") { stdout = "" } }`,
+			want: "expected",
+		},
+		{
+			name: "group wrapper not string",
+			body: `group("g") { weight = 1.0 wrapper = X test("t") { stdout = "" } }`,
+			want: "expected",
+		},
+		{
+			name: "group unclosed",
+			body: `group("g") { weight = 1.0 test("t") { stdout = "" }`,
+			want: "",
+		},
+		{
+			name: "group pattern error",
+			body: `group("g") { weight = 1.0 pattern { bogus = 1 } }`,
+			want: "",
+		},
+
+		{
+			name: "test missing lparen",
+			body: `group("g") { weight = 1.0 test "t" { stdout = "" } }`,
+			want: "expected",
+		},
+		{
+			name: "test name not string",
+			body: `group("g") { weight = 1.0 test(t) { stdout = "" } }`,
+			want: "expected",
+		},
+		{
+			name: "test missing rparen",
+			body: `group("g") { weight = 1.0 test("t" { stdout = "" } }`,
+			want: "expected",
+		},
+		{
+			name: "test missing lbrace",
+			body: `group("g") { weight = 1.0 test("t") stdout = "" }`,
+			want: "expected",
+		},
+		{
+			name: "test non-ident in body",
+			body: `group("g") { weight = 1.0 test("t") { 42 } }`,
+			want: "unexpected token",
+		},
+		{
+			name: "test stdin missing assign",
+			body: `group("g") { weight = 1.0 test("t") { stdin "x" stdout = "" } }`,
+			want: "expected",
+		},
+		{
+			name: "test stdin not string",
+			body: `group("g") { weight = 1.0 test("t") { stdin = 1 stdout = "" } }`,
+			want: "expected",
+		},
+		{
+			name: "test args missing assign",
+			body: `group("g") { weight = 1.0 test("t") { args "x" stdout = "" } }`,
+			want: "expected",
+		},
+		{
+			name: "test exitCode missing assign",
+			body: `group("g") { weight = 1.0 test("t") { exitCode 0 stdout = "" } }`,
+			want: "expected",
+		},
+		{
+			name: "test exitCode not int",
+			body: `group("g") { weight = 1.0 test("t") { exitCode = "x" stdout = "" } }`,
+			want: "",
+		},
+		{
+			name: "test timeout missing assign",
+			body: `group("g") { weight = 1.0 test("t") { timeout 2s stdout = "" } }`,
+			want: "expected",
+		},
+		{
+			name: "test timeout not duration",
+			body: `group("g") { weight = 1.0 test("t") { timeout = "2s" stdout = "" } }`,
+			want: "",
+		},
+		{
+			name: "test memory_limit missing assign",
+			body: `group("g") { weight = 1.0 test("t") { memory_limit 64MB stdout = "" } }`,
+			want: "expected",
+		},
+		{
+			name: "test memory_limit not size",
+			body: `group("g") { weight = 1.0 test("t") { memory_limit = "64MB" stdout = "" } }`,
+			want: "",
+		},
+		{
+			name: "test wrapper missing assign",
+			body: `group("g") { weight = 1.0 test("t") { wrapper "x" stdout = "" } }`,
+			want: "expected",
+		},
+		{
+			name: "test wrapper not string",
+			body: `group("g") { weight = 1.0 test("t") { wrapper = X stdout = "" } }`,
+			want: "expected",
+		},
+		{
+			name: "test env missing lparen",
+			body: `group("g") { weight = 1.0 test("t") { env "K" = "v" stdout = "" } }`,
+			want: "expected",
+		},
+		{
+			name: "test env key not string",
+			body: `group("g") { weight = 1.0 test("t") { env(K) = "v" stdout = "" } }`,
+			want: "expected",
+		},
+		{
+			name: "test env missing rparen",
+			body: `group("g") { weight = 1.0 test("t") { env("K" = "v" stdout = "" } }`,
+			want: "expected",
+		},
+		{
+			name: "test env missing assign",
+			body: `group("g") { weight = 1.0 test("t") { env("K") "v" stdout = "" } }`,
+			want: "expected",
+		},
+		{
+			name: "test env value not string",
+			body: `group("g") { weight = 1.0 test("t") { env("K") = K stdout = "" } }`,
+			want: "expected",
+		},
+		{
+			name: "test file missing lparen",
+			body: `group("g") { weight = 1.0 test("t") { file "x" = "y" stdout = "" } }`,
+			want: "expected",
+		},
+		{
+			name: "test file key not string",
+			body: `group("g") { weight = 1.0 test("t") { file(x) = "y" stdout = "" } }`,
+			want: "expected",
+		},
+		{
+			name: "test file missing rparen",
+			body: `group("g") { weight = 1.0 test("t") { file("x" = "y" stdout = "" } }`,
+			want: "expected",
+		},
+		{
+			name: "test file missing assign",
+			body: `group("g") { weight = 1.0 test("t") { file("x") "y" stdout = "" } }`,
+			want: "expected",
+		},
+		{
+			name: "test file value not string",
+			body: `group("g") { weight = 1.0 test("t") { file("x") = y stdout = "" } }`,
+			want: "expected",
+		},
+		{
+			name: "test outFile missing lparen",
+			body: `group("g") { weight = 1.0 test("t") { outFile "x" = "y" stdout = "" } }`,
+			want: "expected",
+		},
+		{
+			name: "test outFile key not string",
+			body: `group("g") { weight = 1.0 test("t") { outFile(x) = "y" stdout = "" } }`,
+			want: "expected",
+		},
+		{
+			name: "test outFile missing rparen",
+			body: `group("g") { weight = 1.0 test("t") { outFile("x" = "y" stdout = "" } }`,
+			want: "expected",
+		},
+		{
+			name: "test outFile missing assign",
+			body: `group("g") { weight = 1.0 test("t") { outFile("x") "y" stdout = "" } }`,
+			want: "expected",
+		},
+		{
+			name: "test outFile value not string",
+			body: `group("g") { weight = 1.0 test("t") { outFile("x") = y stdout = "" } }`,
+			want: "expected",
+		},
+		{
+			name: "test unclosed",
+			body: `group("g") { weight = 1.0 test("t") { stdout = ""`,
+			want: "",
+		},
+	}
+
+	for _, c := range cases {
+		t.Run(c.name, func(t *testing.T) {
+			_, _, err := Parse(prefix + c.body)
+			if err == nil {
+				t.Fatalf("expected error, got nil")
+			}
+			if c.want != "" && !strings.Contains(err.Error(), c.want) {
+				t.Errorf("error = %q, want substring %q", err.Error(), c.want)
+			}
+		})
+	}
+}
--- a/dsl/parser_features_test.go
+++ b/dsl/parser_features_test.go
@@ -0,0 +1,237 @@
+package dsl
+
+import (
+	"strings"
+	"testing"
+	"time"
+)
+
+func parseOrFatal(t *testing.T, src string) *File {
+	t.Helper()
+	f, _, err := Parse(src)
+	if err != nil {
+		t.Fatalf("parse error: %v", err)
+	}
+	return f
+}
+
+func parseExpectError(t *testing.T, src, wantSubstr string) {
+	t.Helper()
+	_, _, err := Parse(src)
+	if err == nil {
+		t.Fatalf("expected error containing %q, got nil", wantSubstr)
+	}
+	if !strings.Contains(err.Error(), wantSubstr) {
+		t.Errorf("error = %q, want substring %q", err.Error(), wantSubstr)
+	}
+}
+
+func TestParseTestFileAndOutFile(t *testing.T) {
+	src := `
+build "go build ."
+
+group("g") {
+    weight = 1.0
+    test("files") {
+        file("input.txt") = "1 2 3\n"
+        outFile("result.txt") = "6\n"
+        stdout = ""
+    }
+}
+`
+	f := parseOrFatal(t, src)
+	tst := f.Groups[0].Tests[0]
+	if tst.InFiles["input.txt"] != "1 2 3\n" {
+		t.Errorf("InFiles[input.txt] = %q", tst.InFiles["input.txt"])
+	}
+	if tst.OutFiles["result.txt"] != "6\n" {
+		t.Errorf("OutFiles[result.txt] = %q", tst.OutFiles["result.txt"])
+	}
+}
+
+func TestParseTestEnvAndWrapper(t *testing.T) {
+	src := `
+build "go build ."
+
+group("g") {
+    weight = 1.0
+    test("env") {
+        env("FOO") = "bar"
+        env("BAZ") = "qux"
+        wrapper = "valgrind"
+        stdout = ""
+    }
+}
+`
+	f := parseOrFatal(t, src)
+	tst := f.Groups[0].Tests[0]
+	if tst.Env["FOO"] != "bar" || tst.Env["BAZ"] != "qux" {
+		t.Errorf("Env = %v", tst.Env)
+	}
+	if tst.Wrapper != "valgrind" {
+		t.Errorf("Wrapper = %q", tst.Wrapper)
+	}
+}
+
+func TestParseTestTimeoutAndMemoryOverride(t *testing.T) {
+	src := `
+build "go build ."
+timeout 10s
+
+group("g") {
+    weight = 1.0
+    memory_limit = 256MB
+    test("override") {
+        timeout = 2s
+        memory_limit = 64MB
+        stdout = ""
+    }
+}
+`
+	f := parseOrFatal(t, src)
+	tst := f.Groups[0].Tests[0]
+	if tst.Timeout != 2*time.Second {
+		t.Errorf("Timeout = %v, want 2s", tst.Timeout)
+	}
+	if tst.MemoryLimit != 64*1024*1024 {
+		t.Errorf("MemoryLimit = %d, want %d", tst.MemoryLimit, 64*1024*1024)
+	}
+}
+
+func TestParseTestExitCodeNonZero(t *testing.T) {
+	src := `
+build "go build ."
+
+group("g") {
+    weight = 1.0
+    test("fail") {
+        exitCode = 42
+        stdout = ""
+    }
+}
+`
+	f := parseOrFatal(t, src)
+	tst := f.Groups[0].Tests[0]
+	if tst.ExitCode == nil || *tst.ExitCode != 42 {
+		t.Errorf("ExitCode = %v, want 42", tst.ExitCode)
+	}
+}
+
+func TestParseTestUnknownKeyword(t *testing.T) {
+	src := `
+build "go build ."
+group("g") {
+    weight = 1.0
+    test("bad") {
+        bogus = "x"
+    }
+}
+`
+	parseExpectError(t, src, `"bogus"`)
+}
+
+func TestParseGroupScoringPartial(t *testing.T) {
+	src := `
+build "go build ."
+group("g") {
+    weight = 1.0
+    scoring = partial
+    test("t") { stdout = "" }
+}
+`
+	f := parseOrFatal(t, src)
+	if f.Groups[0].Scoring != ScoringPartial {
+		t.Errorf("Scoring = %v, want partial", f.Groups[0].Scoring)
+	}
+}
+
+func TestParseGroupScoringAllOrNone(t *testing.T) {
+	src := `
+build "go build ."
+group("g") {
+    weight = 1.0
+    scoring = all_or_none
+    test("t") { stdout = "" }
+}
+`
+	f := parseOrFatal(t, src)
+	if f.Groups[0].Scoring != ScoringAllOrNone {
+		t.Errorf("Scoring = %v, want all_or_none", f.Groups[0].Scoring)
+	}
+}
+
+func TestParseGroupScoringUnknown(t *testing.T) {
+	src := `
+build "go build ."
+group("g") {
+    weight = 1.0
+    scoring = magic
+    test("t") { stdout = "" }
+}
+`
+	parseExpectError(t, src, "unknown scoring mode")
+}
+
+func TestParseGroupEnvAndWrapper(t *testing.T) {
+	src := `
+build "go build ."
+group("g") {
+    weight = 1.0
+    env("LANG") = "C"
+    env("DEBUG") = "1"
+    wrapper = "strace"
+    test("t") { stdout = "" }
+}
+`
+	f := parseOrFatal(t, src)
+	g := f.Groups[0]
+	if g.Env["LANG"] != "C" || g.Env["DEBUG"] != "1" {
+		t.Errorf("group Env = %v", g.Env)
+	}
+	if g.Wrapper != "strace" {
+		t.Errorf("group Wrapper = %q", g.Wrapper)
+	}
+}
+
+func TestParseGroupMemoryLimit(t *testing.T) {
+	src := `
+build "go build ."
+group("g") {
+    weight = 1.0
+    memory_limit = 128MB
+    test("t") { stdout = "" }
+}
+`
+	f := parseOrFatal(t, src)
+	if f.Groups[0].MemoryLimit != 128*1024*1024 {
+		t.Errorf("group MemoryLimit = %d", f.Groups[0].MemoryLimit)
+	}
+}
+
+func TestParseGroupUnknownKeyword(t *testing.T) {
+	src := `
+build "go build ."
+group("g") {
+    weight = 1.0
+    foobar = 1
+    test("t") { stdout = "" }
+}
+`
+	parseExpectError(t, src, `"foobar"`)
+}
+
+func TestParseGroupMissingWeight(t *testing.T) {
+	src := `
+build "go build ."
+group("g") {
+    test("t") { stdout = "" }
+}
+`
+	_, warns, err := Parse(src)
+	if err != nil && !strings.Contains(err.Error(), "weight") {
+		return
+	}
+	if err == nil && len(warns) == 0 {
+		t.Error("expected error or warning about missing weight")
+	}
+}
--- a/dsl/parser_misc_test.go
+++ b/dsl/parser_misc_test.go
@@ -0,0 +1,193 @@
+package dsl
+
+import (
+	"strings"
+	"testing"
+)
+
+func TestParsePatternDirsMode(t *testing.T) {
+	src := `
+build "make"
+group("g") {
+    weight = 1.0
+    pattern {
+        dirs   = "tests/*"
+        input  = "in.txt"
+        output = "out.txt"
+        args   = "--case" "{name}"
+    }
+}
+`
+	f, _, err := Parse(src)
+	if err != nil {
+		t.Fatalf("parse: %v", err)
+	}
+	pat := f.Groups[0].Pattern
+	if pat == nil {
+		t.Fatal("no pattern")
+	}
+	if pat.DirsGlob != "tests/*" {
+		t.Errorf("DirsGlob = %q", pat.DirsGlob)
+	}
+	if pat.InputFile != "in.txt" || pat.OutputFile != "out.txt" {
+		t.Errorf("InputFile/OutputFile = %q/%q", pat.InputFile, pat.OutputFile)
+	}
+	if len(pat.Args) != 2 || pat.Args[0] != "--case" || pat.Args[1] != "{name}" {
+		t.Errorf("Args = %v", pat.Args)
+	}
+}
+
+func TestParsePatternUnknownField(t *testing.T) {
+	src := `
+build "make"
+group("g") {
+    weight = 1.0
+    pattern { bogus = "x" }
+}
+`
+	_, _, err := Parse(src)
+	if err == nil || !strings.Contains(err.Error(), "unknown pattern field") {
+		t.Errorf("want unknown pattern field error, got %v", err)
+	}
+}
+
+func TestParsePatternNonIdent(t *testing.T) {
+	src := `
+build "make"
+group("g") {
+    weight = 1.0
+    pattern { "x" = "y" }
+}
+`
+	_, _, err := Parse(src)
+	if err == nil {
+		t.Error("expected error on non-ident in pattern")
+	}
+}
+
+func TestParseNormalizeCRLFAndTrimWS(t *testing.T) {
+	src := `
+build "make"
+normalize_crlf = true
+trim_trailing_ws = false
+group("g") { weight = 1.0 test("t") { stdout = "" } }
+`
+	f, _, err := Parse(src)
+	if err != nil {
+		t.Fatalf("parse: %v", err)
+	}
+	if !f.NormalizeCRLF {
+		t.Error("NormalizeCRLF should be true")
+	}
+	if f.TrimTrailingWS {
+		t.Error("TrimTrailingWS should be false")
+	}
+}
+
+func TestParseBinaryAndSources(t *testing.T) {
+	src := `
+build "make"
+binary = "sol"
+sources = "main.c"
+group("g") { weight = 1.0 test("t") { stdout = "" } }
+`
+	f, _, err := Parse(src)
+	if err != nil {
+		t.Fatalf("parse: %v", err)
+	}
+	if f.Binary != "sol" {
+		t.Errorf("Binary = %q", f.Binary)
+	}
+	if f.Sources != "main.c" {
+		t.Errorf("Sources = %q", f.Sources)
+	}
+}
+
+func TestParseMemoryLimitBareInt(t *testing.T) {
+	src := `
+build "make"
+memory_limit = 1024
+group("g") { weight = 1.0 test("t") { stdout = "" } }
+`
+	f, _, err := Parse(src)
+	if err != nil {
+		t.Fatalf("parse: %v", err)
+	}
+	if f.MemoryLimit != 1024 {
+		t.Errorf("MemoryLimit = %d, want 1024", f.MemoryLimit)
+	}
+}
+
+func TestParseBoolInvalidIdent(t *testing.T) {
+	_, _, err := Parse("build \"make\"\nnormalize_crlf = maybe\n")
+	if err == nil || !strings.Contains(err.Error(), "true/false") {
+		t.Errorf("want true/false error, got %v", err)
+	}
+}
+
+func TestParseBoolNonIdent(t *testing.T) {
+	_, _, err := Parse("build \"make\"\nnormalize_crlf = \"true\"\n")
+	if err == nil || !strings.Contains(err.Error(), "true/false") {
+		t.Errorf("want true/false error, got %v", err)
+	}
+}
+
+func TestParseTopLevelNonIdent(t *testing.T) {
+	_, _, err := Parse(`"stray"`)
+	if err == nil || !strings.Contains(err.Error(), "unexpected token") {
+		t.Errorf("want unexpected token error, got %v", err)
+	}
+}
+
+func TestParseMatcherOrAssignNoMatcher(t *testing.T) {
+	src := `
+build "make"
+group("g") {
+    weight = 1.0
+    test("t") { stdout 42 }
+}
+`
+	_, _, err := Parse(src)
+	if err == nil || !strings.Contains(err.Error(), "expected matcher") {
+		t.Errorf("want matcher error, got %v", err)
+	}
+}
+
+func TestParseStringListEmpty(t *testing.T) {
+	src := `
+build "make"
+group("g") {
+    weight = 1.0
+    test("t") { args = stdout = "" }
+}
+`
+	_, _, err := Parse(src)
+	if err == nil {
+		t.Error("expected error on empty string list")
+	}
+}
+
+func TestParseFileMissing(t *testing.T) {
+	_, _, err := ParseFile("/nonexistent/judge-test-does-not-exist.jdg")
+	if err == nil {
+		t.Error("expected error on missing file")
+	}
+}
+
+func TestValidateBuildsMixedLegacyAndStructured(t *testing.T) {
+	src := `
+build "legacy shell"
+build_defaults {
+    language = "c"
+    standard = "c11"
+    sources  = "*.c"
+    output   = "sol"
+    warnings = strict
+}
+group("g") { weight = 1.0 test("t") { stdout = "" } }
+`
+	_, _, err := Parse(src)
+	if err == nil || !strings.Contains(err.Error(), "cannot mix legacy") {
+		t.Errorf("want mix-legacy error, got %v", err)
+	}
+}
--- a/editor/vscode-jdg/jdg-language-0.1.0.vsix
+++ b/editor/vscode-jdg/jdg-language-0.1.0.vsix
--- a/editor/vscode-jdg/syntaxes/jdg.tmLanguage.json
+++ b/editor/vscode-jdg/syntaxes/jdg.tmLanguage.json
@@ -7,13 +7,19 @@
    { "include": "#comments" },
    { "include": "#heredoc" },
    { "include": "#string" },
+    { "include": "#size" },
    { "include": "#duration" },
    { "include": "#number" },
    { "include": "#boolean" },
    { "include": "#top-keywords" },
+    { "include": "#os-block" },
+    { "include": "#build-fields" },
    { "include": "#block-keywords" },
    { "include": "#matcher-keywords" },
+    { "include": "#profile-values" },
+    { "include": "#warnings-values" },
    { "include": "#scoring-values" },
+    { "include": "#class-values" },
    { "include": "#operators" },
    { "include": "#braces" },
    { "include": "#identifier" }
@@ -39,6 +45,10 @@
        { "name": "constant.character.escape.jdg", "match": "\\\\[nt\\\\\"]" }
      ]
    },
+    "size": {
+      "name": "constant.numeric.size.jdg",
+      "match": "\\b\\d+(?:KiB|MiB|GiB|KB|MB|GB|B|K|M|G)\\b"
+    },
    "duration": {
      "name": "constant.numeric.duration.jdg",
      "match": "\\b-?\\d+(?:\\.\\d+)?(?:ms|s|m|h)\\b"
@@ -53,20 +63,40 @@
    },
    "top-keywords": {
      "name": "keyword.control.jdg",
-      "match": "\\b(?:build|build_linux|build_windows|build_darwin|binary|sources|timeout|normalize_crlf|trim_trailing_ws|group|test|pattern|env|file|outFile)\\b"
+      "match": "\\b(?:include|build|build_defaults|build_linux|build_windows|build_darwin|toolchains|binary|sources|timeout|memory_limit|normalize_crlf|trim_trailing_ws|group|test|pattern|env|file|outFile|define)\\b"
+    },
+    "os-block": {
+      "name": "support.type.os.jdg",
+      "match": "\\b(?:linux|windows|darwin)\\b(?=\\s*\\{)"
+    },
+    "build-fields": {
+      "name": "variable.parameter.build.jdg",
+      "match": "\\b(?:language|standard|output|wrapper|includes|sanitize|link|extra|platforms|compilers|profile|warnings|class)\\b"
    },
    "block-keywords": {
      "name": "variable.parameter.jdg",
-      "match": "\\b(?:weight|scoring|wrapper|stdin|stdout|stderr|args|exitCode|input|output|dirs)\\b"
+      "match": "\\b(?:weight|scoring|stdin|stdout|stderr|args|exitCode|input|dirs)\\b"
    },
    "matcher-keywords": {
      "name": "support.function.jdg",
      "match": "\\b(?:contains|matches|anyOrder|of)\\b"
    },
+    "profile-values": {
+      "name": "constant.language.profile.jdg",
+      "match": "\\b(?:release|debug|sanitized)\\b"
+    },
+    "warnings-values": {
+      "name": "constant.language.warnings.jdg",
+      "match": "\\b(?:default|strict|pedantic)\\b"
+    },
    "scoring-values": {
-      "name": "constant.language.jdg",
+      "name": "constant.language.scoring.jdg",
      "match": "\\b(?:partial|all_or_none)\\b"
    },
+    "class-values": {
+      "name": "constant.language.class.jdg",
+      "match": "\\b(?:gnu|msvc)\\b"
+    },
    "operators": {
      "name": "keyword.operator.jdg",
      "match": "[=~]"
--- a/editor/zed/.gitignore
+++ b/editor/zed/.gitignore
@@ -0,0 +1,17 @@
+grammars/*.wasm
+
+tree-sitter-jdg/tree-sitter-jdg.wasm
+tree-sitter-jdg/node_modules/
+tree-sitter-jdg/bindings/
+tree-sitter-jdg/binding.gyp
+tree-sitter-jdg/Cargo.toml
+tree-sitter-jdg/Cargo.lock
+tree-sitter-jdg/Makefile
+tree-sitter-jdg/Package.swift
+tree-sitter-jdg/pyproject.toml
+tree-sitter-jdg/setup.py
+tree-sitter-jdg/go.mod
+tree-sitter-jdg/go.sum
+tree-sitter-jdg/bindings.gyp
+tree-sitter-jdg/prebuilds/
+tree-sitter-jdg/target/
--- a/editor/zed/extension.toml
+++ b/editor/zed/extension.toml
@@ -0,0 +1,23 @@
+id = "jdg"
+name = "JDG"
+version = "0.1.0"
+schema_version = 1
+description = "Syntax highlighting for the judge DSL (.jdg)"
+repository = "https://github.com/Mond1c/judge"
+authors = ["judge contributors"]
+themes = []
+icon_themes = []
+languages = ["languages/jdg"]
+capabilities = []
+
+[lib]
+
+[grammars.jdg]
+repository = "https://github.com/Mond1c/judge"
+rev = "main"
+
+[language_servers]
+
+[context_servers]
+
+[slash_commands]
--- a/editor/zed/languages/jdg/brackets.scm
+++ b/editor/zed/languages/jdg/brackets.scm
@@ -0,0 +1,2 @@
+("{" @open "}" @close)
+("(" @open ")" @close)
--- a/editor/zed/languages/jdg/config.toml
+++ b/editor/zed/languages/jdg/config.toml
@@ -0,0 +1,12 @@
+name = "JDG"
+grammar = "jdg"
+path_suffixes = ["jdg"]
+line_comments = ["// "]
+autoclose_before = ";:.,=}])>"
+brackets = [
+  { start = "{", end = "}", close = true, newline = true },
+  { start = "(", end = ")", close = true, newline = false },
+  { start = "\"", end = "\"", close = true, newline = false, not_in = ["string", "comment"] },
+]
+word_characters = ["_"]
+tab_size = 4
--- a/editor/zed/languages/jdg/highlights.scm
+++ b/editor/zed/languages/jdg/highlights.scm
@@ -0,0 +1,100 @@
+(comment) @comment
+
+[
+  "include"
+  "build"
+  "build_defaults"
+  "build_linux"
+  "build_windows"
+  "build_darwin"
+  "toolchains"
+  "binary"
+  "sources"
+  "timeout"
+  "memory_limit"
+  "group"
+  "test"
+  "pattern"
+] @keyword
+
+[
+  "normalize_crlf"
+  "trim_trailing_ws"
+] @keyword
+
+(build_scalar_field  name: _ @property)
+(build_list_field    name: _ @property)
+(bool_decl           name: _ @property)
+(legacy_build_platform keyword: _ @keyword)
+
+[
+  "profile"
+  "warnings"
+  "define"
+  "platforms"
+  "binary"
+  "class"
+  "weight"
+  "scoring"
+  "wrapper"
+  "env"
+  "stdin"
+  "stdout"
+  "stderr"
+  "args"
+  "exitCode"
+  "file"
+  "outFile"
+  "input"
+  "output"
+  "dirs"
+] @property
+
+[
+  "linux"
+  "windows"
+  "darwin"
+] @type.builtin
+
+(profile_value)  @constant.builtin
+(warnings_value) @constant.builtin
+(scoring_value)  @constant.builtin
+(class_value)    @constant.builtin
+(bool)           @constant.builtin
+
+"of" @keyword.operator
+
+[
+  "contains"
+  "matches"
+  "anyOrder"
+] @function.builtin
+
+[
+  "="
+  "~"
+] @operator
+
+[
+  "{"
+  "}"
+  "("
+  ")"
+] @punctuation.bracket
+
+(simple_string)  @string
+(heredoc_string) @string
+(regex_string)   @string.regex
+
+(structured_build name: (string) @string.special)
+(legacy_build     command: (string) @string.special)
+(toolchain_entry  name: (string) @type)
+(toolchain_entry  name: (identifier) @type)
+(group            name: (string) @string.special)
+(test             name: (string) @string.special)
+(include          path: (string) @string.special.path)
+
+(define_field  key: (string) @constant.macro)
+(env_decl      key: (string) @constant.macro)
+(file_decl     key: (string) @string.special.path)
+(out_file_decl key: (string) @string.special.path)
--- a/editor/zed/languages/jdg/indents.scm
+++ b/editor/zed/languages/jdg/indents.scm
@@ -0,0 +1,11 @@
+[
+  (build_block)
+  (toolchains)
+  (toolchain_entry)
+  (group)
+  (test)
+  (pattern)
+  (any_order_matcher)
+] @indent
+
+"}" @end
--- a/editor/zed/languages/jdg/injections.scm
+++ b/editor/zed/languages/jdg/injections.scm
@@ -0,0 +1,7 @@
+((comment) @injection.content
+  (#set! injection.language "comment"))
+
+((regex_matcher
+  pattern: (regex_string) @injection.content)
+  (#set! injection.language "regex")
+  (#set! injection.include-children))
--- a/editor/zed/languages/jdg/outline.scm
+++ b/editor/zed/languages/jdg/outline.scm
@@ -0,0 +1,23 @@
+(structured_build
+  "build" @context
+  name: (string) @name) @item
+
+(build_defaults
+  "build_defaults" @context) @item
+
+(toolchains
+  "toolchains" @context) @item
+
+(toolchain_entry
+  name: (_) @name) @item
+
+(group
+  "group" @context
+  name: (string) @name) @item
+
+(test
+  "test" @context
+  name: (string) @name) @item
+
+(pattern
+  "pattern" @context) @item
--- a/editor/zed/tree-sitter-jdg/.editorconfig
+++ b/editor/zed/tree-sitter-jdg/.editorconfig
@@ -0,0 +1,39 @@
+root = true
+
+[*]
+charset = utf-8
+end_of_line = lf
+insert_final_newline = true
+trim_trailing_whitespace = true
+
+[*.{json,toml,yml,gyp}]
+indent_style = space
+indent_size = 2
+
+[*.js]
+indent_style = space
+indent_size = 2
+
+[*.rs]
+indent_style = space
+indent_size = 4
+
+[*.{c,cc,h}]
+indent_style = space
+indent_size = 4
+
+[*.{py,pyi}]
+indent_style = space
+indent_size = 4
+
+[*.swift]
+indent_style = space
+indent_size = 4
+
+[*.go]
+indent_style = tab
+indent_size = 8
+
+[Makefile]
+indent_style = tab
+indent_size = 8
--- a/editor/zed/tree-sitter-jdg/.gitattributes
+++ b/editor/zed/tree-sitter-jdg/.gitattributes
@@ -0,0 +1,11 @@
+* text eol=lf
+
+src/*.json linguist-generated
+src/parser.c linguist-generated
+src/tree_sitter/* linguist-generated
+
+bindings/** linguist-generated
+binding.gyp linguist-generated
+setup.py linguist-generated
+Makefile linguist-generated
+Package.swift linguist-generated
--- a/editor/zed/tree-sitter-jdg/.gitignore
+++ b/editor/zed/tree-sitter-jdg/.gitignore
@@ -0,0 +1,38 @@
+# Rust artifacts
+Cargo.lock
+target/
+
+# Node artifacts
+build/
+prebuilds/
+node_modules/
+*.tgz
+
+# Swift artifacts
+.build/
+
+# Go artifacts
+go.sum
+_obj/
+
+# Python artifacts
+.venv/
+dist/
+*.egg-info
+*.whl
+
+# C artifacts
+*.a
+*.so
+*.so.*
+*.dylib
+*.dll
+*.pc
+
+# Example dirs
+/examples/*/
+
+# Grammar volatiles
+*.wasm
+*.obj
+*.o
--- a/editor/zed/tree-sitter-jdg/grammar.js
+++ b/editor/zed/tree-sitter-jdg/grammar.js
@@ -0,0 +1,342 @@
+module.exports = grammar({
+  name: 'jdg',
+
+  extras: $ => [/\s+/, $.comment],
+
+  word: $ => $.identifier,
+
+  conflicts: $ => [],
+
+  rules: {
+    source_file: $ => repeat($._top_statement),
+
+    comment: $ => token(seq('//', /[^\n]*/)),
+
+    _top_statement: $ => choice(
+      $.include,
+      $.toolchains,
+      $.build_defaults,
+      $.structured_build,
+      $.legacy_build_platform,
+      $.legacy_build,
+      $.binary_decl,
+      $.sources_decl,
+      $.bool_decl,
+      $.timeout_decl,
+      $.memory_limit_decl,
+      $.group,
+    ),
+
+    include: $ => seq(
+      'include',
+      field('path', $.string)
+    ),
+
+    legacy_build: $ => prec(1, seq(
+      'build',
+      field('command', $.string)
+    )),
+
+    legacy_build_platform: $ => seq(
+      field('keyword', choice('build_linux', 'build_windows', 'build_darwin')),
+      field('command', $.string)
+    ),
+
+    structured_build: $ => prec(2, seq(
+      'build',
+      field('name', $.string),
+      field('body', $.build_block)
+    )),
+
+    build_defaults: $ => seq(
+      'build_defaults',
+      field('body', $.build_block)
+    ),
+
+    build_block: $ => seq(
+      '{',
+      repeat($._build_field),
+      '}'
+    ),
+
+    _build_field: $ => choice(
+      $.build_scalar_field,
+      $.build_list_field,
+      $.build_profile_field,
+      $.build_warnings_field,
+      $.define_field,
+      $.os_override,
+    ),
+
+    build_scalar_field: $ => seq(
+      field('name', choice('language', 'standard', 'output', 'wrapper')),
+      '=',
+      field('value', $.string)
+    ),
+
+    build_list_field: $ => seq(
+      field('name', choice('sources', 'includes', 'sanitize', 'link', 'extra', 'platforms', 'compilers')),
+      '=',
+      field('value', repeat1($.string))
+    ),
+
+    build_profile_field: $ => seq(
+      'profile',
+      '=',
+      field('value', $.profile_value)
+    ),
+    profile_value: $ => choice('release', 'debug', 'sanitized'),
+
+    build_warnings_field: $ => seq(
+      'warnings',
+      '=',
+      field('value', $.warnings_value)
+    ),
+    warnings_value: $ => choice('default', 'strict', 'pedantic'),
+
+    define_field: $ => seq(
+      'define',
+      '(',
+      field('key', $.string),
+      ')',
+      '=',
+      field('value', $.string)
+    ),
+
+    os_override: $ => seq(
+      field('os', $.os_name),
+      field('body', $.build_block)
+    ),
+    os_name: $ => choice('linux', 'windows', 'darwin'),
+
+    toolchains: $ => seq(
+      'toolchains',
+      '{',
+      repeat($.toolchain_entry),
+      '}'
+    ),
+
+    toolchain_entry: $ => seq(
+      field('name', choice($.identifier, $.string)),
+      '{',
+      repeat($._toolchain_field),
+      '}'
+    ),
+
+    _toolchain_field: $ => choice(
+      $.toolchain_platforms_field,
+      $.toolchain_binary_field,
+      $.toolchain_class_field,
+    ),
+
+    toolchain_platforms_field: $ => seq(
+      'platforms',
+      '=',
+      field('value', repeat1($.string))
+    ),
+    toolchain_binary_field: $ => seq(
+      'binary',
+      '=',
+      field('value', $.string)
+    ),
+    toolchain_class_field: $ => seq(
+      'class',
+      '=',
+      field('value', $.class_value)
+    ),
+    class_value: $ => choice('gnu', 'msvc'),
+
+    binary_decl: $ => seq(
+      'binary',
+      '=',
+      field('value', $.string)
+    ),
+
+    sources_decl: $ => seq(
+      'sources',
+      '=',
+      field('value', $.string)
+    ),
+
+    bool_decl: $ => seq(
+      field('name', choice('normalize_crlf', 'trim_trailing_ws')),
+      '=',
+      field('value', $.bool)
+    ),
+
+    timeout_decl: $ => seq(
+      'timeout',
+      field('value', $.duration)
+    ),
+
+    memory_limit_decl: $ => seq(
+      'memory_limit',
+      '=',
+      field('value', $.size)
+    ),
+
+    group: $ => seq(
+      'group',
+      '(',
+      field('name', $.string),
+      ')',
+      '{',
+      repeat($._group_field),
+      '}'
+    ),
+
+    _group_field: $ => choice(
+      $.group_weight_field,
+      $.group_timeout_field,
+      $.group_memory_field,
+      $.group_scoring_field,
+      $.group_wrapper_field,
+      $.env_decl,
+      $.test,
+      $.pattern,
+    ),
+
+    group_weight_field: $ => seq('weight', '=', field('value', $.number)),
+    group_timeout_field: $ => seq('timeout', '=', field('value', $.duration)),
+    group_memory_field: $ => seq('memory_limit', '=', field('value', $.size)),
+    group_scoring_field: $ => seq('scoring', '=', field('value', $.scoring_value)),
+    scoring_value: $ => choice('partial', 'all_or_none'),
+    group_wrapper_field: $ => seq('wrapper', '=', field('value', $.string)),
+
+    env_decl: $ => seq(
+      'env',
+      '(',
+      field('key', $.string),
+      ')',
+      '=',
+      field('value', $.string)
+    ),
+
+    test: $ => seq(
+      'test',
+      '(',
+      field('name', $.string),
+      ')',
+      '{',
+      repeat($._test_field),
+      '}'
+    ),
+
+    _test_field: $ => choice(
+      $.test_stdin_field,
+      $.test_stdout_field,
+      $.test_stderr_field,
+      $.test_args_field,
+      $.test_exit_code_field,
+      $.test_timeout_field,
+      $.test_memory_field,
+      $.test_wrapper_field,
+      $.env_decl,
+      $.file_decl,
+      $.out_file_decl,
+    ),
+
+    test_stdin_field: $ => seq('stdin', '=', field('value', $.string)),
+    test_stdout_field: $ => seq('stdout', $._matcher),
+    test_stderr_field: $ => seq('stderr', $._matcher),
+    test_args_field: $ => seq('args', '=', field('value', repeat1($.string))),
+    test_exit_code_field: $ => seq('exitCode', '=', field('value', $.integer)),
+    test_timeout_field: $ => seq('timeout', '=', field('value', $.duration)),
+    test_memory_field: $ => seq('memory_limit', '=', field('value', $.size)),
+    test_wrapper_field: $ => seq('wrapper', '=', field('value', $.string)),
+
+    file_decl: $ => seq(
+      'file',
+      '(',
+      field('key', $.string),
+      ')',
+      '=',
+      field('value', $.string)
+    ),
+
+    out_file_decl: $ => seq(
+      'outFile',
+      '(',
+      field('key', $.string),
+      ')',
+      '=',
+      field('value', $.string)
+    ),
+
+    _matcher: $ => choice(
+      $.exact_matcher,
+      $.numeric_matcher,
+      $.contains_matcher,
+      $.regex_matcher,
+      $.any_order_matcher,
+    ),
+
+    exact_matcher: $ => seq('=', field('value', $.string)),
+    numeric_matcher: $ => seq(
+      '~',
+      field('epsilon', $.number),
+      'of',
+      field('value', $.string)
+    ),
+    contains_matcher: $ => seq('contains', field('value', $.string)),
+    regex_matcher: $ => seq('matches', field('pattern', $.regex_string)),
+    any_order_matcher: $ => seq(
+      'anyOrder',
+      '{',
+      repeat1($.string),
+      '}'
+    ),
+
+    pattern: $ => seq(
+      'pattern',
+      '{',
+      repeat($._pattern_field),
+      '}'
+    ),
+
+    _pattern_field: $ => choice(
+      $.pattern_input_field,
+      $.pattern_output_field,
+      $.pattern_dirs_field,
+      $.pattern_args_field,
+    ),
+
+    pattern_input_field: $ => seq('input', '=', field('value', $.string)),
+    pattern_output_field: $ => seq('output', '=', field('value', $.string)),
+    pattern_dirs_field: $ => seq('dirs', '=', field('value', $.string)),
+    pattern_args_field: $ => seq('args', '=', field('value', repeat1($.string))),
+
+    bool: $ => choice('true', 'false'),
+
+    number: $ => choice($.integer, $.float),
+
+    string: $ => choice(
+      $.heredoc_string,
+      $.simple_string,
+    ),
+
+    simple_string: $ => token(seq(
+      '"',
+      repeat(choice(
+        /[^"\\\n]/,
+        seq('\\', /./)
+      )),
+      '"'
+    )),
+
+    heredoc_string: $ => token(seq(
+      '"""',
+      /([^"]|"[^"]|""[^"])*/,
+      '"""'
+    )),
+
+    regex_string: $ => $.simple_string,
+
+    identifier: $ => /[a-zA-Z_][a-zA-Z0-9_]*/,
+
+    integer: $ => token(/-?\d+/),
+    float: $ => token(/-?\d+\.\d+/),
+    duration: $ => token(/\d+(ms|s|m)/),
+    size: $ => token(/\d+(KiB|MiB|GiB|KB|MB|GB|B|K|M|G)/),
+  }
+});
--- a/editor/zed/tree-sitter-jdg/package.json
+++ b/editor/zed/tree-sitter-jdg/package.json
@@ -0,0 +1,54 @@
+{
+  "name": "tree-sitter-jdg",
+  "version": "0.1.0",
+  "description": "Tree-sitter grammar for the judge DSL (.jdg)",
+  "main": "bindings/node",
+  "types": "bindings/node",
+  "keywords": [
+    "tree-sitter",
+    "parser",
+    "jdg",
+    "judge"
+  ],
+  "files": [
+    "grammar.js",
+    "binding.gyp",
+    "prebuilds/**",
+    "bindings/node/*",
+    "queries/*",
+    "src/**"
+  ],
+  "license": "MIT",
+  "dependencies": {
+    "node-addon-api": "^7.1.0",
+    "node-gyp-build": "^4.8.0"
+  },
+  "peerDependencies": {
+    "tree-sitter": "^0.21.0"
+  },
+  "peerDependenciesMeta": {
+    "tree_sitter": {
+      "optional": true
+    }
+  },
+  "devDependencies": {
+    "tree-sitter-cli": "^0.22.0",
+    "prebuildify": "^6.0.0"
+  },
+  "scripts": {
+    "build": "tree-sitter generate",
+    "test": "tree-sitter test",
+    "install": "node-gyp-build",
+    "prebuildify": "prebuildify --napi --strip"
+  },
+  "tree-sitter": [
+    {
+      "scope": "source.jdg",
+      "file-types": [
+        "jdg"
+      ],
+      "highlights": "queries/highlights.scm",
+      "injections": "queries/injections.scm"
+    }
+  ]
+}
--- a/editor/zed/tree-sitter-jdg/src/grammar.json
+++ b/editor/zed/tree-sitter-jdg/src/grammar.json
--- a/editor/zed/tree-sitter-jdg/src/node-types.json
+++ b/editor/zed/tree-sitter-jdg/src/node-types.json
--- a/editor/zed/tree-sitter-jdg/src/parser.c
+++ b/editor/zed/tree-sitter-jdg/src/parser.c
--- a/editor/zed/tree-sitter-jdg/src/tree_sitter/alloc.h
+++ b/editor/zed/tree-sitter-jdg/src/tree_sitter/alloc.h
@@ -0,0 +1,54 @@
+#ifndef TREE_SITTER_ALLOC_H_
+#define TREE_SITTER_ALLOC_H_
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include <stdbool.h>
+#include <stdio.h>
+#include <stdlib.h>
+
+// Allow clients to override allocation functions
+#ifdef TREE_SITTER_REUSE_ALLOCATOR
+
+extern void *(*ts_current_malloc)(size_t);
+extern void *(*ts_current_calloc)(size_t, size_t);
+extern void *(*ts_current_realloc)(void *, size_t);
+extern void (*ts_current_free)(void *);
+
+#ifndef ts_malloc
+#define ts_malloc  ts_current_malloc
+#endif
+#ifndef ts_calloc
+#define ts_calloc  ts_current_calloc
+#endif
+#ifndef ts_realloc
+#define ts_realloc ts_current_realloc
+#endif
+#ifndef ts_free
+#define ts_free    ts_current_free
+#endif
+
+#else
+
+#ifndef ts_malloc
+#define ts_malloc  malloc
+#endif
+#ifndef ts_calloc
+#define ts_calloc  calloc
+#endif
+#ifndef ts_realloc
+#define ts_realloc realloc
+#endif
+#ifndef ts_free
+#define ts_free    free
+#endif
+
+#endif
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif // TREE_SITTER_ALLOC_H_
--- a/editor/zed/tree-sitter-jdg/src/tree_sitter/array.h
+++ b/editor/zed/tree-sitter-jdg/src/tree_sitter/array.h
@@ -0,0 +1,290 @@
+#ifndef TREE_SITTER_ARRAY_H_
+#define TREE_SITTER_ARRAY_H_
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include "./alloc.h"
+
+#include <assert.h>
+#include <stdbool.h>
+#include <stdint.h>
+#include <stdlib.h>
+#include <string.h>
+
+#ifdef _MSC_VER
+#pragma warning(disable : 4101)
+#elif defined(__GNUC__) || defined(__clang__)
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wunused-variable"
+#endif
+
+#define Array(T)       \
+  struct {             \
+    T *contents;       \
+    uint32_t size;     \
+    uint32_t capacity; \
+  }
+
+/// Initialize an array.
+#define array_init(self) \
+  ((self)->size = 0, (self)->capacity = 0, (self)->contents = NULL)
+
+/// Create an empty array.
+#define array_new() \
+  { NULL, 0, 0 }
+
+/// Get a pointer to the element at a given `index` in the array.
+#define array_get(self, _index) \
+  (assert((uint32_t)(_index) < (self)->size), &(self)->contents[_index])
+
+/// Get a pointer to the first element in the array.
+#define array_front(self) array_get(self, 0)
+
+/// Get a pointer to the last element in the array.
+#define array_back(self) array_get(self, (self)->size - 1)
+
+/// Clear the array, setting its size to zero. Note that this does not free any
+/// memory allocated for the array's contents.
+#define array_clear(self) ((self)->size = 0)
+
+/// Reserve `new_capacity` elements of space in the array. If `new_capacity` is
+/// less than the array's current capacity, this function has no effect.
+#define array_reserve(self, new_capacity) \
+  _array__reserve((Array *)(self), array_elem_size(self), new_capacity)
+
+/// Free any memory allocated for this array. Note that this does not free any
+/// memory allocated for the array's contents.
+#define array_delete(self) _array__delete((Array *)(self))
+
+/// Push a new `element` onto the end of the array.
+#define array_push(self, element)                            \
+  (_array__grow((Array *)(self), 1, array_elem_size(self)), \
+   (self)->contents[(self)->size++] = (element))
+
+/// Increase the array's size by `count` elements.
+/// New elements are zero-initialized.
+#define array_grow_by(self, count) \
+  do { \
+    if ((count) == 0) break; \
+    _array__grow((Array *)(self), count, array_elem_size(self)); \
+    memset((self)->contents + (self)->size, 0, (count) * array_elem_size(self)); \
+    (self)->size += (count); \
+  } while (0)
+
+/// Append all elements from one array to the end of another.
+#define array_push_all(self, other)                                       \
+  array_extend((self), (other)->size, (other)->contents)
+
+/// Append `count` elements to the end of the array, reading their values from the
+/// `contents` pointer.
+#define array_extend(self, count, contents)                    \
+  _array__splice(                                               \
+    (Array *)(self), array_elem_size(self), (self)->size, \
+    0, count,  contents                                        \
+  )
+
+/// Remove `old_count` elements from the array starting at the given `index`. At
+/// the same index, insert `new_count` new elements, reading their values from the
+/// `new_contents` pointer.
+#define array_splice(self, _index, old_count, new_count, new_contents)  \
+  _array__splice(                                                       \
+    (Array *)(self), array_elem_size(self), _index,                \
+    old_count, new_count, new_contents                                 \
+  )
+
+/// Insert one `element` into the array at the given `index`.
+#define array_insert(self, _index, element) \
+  _array__splice((Array *)(self), array_elem_size(self), _index, 0, 1, &(element))
+
+/// Remove one element from the array at the given `index`.
+#define array_erase(self, _index) \
+  _array__erase((Array *)(self), array_elem_size(self), _index)
+
+/// Pop the last element off the array, returning the element by value.
+#define array_pop(self) ((self)->contents[--(self)->size])
+
+/// Assign the contents of one array to another, reallocating if necessary.
+#define array_assign(self, other) \
+  _array__assign((Array *)(self), (const Array *)(other), array_elem_size(self))
+
+/// Swap one array with another
+#define array_swap(self, other) \
+  _array__swap((Array *)(self), (Array *)(other))
+
+/// Get the size of the array contents
+#define array_elem_size(self) (sizeof *(self)->contents)
+
+/// Search a sorted array for a given `needle` value, using the given `compare`
+/// callback to determine the order.
+///
+/// If an existing element is found to be equal to `needle`, then the `index`
+/// out-parameter is set to the existing value's index, and the `exists`
+/// out-parameter is set to true. Otherwise, `index` is set to an index where
+/// `needle` should be inserted in order to preserve the sorting, and `exists`
+/// is set to false.
+#define array_search_sorted_with(self, compare, needle, _index, _exists) \
+  _array__search_sorted(self, 0, compare, , needle, _index, _exists)
+
+/// Search a sorted array for a given `needle` value, using integer comparisons
+/// of a given struct field (specified with a leading dot) to determine the order.
+///
+/// See also `array_search_sorted_with`.
+#define array_search_sorted_by(self, field, needle, _index, _exists) \
+  _array__search_sorted(self, 0, _compare_int, field, needle, _index, _exists)
+
+/// Insert a given `value` into a sorted array, using the given `compare`
+/// callback to determine the order.
+#define array_insert_sorted_with(self, compare, value) \
+  do { \
+    unsigned _index, _exists; \
+    array_search_sorted_with(self, compare, &(value), &_index, &_exists); \
+    if (!_exists) array_insert(self, _index, value); \
+  } while (0)
+
+/// Insert a given `value` into a sorted array, using integer comparisons of
+/// a given struct field (specified with a leading dot) to determine the order.
+///
+/// See also `array_search_sorted_by`.
+#define array_insert_sorted_by(self, field, value) \
+  do { \
+    unsigned _index, _exists; \
+    array_search_sorted_by(self, field, (value) field, &_index, &_exists); \
+    if (!_exists) array_insert(self, _index, value); \
+  } while (0)
+
+// Private
+
+typedef Array(void) Array;
+
+/// This is not what you're looking for, see `array_delete`.
+static inline void _array__delete(Array *self) {
+  if (self->contents) {
+    ts_free(self->contents);
+    self->contents = NULL;
+    self->size = 0;
+    self->capacity = 0;
+  }
+}
+
+/// This is not what you're looking for, see `array_erase`.
+static inline void _array__erase(Array *self, size_t element_size,
+                                uint32_t index) {
+  assert(index < self->size);
+  char *contents = (char *)self->contents;
+  memmove(contents + index * element_size, contents + (index + 1) * element_size,
+          (self->size - index - 1) * element_size);
+  self->size--;
+}
+
+/// This is not what you're looking for, see `array_reserve`.
+static inline void _array__reserve(Array *self, size_t element_size, uint32_t new_capacity) {
+  if (new_capacity > self->capacity) {
+    if (self->contents) {
+      self->contents = ts_realloc(self->contents, new_capacity * element_size);
+    } else {
+      self->contents = ts_malloc(new_capacity * element_size);
+    }
+    self->capacity = new_capacity;
+  }
+}
+
+/// This is not what you're looking for, see `array_assign`.
+static inline void _array__assign(Array *self, const Array *other, size_t element_size) {
+  _array__reserve(self, element_size, other->size);
+  self->size = other->size;
+  memcpy(self->contents, other->contents, self->size * element_size);
+}
+
+/// This is not what you're looking for, see `array_swap`.
+static inline void _array__swap(Array *self, Array *other) {
+  Array swap = *other;
+  *other = *self;
+  *self = swap;
+}
+
+/// This is not what you're looking for, see `array_push` or `array_grow_by`.
+static inline void _array__grow(Array *self, uint32_t count, size_t element_size) {
+  uint32_t new_size = self->size + count;
+  if (new_size > self->capacity) {
+    uint32_t new_capacity = self->capacity * 2;
+    if (new_capacity < 8) new_capacity = 8;
+    if (new_capacity < new_size) new_capacity = new_size;
+    _array__reserve(self, element_size, new_capacity);
+  }
+}
+
+/// This is not what you're looking for, see `array_splice`.
+static inline void _array__splice(Array *self, size_t element_size,
+                                 uint32_t index, uint32_t old_count,
+                                 uint32_t new_count, const void *elements) {
+  uint32_t new_size = self->size + new_count - old_count;
+  uint32_t old_end = index + old_count;
+  uint32_t new_end = index + new_count;
+  assert(old_end <= self->size);
+
+  _array__reserve(self, element_size, new_size);
+
+  char *contents = (char *)self->contents;
+  if (self->size > old_end) {
+    memmove(
+      contents + new_end * element_size,
+      contents + old_end * element_size,
+      (self->size - old_end) * element_size
+    );
+  }
+  if (new_count > 0) {
+    if (elements) {
+      memcpy(
+        (contents + index * element_size),
+        elements,
+        new_count * element_size
+      );
+    } else {
+      memset(
+        (contents + index * element_size),
+        0,
+        new_count * element_size
+      );
+    }
+  }
+  self->size += new_count - old_count;
+}
+
+/// A binary search routine, based on Rust's `std::slice::binary_search_by`.
+/// This is not what you're looking for, see `array_search_sorted_with` or `array_search_sorted_by`.
+#define _array__search_sorted(self, start, compare, suffix, needle, _index, _exists) \
+  do { \
+    *(_index) = start; \
+    *(_exists) = false; \
+    uint32_t size = (self)->size - *(_index); \
+    if (size == 0) break; \
+    int comparison; \
+    while (size > 1) { \
+      uint32_t half_size = size / 2; \
+      uint32_t mid_index = *(_index) + half_size; \
+      comparison = compare(&((self)->contents[mid_index] suffix), (needle)); \
+      if (comparison <= 0) *(_index) = mid_index; \
+      size -= half_size; \
+    } \
+    comparison = compare(&((self)->contents[*(_index)] suffix), (needle)); \
+    if (comparison == 0) *(_exists) = true; \
+    else if (comparison < 0) *(_index) += 1; \
+  } while (0)
+
+/// Helper macro for the `_sorted_by` routines below. This takes the left (existing)
+/// parameter by reference in order to work with the generic sorting function above.
+#define _compare_int(a, b) ((int)*(a) - (int)(b))
+
+#ifdef _MSC_VER
+#pragma warning(default : 4101)
+#elif defined(__GNUC__) || defined(__clang__)
+#pragma GCC diagnostic pop
+#endif
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif  // TREE_SITTER_ARRAY_H_
--- a/editor/zed/tree-sitter-jdg/src/tree_sitter/parser.h
+++ b/editor/zed/tree-sitter-jdg/src/tree_sitter/parser.h
@@ -0,0 +1,265 @@
+#ifndef TREE_SITTER_PARSER_H_
+#define TREE_SITTER_PARSER_H_
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include <stdbool.h>
+#include <stdint.h>
+#include <stdlib.h>
+
+#define ts_builtin_sym_error ((TSSymbol)-1)
+#define ts_builtin_sym_end 0
+#define TREE_SITTER_SERIALIZATION_BUFFER_SIZE 1024
+
+#ifndef TREE_SITTER_API_H_
+typedef uint16_t TSStateId;
+typedef uint16_t TSSymbol;
+typedef uint16_t TSFieldId;
+typedef struct TSLanguage TSLanguage;
+#endif
+
+typedef struct {
+  TSFieldId field_id;
+  uint8_t child_index;
+  bool inherited;
+} TSFieldMapEntry;
+
+typedef struct {
+  uint16_t index;
+  uint16_t length;
+} TSFieldMapSlice;
+
+typedef struct {
+  bool visible;
+  bool named;
+  bool supertype;
+} TSSymbolMetadata;
+
+typedef struct TSLexer TSLexer;
+
+struct TSLexer {
+  int32_t lookahead;
+  TSSymbol result_symbol;
+  void (*advance)(TSLexer *, bool);
+  void (*mark_end)(TSLexer *);
+  uint32_t (*get_column)(TSLexer *);
+  bool (*is_at_included_range_start)(const TSLexer *);
+  bool (*eof)(const TSLexer *);
+};
+
+typedef enum {
+  TSParseActionTypeShift,
+  TSParseActionTypeReduce,
+  TSParseActionTypeAccept,
+  TSParseActionTypeRecover,
+} TSParseActionType;
+
+typedef union {
+  struct {
+    uint8_t type;
+    TSStateId state;
+    bool extra;
+    bool repetition;
+  } shift;
+  struct {
+    uint8_t type;
+    uint8_t child_count;
+    TSSymbol symbol;
+    int16_t dynamic_precedence;
+    uint16_t production_id;
+  } reduce;
+  uint8_t type;
+} TSParseAction;
+
+typedef struct {
+  uint16_t lex_state;
+  uint16_t external_lex_state;
+} TSLexMode;
+
+typedef union {
+  TSParseAction action;
+  struct {
+    uint8_t count;
+    bool reusable;
+  } entry;
+} TSParseActionEntry;
+
+typedef struct {
+  int32_t start;
+  int32_t end;
+} TSCharacterRange;
+
+struct TSLanguage {
+  uint32_t version;
+  uint32_t symbol_count;
+  uint32_t alias_count;
+  uint32_t token_count;
+  uint32_t external_token_count;
+  uint32_t state_count;
+  uint32_t large_state_count;
+  uint32_t production_id_count;
+  uint32_t field_count;
+  uint16_t max_alias_sequence_length;
+  const uint16_t *parse_table;
+  const uint16_t *small_parse_table;
+  const uint32_t *small_parse_table_map;
+  const TSParseActionEntry *parse_actions;
+  const char * const *symbol_names;
+  const char * const *field_names;
+  const TSFieldMapSlice *field_map_slices;
+  const TSFieldMapEntry *field_map_entries;
+  const TSSymbolMetadata *symbol_metadata;
+  const TSSymbol *public_symbol_map;
+  const uint16_t *alias_map;
+  const TSSymbol *alias_sequences;
+  const TSLexMode *lex_modes;
+  bool (*lex_fn)(TSLexer *, TSStateId);
+  bool (*keyword_lex_fn)(TSLexer *, TSStateId);
+  TSSymbol keyword_capture_token;
+  struct {
+    const bool *states;
+    const TSSymbol *symbol_map;
+    void *(*create)(void);
+    void (*destroy)(void *);
+    bool (*scan)(void *, TSLexer *, const bool *symbol_whitelist);
+    unsigned (*serialize)(void *, char *);
+    void (*deserialize)(void *, const char *, unsigned);
+  } external_scanner;
+  const TSStateId *primary_state_ids;
+};
+
+static inline bool set_contains(TSCharacterRange *ranges, uint32_t len, int32_t lookahead) {
+  uint32_t index = 0;
+  uint32_t size = len - index;
+  while (size > 1) {
+    uint32_t half_size = size / 2;
+    uint32_t mid_index = index + half_size;
+    TSCharacterRange *range = &ranges[mid_index];
+    if (lookahead >= range->start && lookahead <= range->end) {
+      return true;
+    } else if (lookahead > range->end) {
+      index = mid_index;
+    }
+    size -= half_size;
+  }
+  TSCharacterRange *range = &ranges[index];
+  return (lookahead >= range->start && lookahead <= range->end);
+}
+
+/*
+ *  Lexer Macros
+ */
+
+#ifdef _MSC_VER
+#define UNUSED __pragma(warning(suppress : 4101))
+#else
+#define UNUSED __attribute__((unused))
+#endif
+
+#define START_LEXER()           \
+  bool result = false;          \
+  bool skip = false;            \
+  UNUSED                        \
+  bool eof = false;             \
+  int32_t lookahead;            \
+  goto start;                   \
+  next_state:                   \
+  lexer->advance(lexer, skip);  \
+  start:                        \
+  skip = false;                 \
+  lookahead = lexer->lookahead;
+
+#define ADVANCE(state_value) \
+  {                          \
+    state = state_value;     \
+    goto next_state;         \
+  }
+
+#define ADVANCE_MAP(...)                                              \
+  {                                                                   \
+    static const uint16_t map[] = { __VA_ARGS__ };                    \
+    for (uint32_t i = 0; i < sizeof(map) / sizeof(map[0]); i += 2) {  \
+      if (map[i] == lookahead) {                                      \
+        state = map[i + 1];                                           \
+        goto next_state;                                              \
+      }                                                               \
+    }                                                                 \
+  }
+
+#define SKIP(state_value) \
+  {                       \
+    skip = true;          \
+    state = state_value;  \
+    goto next_state;      \
+  }
+
+#define ACCEPT_TOKEN(symbol_value)     \
+  result = true;                       \
+  lexer->result_symbol = symbol_value; \
+  lexer->mark_end(lexer);
+
+#define END_STATE() return result;
+
+/*
+ *  Parse Table Macros
+ */
+
+#define SMALL_STATE(id) ((id) - LARGE_STATE_COUNT)
+
+#define STATE(id) id
+
+#define ACTIONS(id) id
+
+#define SHIFT(state_value)            \
+  {{                                  \
+    .shift = {                        \
+      .type = TSParseActionTypeShift, \
+      .state = (state_value)          \
+    }                                 \
+  }}
+
+#define SHIFT_REPEAT(state_value)     \
+  {{                                  \
+    .shift = {                        \
+      .type = TSParseActionTypeShift, \
+      .state = (state_value),         \
+      .repetition = true              \
+    }                                 \
+  }}
+
+#define SHIFT_EXTRA()                 \
+  {{                                  \
+    .shift = {                        \
+      .type = TSParseActionTypeShift, \
+      .extra = true                   \
+    }                                 \
+  }}
+
+#define REDUCE(symbol_name, children, precedence, prod_id) \
+  {{                                                       \
+    .reduce = {                                            \
+      .type = TSParseActionTypeReduce,                     \
+      .symbol = symbol_name,                               \
+      .child_count = children,                             \
+      .dynamic_precedence = precedence,                    \
+      .production_id = prod_id                             \
+    },                                                     \
+  }}
+
+#define RECOVER()                    \
+  {{                                 \
+    .type = TSParseActionTypeRecover \
+  }}
+
+#define ACCEPT_INPUT()              \
+  {{                                \
+    .type = TSParseActionTypeAccept \
+  }}
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif  // TREE_SITTER_PARSER_H_
--- a/example/c-sum-v2/solution.c
+++ b/example/c-sum-v2/solution.c
@@ -0,0 +1,19 @@
+#include <stdio.h>
+#include <stdlib.h>
+
+int main(void) {
+    long n;
+    if (scanf("%ld", &n) != 1) {
+        return 1;
+    }
+    long long sum = 0;
+    for (long i = 0; i < n; i++) {
+        long long x;
+        if (scanf("%lld", &x) != 1) {
+            return 1;
+        }
+        sum += x;
+    }
+    printf("%lld\n", sum);
+    return 0;
+}
--- a/example/c-sum-v2/sum.jdg
+++ b/example/c-sum-v2/sum.jdg
@@ -0,0 +1,82 @@
+// Smoke test for the structured build DSL. Same task as example/c-sum
+// (sum of N integers) but described declaratively — no shell strings,
+// no per-platform `if /I "%CC%"=="msvc"` branching. The compiler is
+// selected via the JUDGE_CC env variable set by the CI matrix.
+
+toolchains {
+    gcc   { platforms = "linux"             }
+    clang { platforms = "linux" "windows"   }
+    msvc  { platforms = "windows"           }
+}
+
+build_defaults {
+    language = "c"
+    standard = "c11"
+    sources  = "solution.c"
+    output   = "solution"
+    warnings = strict
+}
+
+build "release" {
+    profile = release
+}
+
+build "debug" {
+    profile = debug
+}
+
+build "sanitized" {
+    profile   = sanitized
+    sanitize  = "address" "undefined"
+    platforms = "linux"
+    compilers = "gcc" "clang"
+}
+
+build "debug-valgrind" {
+    profile   = debug
+    wrapper   = "valgrind --error-exitcode=99 --leak-check=full -q"
+    platforms = "linux"
+    compilers = "gcc"
+}
+
+timeout 5s
+normalize_crlf = true
+trim_trailing_ws = true
+
+group("basic") {
+    weight = 0.5
+
+    test("one number") {
+        stdin  = "1\n42\n"
+        stdout = "42\n"
+    }
+
+    test("three numbers") {
+        stdin  = "3\n1 2 3\n"
+        stdout = "6\n"
+    }
+
+    test("negatives") {
+        stdin  = "4\n-1 -2 3 5\n"
+        stdout = "5\n"
+    }
+
+    test("zero count") {
+        stdin  = "0\n"
+        stdout = "0\n"
+    }
+}
+
+group("edge") {
+    weight = 0.5
+
+    test("large sum fits in int64") {
+        stdin  = "3\n2000000000 2000000000 2000000000\n"
+        stdout = "6000000000\n"
+    }
+
+    test("multiline input") {
+        stdin  = "5\n10\n20\n30\n40\n50\n"
+        stdout = "150\n"
+    }
+}
--- a/example/showcase/01-minimal.jdg
+++ b/example/showcase/01-minimal.jdg
@@ -0,0 +1,23 @@
+// Минимальный рабочий suite на structured DSL.
+// Декларируем один toolchain, один build, одну группу с одним тестом.
+// С этого можно копипастить для новых задач.
+
+toolchains {
+    gcc { platforms = "linux" }
+}
+
+build "release" {
+    language = "c"
+    standard = "c11"
+    sources  = "solution.c"
+    output   = "solution"
+    profile  = release
+}
+
+group("basic") {
+    weight = 1.0
+
+    test("hello") {
+        stdout = "Hello, World!\n"
+    }
+}
--- a/example/showcase/02-structured-build.jdg
+++ b/example/showcase/02-structured-build.jdg
@@ -0,0 +1,72 @@
+// Полный showcase возможностей build-блока.
+// Показывает: toolchains c платформами / переопределением binary / явным class,
+// build_defaults как общая основа, три профиля (release/debug/sanitized),
+// фильтры platforms и compilers, OS-оверрайды, wrapper, sanitize, define,
+// includes, link, extra, а также build-level `wrapper` для valgrind.
+
+toolchains {
+    gcc   { platforms = "linux" }
+    clang {
+        platforms = "linux" "windows"
+        binary    = "clang-17"   // конкретная версия в PATH
+        class     = gnu           // явный class если имя нестандартное
+    }
+    msvc  { platforms = "windows" }
+    nvcc  {
+        platforms = "linux"
+        class     = gnu           // nvcc принимает gnu-like флаги
+    }
+}
+
+build_defaults {
+    language = "c"
+    standard = "c11"
+    sources  = "src/*.c"          // glob, раскрывается относительно work dir
+    output   = "solution"         // на Windows автоматически станет solution.exe
+    includes = "include" "vendor/include"
+    link     = "m" "pthread"
+    warnings = strict             // → -Wall -Wextra / /W4
+    define("BUILD_ID") = "42"
+}
+
+build "release" {
+    profile = release             // → -O2 / /O2
+
+    // OS-специфичные правки поверх defaults. Скаляры переопределяются,
+    // списки (extra, sanitize, sources, includes) аккумулируются.
+    linux   { extra = "-fPIC" }
+    windows { extra = "/bigobj" }
+    darwin  { extra = "-mmacosx-version-min=11" }
+}
+
+build "debug" {
+    profile  = debug              // → -O0 -g / /Od /Zi
+    warnings = pedantic           // → добавляет -Wpedantic / /permissive-
+    define("DEBUG") = "1"
+}
+
+build "sanitized" {
+    profile   = sanitized
+    sanitize  = "address" "undefined" "leak"
+    platforms = "linux"           // UBSan не работает на MSVC
+    compilers = "gcc" "clang"     // и только gnu-like
+    extra     = "-fno-sanitize-recover=all"
+}
+
+build "debug-valgrind" {
+    profile   = debug
+    wrapper   = "valgrind --error-exitcode=99 --leak-check=full -q"
+    platforms = "linux"
+    compilers = "gcc"
+}
+
+timeout 10s
+memory_limit = 256M
+
+group("basic") {
+    weight = 1.0
+
+    test("smoke") {
+        stdout = "ok\n"
+    }
+}
--- a/example/showcase/03-matchers.jdg
+++ b/example/showcase/03-matchers.jdg
@@ -0,0 +1,69 @@
+// Все способы сматчить stdout/stderr: exact, contains, regex, anyOrder,
+// numeric-eps. Плюс случай, когда сравнение вывода вообще не нужно
+// (достаточно exitCode).
+
+toolchains {
+    gcc { platforms = "linux" }
+}
+
+build "release" {
+    language = "c"
+    standard = "c11"
+    sources  = "solution.c"
+    output   = "solution"
+    profile  = release
+}
+
+group("matchers") {
+    weight = 1.0
+
+    // Точное совпадение (по умолчанию). Пробелы, переводы строк и
+    // регистр важны. Нормализация CRLF / trim_ws настраивается на
+    // уровне файла.
+    test("exact") {
+        stdin  = "42\n"
+        stdout = "42\n"
+    }
+
+    // Подстрока в любом месте вывода.
+    test("contains") {
+        stdin  = "log me\n"
+        stdout contains "log me"
+    }
+
+    // Регулярное выражение (Go regexp syntax).
+    test("regex") {
+        stdin  = "7\n"
+        stdout matches "^result:\\s+[0-9]+\\s*$"
+    }
+
+    // Лайны могут прийти в любом порядке — полезно для stress-тестов
+    // с параллельным выводом.
+    test("any-order") {
+        stdin = "start\n"
+        stdout anyOrder {
+            "alpha"
+            "beta"
+            "gamma"
+        }
+    }
+
+    // Числовое сравнение с epsilon — для задач с плавающей точкой.
+    // Синтаксис: ~ <float> of "<ожидаемое значение как строка>"
+    test("numeric") {
+        stdin  = "3.14159\n"
+        stdout ~ 0.0001 of "3.14159\n"
+    }
+
+    // Проверяем только stderr; stdout не трогаем вообще (NoMatcher).
+    test("stderr-only") {
+        args   = "--version"
+        stderr contains "version 1."
+    }
+
+    // Никакого матчинга вывода — нас интересует только exit code.
+    test("exit-code-only") {
+        args     = "--help"
+        exitCode = 0
+    }
+}
--- a/example/showcase/04-limits-scoring-env.jdg
+++ b/example/showcase/04-limits-scoring-env.jdg
@@ -0,0 +1,77 @@
+// Тонкая настройка: timeout и memory_limit на разных уровнях (file → group
+// → test), режимы скоринга (partial / all_or_none), переменные окружения,
+// проверка exit code, передача аргументов.
+
+toolchains {
+    gcc { platforms = "linux" }
+}
+
+build "release" {
+    language = "c"
+    standard = "c11"
+    sources  = "solution.c"
+    output   = "solution"
+    profile  = release
+}
+
+// Глобальные дефолты. Применяются к каждому тесту, если тот не
+// переопределит у себя или на уровне группы.
+timeout 5s
+memory_limit = 256M
+normalize_crlf   = true   // стрип \r перед сравнением — полезно на Windows
+trim_trailing_ws = true   // trim trailing space/tab на каждой строке
+
+group("strict-subset") {
+    weight  = 0.4
+    scoring = all_or_none     // группа даёт weight или 0 — без частичного
+
+    // Эти дефолты применятся к каждому тесту группы.
+    timeout = 2s
+    memory_limit = 64M
+    env("LC_ALL") = "C"
+    env("LANG")   = "C"
+
+    test("fast") {
+        args   = "--mode" "fast"
+        stdout = "ok\n"
+    }
+
+    test("strict-timeout") {
+        timeout = 500ms          // переопределяет group timeout
+        stdout  = "ok\n"
+    }
+
+    test("high-memory") {
+        memory_limit = 128M      // переопределяет group memory_limit
+        stdout       = "ok\n"
+    }
+
+    test("own-env") {
+        env("FOO") = "bar"       // добавляется к group env
+        stdout     = "ok\n"
+    }
+}
+
+group("partial-credit") {
+    weight  = 0.6
+    scoring = partial             // default: weight * passed/total
+
+    test("case-1") {
+        stdin    = "1\n"
+        stdout   = "1\n"
+        exitCode = 0
+    }
+
+    test("case-2") {
+        stdin    = "2\n"
+        stdout   = "4\n"
+        exitCode = 0
+    }
+
+    test("case-error") {
+        // Программа ДОЛЖНА выйти с кодом 1 на плохом вводе.
+        stdin    = "abc\n"
+        exitCode = 1
+        stderr contains "invalid input"
+    }
+}
--- a/example/showcase/05-pattern-expansion.jdg
+++ b/example/showcase/05-pattern-expansion.jdg
@@ -0,0 +1,80 @@
+// Автогенерация тестов из файлов/папок. Покрывает четыре типовых сценария:
+//
+//   1. Stdio-режим с парой globs     — классика: stdin из .in, stdout = .ans
+//   2. Shared expected               — N входов, один общий ожидаемый файл
+//   3. File-mode по args             — программа читает файл по аргументу
+//   4. Dir-mode с input+output файлами — программа читает И пишет файлы
+//
+// Плейсхолдеры в `args` pattern'а:
+//   {input_path}  — путь к input-файлу текущего теста (для glob — разный,
+//                   для литерала — один и тот же во всех тестах)
+//   {output_path} — аналогично для output
+//   {name}        — wildcard-часть имени (glob) или basename директории (dir)
+//   {dir}         — dir-mode: полный путь к директории теста
+//
+// Правило режима: наличие `{input_path}` в args → input кладётся как InFile,
+// а не в stdin. Наличие `{output_path}` → expected сравнивается как OutFile,
+// а не stdout.
+
+toolchains {
+    gcc { platforms = "linux" }
+}
+
+build "release" {
+    language = "c"
+    standard = "c11"
+    sources  = "solution.c"
+    output   = "solution"
+    profile  = release
+}
+
+// 1) Классика: каждая пара (*.in, *.ans) → один тест.
+//    Stdin ← .in, Stdout сравнивается с .ans.
+group("stdio-pair") {
+    weight = 0.25
+
+    pattern {
+        input  = "tests/*.in"
+        output = "tests/*.ans"
+    }
+}
+
+// 2) Один общий expected для всех входов.
+//    Удобно для задач где «для любого корректного входа ответ одинаковый»
+//    (edge-case fuzzing) или наоборот «все некорректные входы → error»
+//    (смотрим на один и тот же файл с сообщением об ошибке).
+group("shared-output") {
+    weight = 0.25
+
+    pattern {
+        input  = "edge/*.in"
+        output = "edge/expected.ans"   // литерал без *
+    }
+}
+
+// 3) Программа читает файл по аргументу командной строки, а не из stdin.
+//    Expander создаёт InFiles[01.in]=..., запускает `./solution 01.in`,
+//    сравнивает stdout программы с содержимым соответствующего .ans.
+group("file-input") {
+    weight = 0.25
+
+    pattern {
+        input  = "file-tests/*.in"
+        output = "file-tests/*.ans"
+        args   = "{input_path}"
+    }
+}
+
+// 4) Программа и читает, и пишет через файлы — никакого stdin/stdout.
+//    Expander кладёт input в InFiles, expected в OutFiles, запускает
+//    `./solution in.txt out.txt` и сравнивает созданный out.txt с expected.
+group("file-io") {
+    weight = 0.25
+
+    pattern {
+        dirs   = "cases/*"          // каждая подпапка — отдельный тест
+        input  = "in.txt"
+        output = "out.txt"
+        args   = "{input_path}" "{output_path}"
+    }
+}
--- a/example/showcase/06-files.jdg
+++ b/example/showcase/06-files.jdg
@@ -0,0 +1,59 @@
+// Тесты, которые общаются через файлы, а не через stdin/stdout.
+// file("name") = "content"    — файл кладётся в рабочую директорию теста
+//                                перед запуском программы.
+// outFile("name") = "content" — после запуска содержимое файла сравнивается
+//                                с ожидаемым (точное совпадение с учётом
+//                                normalize_crlf / trim_trailing_ws).
+
+toolchains {
+    gcc { platforms = "linux" }
+}
+
+build "release" {
+    language = "c"
+    standard = "c11"
+    sources  = "solution.c"
+    output   = "solution"
+    profile  = release
+}
+
+normalize_crlf   = true
+trim_trailing_ws = true
+
+group("io") {
+    weight = 1.0
+
+    // Программа читает input.txt, пишет output.txt.
+    test("copy") {
+        args = "input.txt" "output.txt"
+        file("input.txt")   = "hello world\n"
+        outFile("output.txt") = "hello world\n"
+    }
+
+    // Несколько входных и выходных файлов сразу — удобно для задач
+    // вроде «посчитать сумму по каждой строке и записать в отдельный
+    // файл».
+    test("multi-file") {
+        args = "a.txt" "b.txt" "result.txt"
+        file("a.txt") = "1 2 3\n"
+        file("b.txt") = "10 20 30\n"
+        outFile("result.txt") = "11 22 33\n"
+    }
+
+    // Вложенные пути тоже работают — создаются автоматически.
+    test("nested") {
+        args = "data/in.txt" "data/out.txt"
+        file("data/in.txt")    = "payload\n"
+        outFile("data/out.txt") = "PAYLOAD\n"
+    }
+
+    // Можно комбинировать с stdin/stdout и переменными окружения.
+    test("combined") {
+        stdin         = "42\n"
+        args          = "config.ini"
+        env("MODE")   = "strict"
+        file("config.ini") = "threshold=10\n"
+        stdout        = "ok\n"
+        outFile("log.txt") = "processed 42\n"
+    }
+}
--- a/example/showcase/07-include.jdg
+++ b/example/showcase/07-include.jdg
@@ -0,0 +1,41 @@
+// Наследование общей конфигурации через `include`. Подключает соседний
+// common.jdg (тулчейны + build_defaults + глобальные настройки), а затем
+// задаёт только специфичные для этой задачи build-варианты и тесты.
+//
+// Пути в include относительны самому этому файлу (не CWD, не корню репо).
+// Абсолютные пути тоже работают. Циклические включения детектятся парсером.
+//
+// Локальные директивы поверх include могут переопределять:
+//   - скаляры в build_defaults (через последующий build_defaults-блок)
+//   - timeout / memory_limit / binary / sources / normalize_crlf / trim_trailing_ws
+// Но не могут переопределять:
+//   - toolchains с тем же именем (ошибка «duplicate toolchain»)
+//   - build с тем же именем
+//   - group с тем же именем
+
+include "common.jdg"
+
+build "release" {
+    profile = release
+}
+
+build "debug" {
+    profile  = debug
+    warnings = pedantic
+}
+
+build "sanitized" {
+    profile   = sanitized
+    sanitize  = "address" "undefined"
+    platforms = "linux"
+    compilers = "gcc" "clang"
+}
+
+group("basic") {
+    weight = 1.0
+
+    test("smoke") {
+        stdin  = "1\n42\n"
+        stdout = "42\n"
+    }
+}
--- a/example/showcase/common.jdg
+++ b/example/showcase/common.jdg
@@ -0,0 +1,22 @@
+// Общая инфраструктура для всех задач курса: тулчейны и базовые настройки
+// билда. Подключается через `include "common.jdg"` из конкретных .jdg
+// файлов задач. Локальные build-блоки наследуют эти дефолты через
+// встроенный merge в build_defaults.
+
+toolchains {
+    gcc   { platforms = "linux" }
+    clang { platforms = "linux" "windows" }
+    msvc  { platforms = "windows" }
+}
+
+build_defaults {
+    language = "c"
+    standard = "c11"
+    sources  = "solution.c"
+    output   = "solution"
+    warnings = strict
+}
+
+timeout 5s
+normalize_crlf   = true
+trim_trailing_ws = true
--- a/reporter/reporter.go
+++ b/reporter/reporter.go
@@ -4,6 +4,7 @@ import (
 	"encoding/json"
 	"fmt"
 	"io"
+	"math"
 	"os"
 	"path/filepath"
 	"sort"
@@ -13,11 +14,47 @@ import (
 )

 func Text(w io.Writer, result *runner.SuiteResult) {
-	if result.BuildLog != "" {
-		fmt.Fprintf(w, "=== BUILD LOG ===\n%s\n", result.BuildLog)
+	if len(result.Builds) == 0 {
+		fmt.Fprintln(w, "(no builds executed)")
+		return
 	}

-	for _, gr := range result.Groups {
+	multi := len(result.Builds) > 1
+	if multi {
+		fmt.Fprintf(w, "=== %d builds ===\n", len(result.Builds))
+		for _, b := range result.Builds {
+			marker := ""
+			if b.Skipped {
+				marker = " (skipped)"
+			}
+			fmt.Fprintf(w, "  • %s%s  score=%.4f\n", b.Name, marker, b.TotalScore)
+		}
+		fmt.Fprintln(w)
+	}
+
+	for _, b := range result.Builds {
+		if multi {
+			fmt.Fprintf(w, "======== build %q ========\n", b.Name)
+		}
+		writeBuildText(w, b, multi)
+	}
+
+	if multi {
+		fmt.Fprintf(w, "\n══ AGGREGATE SCORE (min across builds): %.4f / 1.0000 ══\n", result.TotalScore)
+	}
+}
+
+func writeBuildText(w io.Writer, b *runner.BuildRun, multi bool) {
+	if b.Skipped {
+		fmt.Fprintf(w, "skipped: %s\n", b.SkipReason)
+		return
+	}
+
+	if b.BuildLog != "" {
+		fmt.Fprintf(w, "=== BUILD LOG ===\n%s\n", b.BuildLog)
+	}
+
+	for _, gr := range b.Groups {
 		passed := gr.Passed
 		total := gr.Total
 		pct := 0.0
@@ -53,13 +90,19 @@ func Text(w io.Writer, result *runner.SuiteResult) {
 		fmt.Fprintf(w, "└─\n")
 	}

-	fmt.Fprintf(w, "\n══ TOTAL SCORE: %.4f / 1.0000 ══\n", result.TotalScore)
+	if !multi {
+		fmt.Fprintf(w, "\n══ TOTAL SCORE: %.4f / 1.0000 ══\n", b.TotalScore)
+	}
 }

 func JSON(w io.Writer, result *runner.SuiteResult) error {
 	enc := json.NewEncoder(w)
 	enc.SetIndent("", "  ")
-	return enc.Encode(jsonResult(result))
+
+	if len(result.Builds) <= 1 {
+		return enc.Encode(flatResult(result))
+	}
+	return enc.Encode(nestedResult(result))
 }

 type jsonSuiteResult struct {
@@ -68,6 +111,21 @@ type jsonSuiteResult struct {
 	Groups     []jsonGroupResult `json:"groups"`
 }

+type jsonMultiSuiteResult struct {
+	TotalScore float64                   `json:"total_score"`
+	Builds     map[string]jsonSuiteBuild `json:"builds"`
+}
+
+type jsonSuiteBuild struct {
+	Name       string            `json:"name"`
+	Toolchain  string            `json:"toolchain,omitempty"`
+	Skipped    bool              `json:"skipped,omitempty"`
+	SkipReason string            `json:"skip_reason,omitempty"`
+	TotalScore float64           `json:"total_score"`
+	BuildLog   string            `json:"build_log,omitempty"`
+	Groups     []jsonGroupResult `json:"groups"`
+}
+
 type jsonGroupResult struct {
 	Name   string           `json:"name"`
 	Weight float64          `json:"weight"`
@@ -86,6 +144,62 @@ type jsonTestResult struct {
 	Failures      []string `json:"failures,omitempty"`
 }

+func flatResult(r *runner.SuiteResult) jsonSuiteResult {
+	res := jsonSuiteResult{TotalScore: r.TotalScore}
+	if len(r.Builds) == 0 {
+		return res
+	}
+	b := r.Builds[0]
+	res.BuildLog = b.BuildLog
+	for _, gr := range b.Groups {
+		res.Groups = append(res.Groups, groupJSON(gr))
+	}
+	return res
+}
+
+func nestedResult(r *runner.SuiteResult) jsonMultiSuiteResult {
+	res := jsonMultiSuiteResult{
+		TotalScore: r.TotalScore,
+		Builds:     map[string]jsonSuiteBuild{},
+	}
+	for _, b := range r.Builds {
+		entry := jsonSuiteBuild{
+			Name:       b.Name,
+			Toolchain:  b.Toolchain,
+			Skipped:    b.Skipped,
+			SkipReason: b.SkipReason,
+			TotalScore: b.TotalScore,
+			BuildLog:   b.BuildLog,
+		}
+		for _, gr := range b.Groups {
+			entry.Groups = append(entry.Groups, groupJSON(gr))
+		}
+		res.Builds[b.Name] = entry
+	}
+	return res
+}
+
+func groupJSON(gr *runner.GroupResult) jsonGroupResult {
+	jgr := jsonGroupResult{
+		Name:   gr.Name,
+		Weight: gr.Weight,
+		Score:  gr.Score,
+		Passed: gr.Passed,
+		Total:  gr.Total,
+	}
+	for _, tr := range gr.Tests {
+		jgr.Tests = append(jgr.Tests, jsonTestResult{
+			Name:          tr.Name,
+			Status:        tr.Status.String(),
+			ElapsedMs:     tr.Elapsed.Milliseconds(),
+			PeakMemoryKB:  tr.PeakMemory / 1024,
+			MemoryLimitKB: tr.MemoryLimit / 1024,
+			Failures:      tr.Failures,
+		})
+	}
+	return jgr
+}
+
 func Aggregate(w io.Writer, dir string) error {
 	files, err := filepath.Glob(filepath.Join(dir, "*", "*.json"))
 	if err != nil {
@@ -105,24 +219,24 @@ func Aggregate(w io.Writer, dir string) error {
 	}

 	var entries []entry
-	allPassed := true
+	minScore := math.Inf(1)

 	for _, f := range files {
 		data, err := os.ReadFile(f)
 		if err != nil {
 			return fmt.Errorf("read %s: %w", f, err)
 		}
-		var report jsonSuiteResult
-		if err := json.Unmarshal(data, &report); err != nil {
+		score, err := extractTotalScore(data)
+		if err != nil {
 			return fmt.Errorf("parse %s: %w", f, err)
 		}

 		cfg := filepath.Base(filepath.Dir(f))
 		cfg = strings.TrimPrefix(cfg, "report_")

-		entries = append(entries, entry{Config: cfg, Score: report.TotalScore})
-		if report.TotalScore < 0.9999 {
-			allPassed = false
+		entries = append(entries, entry{Config: cfg, Score: score})
+		if score < minScore {
+			minScore = score
 		}
 	}

@@ -133,13 +247,24 @@ func Aggregate(w io.Writer, dir string) error {
 	for _, e := range entries {
 		fmt.Fprintf(w, "| %s | %.4f |\n", e.Config, e.Score)
 	}
+	fmt.Fprintf(w, "| **Overall (min)** | **%.4f** |\n", minScore)

-	if !allPassed {
-		return fmt.Errorf("one or more configurations scored below 1.0")
+	if minScore < 0.9999 {
+		return fmt.Errorf("minimum score across configurations is %.4f (below 1.0)", minScore)
 	}
 	return nil
 }

+func extractTotalScore(data []byte) (float64, error) {
+	var header struct {
+		TotalScore float64 `json:"total_score"`
+	}
+	if err := json.Unmarshal(data, &header); err != nil {
+		return 0, err
+	}
+	return header.TotalScore, nil
+}
+
 func humanBytes(n int64) string {
 	const (
 		KiB = 1024
@@ -157,31 +282,3 @@ func humanBytes(n int64) string {
 		return fmt.Sprintf("%dB", n)
 	}
 }
-
-func jsonResult(r *runner.SuiteResult) jsonSuiteResult {
-	res := jsonSuiteResult{
-		TotalScore: r.TotalScore,
-		BuildLog:   r.BuildLog,
-	}
-	for _, gr := range r.Groups {
-		jgr := jsonGroupResult{
-			Name:   gr.Name,
-			Weight: gr.Weight,
-			Score:  gr.Score,
-			Passed: gr.Passed,
-			Total:  gr.Total,
-		}
-		for _, tr := range gr.Tests {
-			jgr.Tests = append(jgr.Tests, jsonTestResult{
-				Name:          tr.Name,
-				Status:        tr.Status.String(),
-				ElapsedMs:     tr.Elapsed.Milliseconds(),
-				PeakMemoryKB:  tr.PeakMemory / 1024,
-				MemoryLimitKB: tr.MemoryLimit / 1024,
-				Failures:      tr.Failures,
-			})
-		}
-		res.Groups = append(res.Groups, jgr)
-	}
-	return res
-}
--- a/runner/compiler.go
+++ b/runner/compiler.go
@@ -0,0 +1,198 @@
+package runner
+
+import (
+	"fmt"
+	"runtime"
+	"slices"
+	"sort"
+	"strings"
+
+	"github.com/Mond1c/judge/dsl"
+)
+
+type CompilerClass int
+
+const (
+	CompilerUnknown CompilerClass = iota
+	CompilerGNU
+	CompilerMSVC
+)
+
+type Toolchain struct {
+	Class  CompilerClass
+	Binary string
+	Name   string
+}
+
+func ResolveToolchain(ccEnv string) Toolchain {
+	if ccEnv == "" {
+		ccEnv = defaultCC()
+	}
+	lower := strings.ToLower(ccEnv)
+	switch lower {
+	case "gcc", "g++":
+		return Toolchain{Class: CompilerGNU, Binary: ccEnv, Name: "gcc"}
+	case "clang", "clang++":
+		return Toolchain{Class: CompilerGNU, Binary: ccEnv, Name: "clang"}
+	case "cl", "cl.exe", "msvc":
+		return Toolchain{Class: CompilerMSVC, Binary: "cl", Name: "msvc"}
+	case "clang-cl", "clang-cl.exe":
+		return Toolchain{Class: CompilerMSVC, Binary: "clang-cl", Name: "clang-cl"}
+	case "cc":
+		return Toolchain{Class: CompilerGNU, Binary: "cc", Name: "cc"}
+	default:
+		return Toolchain{Class: CompilerGNU, Binary: ccEnv, Name: lower}
+	}
+}
+
+func ResolveToolchainSpec(spec *dsl.ToolchainSpec) Toolchain {
+	inferred := ResolveToolchain(spec.Name)
+	binary := spec.Binary
+	if binary == "" {
+		binary = inferred.Binary
+	}
+	var class CompilerClass
+	switch spec.Class {
+	case "gnu":
+		class = CompilerGNU
+	case "msvc":
+		class = CompilerMSVC
+	default:
+		class = inferred.Class
+	}
+	return Toolchain{Class: class, Binary: binary, Name: spec.Name}
+}
+
+func defaultCC() string {
+	if runtime.GOOS == "windows" {
+		return "cl"
+	}
+	return "cc"
+}
+
+func Compile(cfg dsl.BuildConfig, tc Toolchain, outputPath string) ([]string, error) {
+	switch tc.Class {
+	case CompilerGNU:
+		return compileGNU(cfg, tc, outputPath), nil
+	case CompilerMSVC:
+		return compileMSVC(cfg, tc, outputPath), nil
+	default:
+		return nil, fmt.Errorf("unknown compiler class for toolchain %q", tc.Name)
+	}
+}
+
+func compileGNU(cfg dsl.BuildConfig, tc Toolchain, outputPath string) []string {
+	argv := []string{tc.Binary}
+
+	if cfg.Standard != "" {
+		argv = append(argv, "-std="+cfg.Standard)
+	}
+
+	switch cfg.Profile {
+	case dsl.ProfileRelease:
+		argv = append(argv, "-O2")
+	case dsl.ProfileDebug:
+		argv = append(argv, "-O0", "-g")
+	case dsl.ProfileSanitized:
+		argv = append(argv, "-O1", "-g", "-fno-omit-frame-pointer")
+	}
+
+	switch cfg.Warnings {
+	case dsl.WarningsStrict:
+		argv = append(argv, "-Wall", "-Wextra")
+	case dsl.WarningsPedantic:
+		argv = append(argv, "-Wall", "-Wextra", "-Wpedantic")
+	}
+
+	if len(cfg.Sanitize) > 0 {
+		argv = append(argv, "-fsanitize="+strings.Join(cfg.Sanitize, ","))
+	}
+
+	for _, inc := range cfg.Includes {
+		argv = append(argv, "-I"+inc)
+	}
+
+	for _, k := range sortedKeys(cfg.Defines) {
+		v := cfg.Defines[k]
+		if v == "" {
+			argv = append(argv, "-D"+k)
+		} else {
+			argv = append(argv, fmt.Sprintf("-D%s=%s", k, v))
+		}
+	}
+
+	argv = append(argv, cfg.Extra...)
+	argv = append(argv, cfg.Sources...)
+	argv = append(argv, "-o", outputPath)
+
+	for _, lib := range cfg.Link {
+		argv = append(argv, "-l"+lib)
+	}
+	return argv
+}
+
+func compileMSVC(cfg dsl.BuildConfig, tc Toolchain, outputPath string) []string {
+	argv := []string{tc.Binary, "/nologo"}
+
+	if cfg.Standard != "" {
+		switch cfg.Standard {
+		case "c11":
+			cfg.Standard = "c17"
+		case "c23":
+			cfg.Standard = "clatest"
+		case "c++23":
+			cfg.Standard = "c++23preview"
+		}
+		argv = append(argv, "/std:"+cfg.Standard)
+	}
+
+	switch cfg.Profile {
+	case dsl.ProfileRelease:
+		argv = append(argv, "/O2")
+	case dsl.ProfileDebug:
+		argv = append(argv, "/Od", "/Zi")
+	case dsl.ProfileSanitized:
+		argv = append(argv, "/Od", "/Zi", "/fsanitize=address")
+	}
+
+	switch cfg.Warnings {
+	case dsl.WarningsStrict:
+		argv = append(argv, "/W4")
+	case dsl.WarningsPedantic:
+		argv = append(argv, "/W4", "/permissive-")
+	}
+
+	if slices.Contains(cfg.Sanitize, "address") && cfg.Profile != dsl.ProfileSanitized {
+		argv = append(argv, "/fsanitize=address")
+	}
+
+	for _, inc := range cfg.Includes {
+		argv = append(argv, "/I"+inc)
+	}
+
+	for _, k := range sortedKeys(cfg.Defines) {
+		v := cfg.Defines[k]
+		if v == "" {
+			argv = append(argv, "/D"+k)
+		} else {
+			argv = append(argv, fmt.Sprintf("/D%s=%s", k, v))
+		}
+	}
+
+	argv = append(argv, cfg.Extra...)
+	argv = append(argv, cfg.Sources...)
+	argv = append(argv, "/Fe:"+outputPath)
+	return argv
+}
+
+func sortedKeys(m map[string]string) []string {
+	if len(m) == 0 {
+		return nil
+	}
+	keys := make([]string, 0, len(m))
+	for k := range m {
+		keys = append(keys, k)
+	}
+	sort.Strings(keys)
+	return keys
+}
--- a/runner/compiler_test.go
+++ b/runner/compiler_test.go
@@ -0,0 +1,294 @@
+package runner
+
+import (
+	"reflect"
+	"strings"
+	"testing"
+
+	"github.com/Mond1c/judge/dsl"
+)
+
+func TestResolveToolchainKnown(t *testing.T) {
+	cases := []struct {
+		in        string
+		wantClass CompilerClass
+		wantName  string
+	}{
+		{"gcc", CompilerGNU, "gcc"},
+		{"g++", CompilerGNU, "gcc"},
+		{"clang", CompilerGNU, "clang"},
+		{"clang++", CompilerGNU, "clang"},
+		{"cl", CompilerMSVC, "msvc"},
+		{"cl.exe", CompilerMSVC, "msvc"},
+		{"msvc", CompilerMSVC, "msvc"},
+		{"clang-cl", CompilerMSVC, "clang-cl"},
+		{"cc", CompilerGNU, "cc"},
+	}
+	for _, c := range cases {
+		t.Run(c.in, func(t *testing.T) {
+			tc := ResolveToolchain(c.in)
+			if tc.Class != c.wantClass {
+				t.Errorf("class: got %v, want %v", tc.Class, c.wantClass)
+			}
+			if tc.Name != c.wantName {
+				t.Errorf("name: got %q, want %q", tc.Name, c.wantName)
+			}
+		})
+	}
+}
+
+func TestResolveToolchainSpec(t *testing.T) {
+	cases := []struct {
+		name      string
+		spec      dsl.ToolchainSpec
+		wantClass CompilerClass
+		wantBin   string
+		wantName  string
+	}{
+		{
+			"gcc inferred",
+			dsl.ToolchainSpec{Name: "gcc", Platforms: []string{"linux"}},
+			CompilerGNU, "gcc", "gcc",
+		},
+		{
+			"msvc inferred",
+			dsl.ToolchainSpec{Name: "msvc", Platforms: []string{"windows"}},
+			CompilerMSVC, "cl", "msvc",
+		},
+		{
+			"nvcc with explicit class",
+			dsl.ToolchainSpec{Name: "nvcc", Platforms: []string{"linux"}, Class: "gnu"},
+			CompilerGNU, "nvcc", "nvcc",
+		},
+		{
+			"custom binary override",
+			dsl.ToolchainSpec{Name: "clang", Platforms: []string{"linux", "windows"}, Binary: "clang-17"},
+			CompilerGNU, "clang-17", "clang",
+		},
+		{
+			"unknown name, explicit class",
+			dsl.ToolchainSpec{Name: "hipcc", Platforms: []string{"linux"}, Class: "gnu"},
+			CompilerGNU, "hipcc", "hipcc",
+		},
+	}
+	for _, c := range cases {
+		t.Run(c.name, func(t *testing.T) {
+			got := ResolveToolchainSpec(&c.spec)
+			if got.Class != c.wantClass {
+				t.Errorf("class: got %v, want %v", got.Class, c.wantClass)
+			}
+			if got.Binary != c.wantBin {
+				t.Errorf("binary: got %q, want %q", got.Binary, c.wantBin)
+			}
+			if got.Name != c.wantName {
+				t.Errorf("name: got %q, want %q", got.Name, c.wantName)
+			}
+		})
+	}
+}
+
+func TestResolveToolchainUnknownFallsBackToGNU(t *testing.T) {
+	tc := ResolveToolchain("gcc-13")
+	if tc.Class != CompilerGNU {
+		t.Errorf("unknown compiler should fall back to GNU, got %v", tc.Class)
+	}
+	if tc.Binary != "gcc-13" {
+		t.Errorf("binary should preserve the original string, got %q", tc.Binary)
+	}
+	if tc.Name != "gcc-13" {
+		t.Errorf("name should be the lowercased original, got %q", tc.Name)
+	}
+}
+
+func TestCompileGNURelease(t *testing.T) {
+	cfg := dsl.BuildConfig{
+		Language: "c",
+		Standard: "c11",
+		Sources:  []string{"solution.c"},
+		Profile:  dsl.ProfileRelease,
+		Warnings: dsl.WarningsStrict,
+	}
+	tc := Toolchain{Class: CompilerGNU, Binary: "gcc", Name: "gcc"}
+	argv, err := Compile(cfg, tc, "solution")
+	if err != nil {
+		t.Fatal(err)
+	}
+	want := []string{"gcc", "-std=c11", "-O2", "-Wall", "-Wextra", "solution.c", "-o", "solution"}
+	if !reflect.DeepEqual(argv, want) {
+		t.Errorf("argv =\n  %v\nwant\n  %v", argv, want)
+	}
+}
+
+func TestCompileGNUDebug(t *testing.T) {
+	cfg := dsl.BuildConfig{
+		Standard: "c11",
+		Sources:  []string{"a.c", "b.c"},
+		Profile:  dsl.ProfileDebug,
+	}
+	tc := Toolchain{Class: CompilerGNU, Binary: "gcc"}
+	argv, _ := Compile(cfg, tc, "out")
+	if !containsSubsequence(argv, []string{"-O0", "-g"}) {
+		t.Errorf("debug flags missing: %v", argv)
+	}
+	if !containsSubsequence(argv, []string{"a.c", "b.c", "-o", "out"}) {
+		t.Errorf("sources and output order wrong: %v", argv)
+	}
+}
+
+func TestCompileGNUSanitized(t *testing.T) {
+	cfg := dsl.BuildConfig{
+		Standard: "c11",
+		Sources:  []string{"s.c"},
+		Profile:  dsl.ProfileSanitized,
+		Sanitize: []string{"address", "undefined"},
+	}
+	tc := Toolchain{Class: CompilerGNU, Binary: "clang"}
+	argv, _ := Compile(cfg, tc, "s")
+	joined := strings.Join(argv, " ")
+	if !strings.Contains(joined, "-fsanitize=address,undefined") {
+		t.Errorf("sanitize flag missing: %v", argv)
+	}
+	if !strings.Contains(joined, "-O1") {
+		t.Errorf("-O1 missing for sanitized profile: %v", argv)
+	}
+}
+
+func TestCompileGNUIncludesAndDefinesAndLink(t *testing.T) {
+	cfg := dsl.BuildConfig{
+		Sources:  []string{"m.c"},
+		Includes: []string{"include", "vendor/inc"},
+		Defines:  map[string]string{"FOO": "1", "BAR": ""},
+		Link:     []string{"m", "pthread"},
+		Extra:    []string{"-fno-strict-aliasing"},
+	}
+	tc := Toolchain{Class: CompilerGNU, Binary: "gcc"}
+	argv, _ := Compile(cfg, tc, "out")
+
+	joined := strings.Join(argv, " ")
+	for _, want := range []string{"-Iinclude", "-Ivendor/inc", "-DFOO=1", "-DBAR", "-lm", "-lpthread", "-fno-strict-aliasing"} {
+		if !strings.Contains(joined, want) {
+			t.Errorf("missing %q in %v", want, argv)
+		}
+	}
+
+	oIdx := indexOf(argv, "-o")
+	lmIdx := indexOf(argv, "-lm")
+	if oIdx == -1 || lmIdx == -1 || lmIdx < oIdx {
+		t.Errorf("-lm must come after -o: %v", argv)
+	}
+}
+
+func TestCompileGNUDefinesOrderDeterministic(t *testing.T) {
+	cfg := dsl.BuildConfig{
+		Sources: []string{"s.c"},
+		Defines: map[string]string{"Z": "1", "A": "2", "M": "3"},
+	}
+	tc := Toolchain{Class: CompilerGNU, Binary: "gcc"}
+	argv1, _ := Compile(cfg, tc, "s")
+	for range 20 {
+		argv2, _ := Compile(cfg, tc, "s")
+		if !reflect.DeepEqual(argv1, argv2) {
+			t.Fatalf("defines order not deterministic:\n  %v\n  %v", argv1, argv2)
+		}
+	}
+}
+
+func TestCompileMSVCRelease(t *testing.T) {
+	cfg := dsl.BuildConfig{
+		Standard: "c11",
+		Sources:  []string{"solution.c"},
+		Profile:  dsl.ProfileRelease,
+		Warnings: dsl.WarningsStrict,
+	}
+	tc := Toolchain{Class: CompilerMSVC, Binary: "cl", Name: "msvc"}
+	argv, _ := Compile(cfg, tc, "solution.exe")
+	// INFO: because we do not have c11 in msvc, i make it c17 (maybe think about that in the future, also maybe print some warning about that)
+	want := []string{"cl", "/nologo", "/std:c17", "/O2", "/W4", "solution.c", "/Fe:solution.exe"}
+	if !reflect.DeepEqual(argv, want) {
+		t.Errorf("argv =\n  %v\nwant\n  %v", argv, want)
+	}
+}
+
+func TestCompileMSVCDebug(t *testing.T) {
+	cfg := dsl.BuildConfig{
+		Sources: []string{"s.c"},
+		Profile: dsl.ProfileDebug,
+	}
+	tc := Toolchain{Class: CompilerMSVC, Binary: "cl"}
+	argv, _ := Compile(cfg, tc, "s.exe")
+	joined := strings.Join(argv, " ")
+	for _, want := range []string{"/Od", "/Zi"} {
+		if !strings.Contains(joined, want) {
+			t.Errorf("missing %q in %v", want, argv)
+		}
+	}
+}
+
+func TestCompileMSVCSanitizedAddressOnly(t *testing.T) {
+	cfg := dsl.BuildConfig{
+		Sources:  []string{"s.c"},
+		Profile:  dsl.ProfileSanitized,
+		Sanitize: []string{"address", "undefined", "thread"},
+	}
+	tc := Toolchain{Class: CompilerMSVC, Binary: "cl"}
+	argv, _ := Compile(cfg, tc, "s.exe")
+	joined := strings.Join(argv, " ")
+	if !strings.Contains(joined, "/fsanitize=address") {
+		t.Errorf("msvc should emit /fsanitize=address for sanitized profile: %v", argv)
+	}
+	if strings.Contains(joined, "undefined") || strings.Contains(joined, "thread") {
+		t.Errorf("msvc should drop unsupported sanitizers, got: %v", argv)
+	}
+}
+
+func TestCompileMSVCIncludesAndDefines(t *testing.T) {
+	cfg := dsl.BuildConfig{
+		Sources:  []string{"m.c"},
+		Includes: []string{"include"},
+		Defines:  map[string]string{"FOO": "1", "BAR": ""},
+	}
+	tc := Toolchain{Class: CompilerMSVC, Binary: "cl"}
+	argv, _ := Compile(cfg, tc, "m.exe")
+	joined := strings.Join(argv, " ")
+	for _, want := range []string{"/Iinclude", "/DFOO=1", "/DBAR"} {
+		if !strings.Contains(joined, want) {
+			t.Errorf("missing %q in %v", want, argv)
+		}
+	}
+}
+
+func TestCompileUnknownClassErrors(t *testing.T) {
+	cfg := dsl.BuildConfig{Sources: []string{"s.c"}}
+	tc := Toolchain{Class: CompilerUnknown, Binary: "weird"}
+	if _, err := Compile(cfg, tc, "s"); err == nil {
+		t.Error("expected error for unknown compiler class")
+	}
+}
+
+func containsSubsequence(haystack, needle []string) bool {
+	if len(needle) == 0 {
+		return true
+	}
+	for i := 0; i+len(needle) <= len(haystack); i++ {
+		match := true
+		for j := range needle {
+			if haystack[i+j] != needle[j] {
+				match = false
+				break
+			}
+		}
+		if match {
+			return true
+		}
+	}
+	return false
+}
+
+func indexOf(xs []string, x string) int {
+	for i, v := range xs {
+		if v == x {
+			return i
+		}
+	}
+	return -1
+}
--- a/runner/expander.go
+++ b/runner/expander.go
@@ -16,47 +16,86 @@ func expandPattern(pattern *dsl.Pattern) ([]*dsl.Test, error) {
 	return expandGlobPattern(pattern)
 }

-func expandGlobPattern(pattern *dsl.Pattern) ([]*dsl.Test, error) {
-	inputFiles, err := filepath.Glob(pattern.InputGlob)
+type patternCase struct {
+	name       string
+	inputPath  string
+	outputPath string
+	dir        string
+}
+
+func globWithAffixes(pattern string) ([]string, string, string, error) {
+	files, err := filepath.Glob(pattern)
 	if err != nil {
-		return nil, fmt.Errorf("invalid input glob %q: %w", pattern.InputGlob, err)
+		return nil, "", "", fmt.Errorf("invalid glob %q: %w", pattern, err)
 	}
-	if len(inputFiles) == 0 {
-		return nil, fmt.Errorf("no files matched input glob %q", pattern.InputGlob)
+	if len(files) == 0 {
+		return nil, "", "", fmt.Errorf("no files matched glob %q", pattern)
+	}
+	prefix, suffix := splitGlob(pattern)
+	return files, prefix, suffix, nil
+}
+
+func expandGlobPattern(pattern *dsl.Pattern) ([]*dsl.Test, error) {
+	inputIsGlob := strings.Contains(pattern.InputGlob, "*")
+	outputIsGlob := strings.Contains(pattern.OutputGlob, "*")
+
+	if pattern.InputGlob == "" && pattern.OutputGlob == "" {
+		return nil, fmt.Errorf("pattern needs at least one of input/output/dirs")
+	}
+	if !inputIsGlob && !outputIsGlob {
+		return nil, fmt.Errorf("pattern needs at least one glob field (input or output must contain *)")
 	}

-	inputPrefix, inputSuffix := splitGlob(pattern.InputGlob)
-	outputPrefix, outputSuffix := splitGlob(pattern.OutputGlob)
+	var cases []patternCase

-	var tests []*dsl.Test
-	for _, inputPath := range inputFiles {
-		wildcard := extractWildcard(inputPath, inputPrefix, inputSuffix)
-		outputPath := outputPrefix + wildcard + outputSuffix
-
-		inputContent, err := os.ReadFile(inputPath)
+	// TODO: i know that this is copypaste, but i do not want make clousers or ifs inside cycle for now
+	switch {
+	case inputIsGlob && outputIsGlob:
+		inputFiles, inputPrefix, inputSuffix, err := globWithAffixes(pattern.InputGlob)
 		if err != nil {
-			return nil, fmt.Errorf("read input %q: %w", inputPath, err)
+			return nil, err
 		}
-		outputContent, err := os.ReadFile(outputPath)
-		if err != nil {
-			return nil, fmt.Errorf("read output %q: %w", outputPath, err)
+		outputPrefix, outputSuffix := splitGlob(pattern.OutputGlob)
+		for _, inputPath := range inputFiles {
+			wildcard := extractWildcard(inputPath, inputPrefix, inputSuffix)
+			outputPath := outputPrefix + wildcard + outputSuffix
+			cases = append(cases, patternCase{
+				name:       wildcard,
+				inputPath:  inputPath,
+				outputPath: outputPath,
+			})
 		}

-		name := fmt.Sprintf("pattern:%s", wildcard)
-		stdin := string(inputContent)
-		expected := string(outputContent)
+	case inputIsGlob && !outputIsGlob:
+		inputFiles, inputPrefix, inputSuffix, err := globWithAffixes(pattern.InputGlob)
+		if err != nil {
+			return nil, err
+		}
+		for _, inputPath := range inputFiles {
+			wildcard := extractWildcard(inputPath, inputPrefix, inputSuffix)
+			cases = append(cases, patternCase{
+				name:       wildcard,
+				inputPath:  inputPath,
+				outputPath: pattern.OutputGlob,
+			})
+		}

-		tests = append(tests, &dsl.Test{
-			Name:     name,
-			Stdin:    &stdin,
-			Env:      map[string]string{},
-			InFiles:  map[string]string{},
-			OutFiles: map[string]string{},
-			Stdout:   dsl.ExactMatcher{Value: expected},
-			Stderr:   dsl.NoMatcher{},
-		})
+	case !inputIsGlob && outputIsGlob:
+		outputFiles, outputPrefix, outputSuffix, err := globWithAffixes(pattern.OutputGlob)
+		if err != nil {
+			return nil, err
+		}
+		for _, outputPath := range outputFiles {
+			wildcard := extractWildcard(outputPath, outputPrefix, outputSuffix)
+			cases = append(cases, patternCase{
+				name:       wildcard,
+				inputPath:  pattern.InputGlob,
+				outputPath: outputPath,
+			})
+		}
 	}
-	return tests, nil
+
+	return buildTests(cases, pattern.Args)
 }

 func expandDirPattern(pattern *dsl.Pattern) ([]*dsl.Test, error) {
@@ -68,42 +107,102 @@ func expandDirPattern(pattern *dsl.Pattern) ([]*dsl.Test, error) {
 		return nil, fmt.Errorf("no directories matched %q", pattern.DirsGlob)
 	}

-	var tests []*dsl.Test
+	var cases []patternCase
 	for _, dir := range dirs {
 		info, err := os.Stat(dir)
 		if err != nil || !info.IsDir() {
 			continue
 		}
-
-		inputPath := filepath.Join(dir, pattern.InputFile)
-		outputPath := filepath.Join(dir, pattern.OutputFile)
-
-		inputContent, err := os.ReadFile(inputPath)
-		if err != nil {
-			return nil, fmt.Errorf("read %q: %w", inputPath, err)
-		}
-		outputContent, err := os.ReadFile(outputPath)
-		if err != nil {
-			return nil, fmt.Errorf("read %q: %w", outputPath, err)
-		}
-
-		name := fmt.Sprintf("pattern:%s", filepath.Base(dir))
-		stdin := string(inputContent)
-		expected := string(outputContent)
-
-		tests = append(tests, &dsl.Test{
-			Name:     name,
-			Stdin:    &stdin,
-			Env:      map[string]string{},
-			InFiles:  map[string]string{},
-			OutFiles: map[string]string{},
-			Stdout:   dsl.ExactMatcher{Value: expected},
-			Stderr:   dsl.NoMatcher{},
+		cases = append(cases, patternCase{
+			name:       filepath.Base(dir),
+			inputPath:  filepath.Join(dir, pattern.InputFile),
+			outputPath: filepath.Join(dir, pattern.OutputFile),
+			dir:        dir,
 		})
 	}
+	return buildTests(cases, pattern.Args)
+}
+
+func buildTest(c *patternCase, argTemplate []string, useInputAsFile, useOutputAsFile bool) (*dsl.Test, error) {
+	inputContent, err := os.ReadFile(c.inputPath)
+	if err != nil {
+		return nil, fmt.Errorf("read input %q: %w", c.inputPath, err)
+	}
+	outputContent, err := os.ReadFile(c.outputPath)
+	if err != nil {
+		return nil, fmt.Errorf("read output %q: %w", c.outputPath, err)
+	}
+
+	t := &dsl.Test{
+		Name:     fmt.Sprintf("pattern:%s", c.name),
+		Env:      map[string]string{},
+		InFiles:  map[string]string{},
+		OutFiles: map[string]string{},
+		Stderr:   dsl.NoMatcher{},
+	}
+
+	inputName := filepath.Base(c.inputPath)
+	outputName := filepath.Base(c.outputPath)
+
+	if useInputAsFile {
+		t.SetInputFile(inputName, inputContent)
+	} else {
+		t.SetStdin(inputContent)
+	}
+
+	if useOutputAsFile {
+		t.SetOutputFile(outputName, outputContent)
+	} else {
+		t.SetStdout(outputContent)
+	}
+
+	if len(argTemplate) > 0 {
+		t.Args = substituteArgs(argTemplate, map[string]string{
+			"{input_path}":  inputName,
+			"{output_path}": outputName,
+			"{name}":        c.name,
+			"{dir}":         c.dir,
+		})
+	}
+
+	return t, nil
+}
+
+func buildTests(cases []patternCase, argTemplate []string) ([]*dsl.Test, error) {
+	useInputAsFile := argsContain(argTemplate, "{input_path}")
+	useOutputAsFile := argsContain(argTemplate, "{output_path}")
+
+	var tests []*dsl.Test
+	for _, c := range cases {
+		t, err := buildTest(&c, argTemplate, useInputAsFile, useOutputAsFile)
+		if err != nil {
+			return nil, err
+		}
+		tests = append(tests, t)
+	}
 	return tests, nil
 }

+func argsContain(args []string, placeholder string) bool {
+	for _, a := range args {
+		if strings.Contains(a, placeholder) {
+			return true
+		}
+	}
+	return false
+}
+
+func substituteArgs(template []string, vars map[string]string) []string {
+	out := make([]string, len(template))
+	for i, a := range template {
+		for k, v := range vars {
+			a = strings.ReplaceAll(a, k, v)
+		}
+		out[i] = a
+	}
+	return out
+}
+
 func splitGlob(pattern string) (prefix, suffix string) {
 	before, after, found := strings.Cut(pattern, "*")
 	if !found {
--- a/runner/expander_test.go
+++ b/runner/expander_test.go
@@ -0,0 +1,240 @@
+package runner
+
+import (
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+
+	"github.com/Mond1c/judge/dsl"
+)
+
+func writeFile(t *testing.T, dir, name, content string) {
+	t.Helper()
+	path := filepath.Join(dir, name)
+	if err := os.MkdirAll(filepath.Dir(path), 0755); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(path, []byte(content), 0644); err != nil {
+		t.Fatal(err)
+	}
+}
+
+func TestExpandGlobPairedStdioMode(t *testing.T) {
+	dir := t.TempDir()
+	writeFile(t, dir, "tests/01.in", "1 2 3\n")
+	writeFile(t, dir, "tests/01.ans", "6\n")
+	writeFile(t, dir, "tests/02.in", "10 20\n")
+	writeFile(t, dir, "tests/02.ans", "30\n")
+
+	cwd, _ := os.Getwd()
+	defer os.Chdir(cwd)
+	os.Chdir(dir)
+
+	tests, err := expandPattern(&dsl.Pattern{
+		InputGlob:  "tests/*.in",
+		OutputGlob: "tests/*.ans",
+	})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if len(tests) != 2 {
+		t.Fatalf("expected 2 tests, got %d", len(tests))
+	}
+	for _, tc := range tests {
+		if tc.Stdin == nil {
+			t.Errorf("%s: stdin should be set in stdio mode", tc.Name)
+		}
+		if len(tc.Args) != 0 {
+			t.Errorf("%s: args should be empty without template", tc.Name)
+		}
+		if _, ok := tc.Stdout.(dsl.ExactMatcher); !ok {
+			t.Errorf("%s: stdout should be ExactMatcher, got %T", tc.Name, tc.Stdout)
+		}
+	}
+}
+
+func TestExpandGlobWithSharedOutput(t *testing.T) {
+	dir := t.TempDir()
+	writeFile(t, dir, "tests/01.in", "1\n")
+	writeFile(t, dir, "tests/02.in", "2\n")
+	writeFile(t, dir, "tests/03.in", "3\n")
+	writeFile(t, dir, "expected.ans", "ok\n")
+
+	cwd, _ := os.Getwd()
+	defer os.Chdir(cwd)
+	os.Chdir(dir)
+
+	tests, err := expandPattern(&dsl.Pattern{
+		InputGlob:  "tests/*.in",
+		OutputGlob: "expected.ans",
+	})
+	if err != nil {
+		t.Fatalf("expand: %v", err)
+	}
+	if len(tests) != 3 {
+		t.Fatalf("expected 3 tests, got %d", len(tests))
+	}
+	for _, tc := range tests {
+		m, ok := tc.Stdout.(dsl.ExactMatcher)
+		if !ok {
+			t.Fatalf("%s: stdout should be ExactMatcher, got %T", tc.Name, tc.Stdout)
+		}
+		if m.Value != "ok\n" {
+			t.Errorf("%s: shared output = %q, want %q", tc.Name, m.Value, "ok\n")
+		}
+	}
+}
+
+func TestExpandGlobFileModeInputOnly(t *testing.T) {
+	dir := t.TempDir()
+	writeFile(t, dir, "tests/01.in", "hello\n")
+	writeFile(t, dir, "tests/01.ans", "HELLO\n")
+
+	cwd, _ := os.Getwd()
+	defer os.Chdir(cwd)
+	os.Chdir(dir)
+
+	tests, err := expandPattern(&dsl.Pattern{
+		InputGlob:  "tests/*.in",
+		OutputGlob: "tests/*.ans",
+		Args:       []string{"{input_path}"},
+	})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if len(tests) != 1 {
+		t.Fatalf("expected 1 test, got %d", len(tests))
+	}
+	tc := tests[0]
+	if tc.Stdin != nil {
+		t.Errorf("stdin should be nil in file mode for input")
+	}
+	if content, ok := tc.InFiles["01.in"]; !ok || content != "hello\n" {
+		t.Errorf("InFiles[01.in] = %q, want %q", content, "hello\n")
+	}
+	if _, ok := tc.Stdout.(dsl.ExactMatcher); !ok {
+		t.Errorf("stdout should still be ExactMatcher when output not in file mode, got %T", tc.Stdout)
+	}
+	if len(tc.Args) != 1 || tc.Args[0] != "01.in" {
+		t.Errorf("args = %v, want [01.in]", tc.Args)
+	}
+}
+
+func TestExpandGlobFileModeBoth(t *testing.T) {
+	dir := t.TempDir()
+	writeFile(t, dir, "tests/01.in", "1 2 3\n")
+	writeFile(t, dir, "tests/01.ans", "6\n")
+
+	cwd, _ := os.Getwd()
+	defer os.Chdir(cwd)
+	os.Chdir(dir)
+
+	tests, err := expandPattern(&dsl.Pattern{
+		InputGlob:  "tests/*.in",
+		OutputGlob: "tests/*.ans",
+		Args:       []string{"{input_path}", "{output_path}"},
+	})
+	if err != nil {
+		t.Fatal(err)
+	}
+	tc := tests[0]
+	if tc.Stdin != nil {
+		t.Error("stdin should be nil")
+	}
+	if _, ok := tc.Stdout.(dsl.NoMatcher); !ok {
+		t.Errorf("stdout should be NoMatcher when output is file, got %T", tc.Stdout)
+	}
+	if content := tc.OutFiles["01.ans"]; content != "6\n" {
+		t.Errorf("OutFiles[01.ans] = %q, want %q", content, "6\n")
+	}
+	if len(tc.Args) != 2 || tc.Args[0] != "01.in" || tc.Args[1] != "01.ans" {
+		t.Errorf("args = %v, want [01.in 01.ans]", tc.Args)
+	}
+}
+
+func TestExpandGlobArgsWithStaticAndPlaceholders(t *testing.T) {
+	dir := t.TempDir()
+	writeFile(t, dir, "tests/01.in", "x\n")
+	writeFile(t, dir, "tests/01.ans", "x\n")
+
+	cwd, _ := os.Getwd()
+	defer os.Chdir(cwd)
+	os.Chdir(dir)
+
+	tests, err := expandPattern(&dsl.Pattern{
+		InputGlob:  "tests/*.in",
+		OutputGlob: "tests/*.ans",
+		Args:       []string{"--mode", "strict", "{input_path}"},
+	})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if len(tests[0].Args) != 3 || tests[0].Args[0] != "--mode" || tests[0].Args[2] != "01.in" {
+		t.Errorf("args = %v", tests[0].Args)
+	}
+}
+
+func TestExpandDirModeWithArgs(t *testing.T) {
+	dir := t.TempDir()
+	writeFile(t, dir, "cases/a/input.txt", "one\n")
+	writeFile(t, dir, "cases/a/expected.txt", "ONE\n")
+	writeFile(t, dir, "cases/b/input.txt", "two\n")
+	writeFile(t, dir, "cases/b/expected.txt", "TWO\n")
+
+	cwd, _ := os.Getwd()
+	defer os.Chdir(cwd)
+	os.Chdir(dir)
+
+	tests, err := expandPattern(&dsl.Pattern{
+		DirsGlob:   "cases/*",
+		InputFile:  "input.txt",
+		OutputFile: "expected.txt",
+		Args:       []string{"{input_path}", "{output_path}"},
+	})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if len(tests) != 2 {
+		t.Fatalf("expected 2 tests, got %d", len(tests))
+	}
+	for _, tc := range tests {
+		if _, ok := tc.InFiles["input.txt"]; !ok {
+			t.Errorf("%s: missing InFiles[input.txt]", tc.Name)
+		}
+		if _, ok := tc.OutFiles["expected.txt"]; !ok {
+			t.Errorf("%s: missing OutFiles[expected.txt]", tc.Name)
+		}
+		if len(tc.Args) != 2 {
+			t.Errorf("%s: args = %v", tc.Name, tc.Args)
+		}
+	}
+}
+
+func TestExpandPatternRejectsAllLiterals(t *testing.T) {
+	_, err := expandPattern(&dsl.Pattern{
+		InputGlob:  "tests/a.in",
+		OutputGlob: "tests/a.ans",
+	})
+	if err == nil {
+		t.Fatal("expected error when no glob fields")
+	}
+	if !strings.Contains(err.Error(), "glob") {
+		t.Errorf("error %q does not mention glob", err.Error())
+	}
+}
+
+func TestExpandPatternNoMatches(t *testing.T) {
+	dir := t.TempDir()
+	cwd, _ := os.Getwd()
+	defer os.Chdir(cwd)
+	os.Chdir(dir)
+
+	_, err := expandPattern(&dsl.Pattern{
+		InputGlob:  "missing/*.in",
+		OutputGlob: "missing/*.ans",
+	})
+	if err == nil {
+		t.Fatal("expected error on zero matches")
+	}
+}
--- a/runner/limiter_linux.go
+++ b/runner/limiter_linux.go
@@ -102,9 +102,9 @@ func createScopeCgroup() (string, error) {
 		return "", fmt.Errorf("read /proc/self/cgroup: %w", err)
 	}
 	var rel string
-	for _, line := range strings.Split(strings.TrimSpace(string(data)), "\n") {
-		if strings.HasPrefix(line, "0::") {
-			rel = strings.TrimPrefix(line, "0::")
+	for line := range strings.SplitSeq(strings.TrimSpace(string(data)), "\n") {
+		if after, ok := strings.CutPrefix(line, "0::"); ok {
+			rel = after
 			break
 		}
 	}
@@ -200,7 +200,7 @@ func (l *linuxLimiter) collect() limitStats {
 		}
 	}
 	if data, err := os.ReadFile(filepath.Join(l.cgPath, "memory.events")); err == nil {
-		for _, line := range strings.Split(string(data), "\n") {
+		for line := range strings.SplitSeq(string(data), "\n") {
 			fields := strings.Fields(line)
 			if len(fields) != 2 {
 				continue
@@ -217,7 +217,7 @@ func (l *linuxLimiter) cleanup() {
 	if l.cgPath == "" {
 		return
 	}
-	for i := 0; i < 10; i++ {
+	for range 10 {
 		err := os.Remove(l.cgPath)
 		if err == nil || os.IsNotExist(err) {
 			l.cgPath = ""
--- a/runner/result.go
+++ b/runner/result.go
@@ -2,6 +2,7 @@ package runner

 import (
 	"fmt"
+	"math"
 	"time"
 )

@@ -40,8 +41,8 @@ type TestResult struct {
 	Status  Status
 	Elapsed time.Duration

-	PeakMemory  int64 // bytes; 0 if not measured
-	MemoryLimit int64 // bytes; 0 if unlimited
+	PeakMemory  int64
+	MemoryLimit int64

 	Failures []string

@@ -64,8 +65,40 @@ type GroupResult struct {
 	Total  int
 }

-type SuiteResult struct {
+type BuildRun struct {
+	Name       string
+	Toolchain  string
+	Skipped    bool
+	SkipReason string
+
+	BuildLog   string
 	Groups     []*GroupResult
 	TotalScore float64
-	BuildLog   string
+}
+
+type SuiteResult struct {
+	Builds []*BuildRun
+
+	TotalScore float64
+}
+
+func (r *SuiteResult) AggregateScore() float64 {
+	if len(r.Builds) == 0 {
+		return 0
+	}
+	min := math.Inf(1)
+	anyRan := false
+	for _, b := range r.Builds {
+		if b.Skipped {
+			continue
+		}
+		anyRan = true
+		if b.TotalScore < min {
+			min = b.TotalScore
+		}
+	}
+	if !anyRan {
+		return 1.0
+	}
+	return min
 }
--- a/runner/result_test.go
+++ b/runner/result_test.go
@@ -0,0 +1,95 @@
+package runner
+
+import (
+	"strings"
+	"testing"
+)
+
+func TestStatusString(t *testing.T) {
+	cases := []struct {
+		s    Status
+		want string
+	}{
+		{StatusPass, "PASS"},
+		{StatusFail, "FAIL"},
+		{StatusTLE, "TLE"},
+		{StatusMLE, "MLE"},
+		{StatusBuildError, "BUILD_ERROR"},
+		{StatusRuntimeError, "RE"},
+		{Status(999), "UNKNOWN"},
+	}
+	for _, c := range cases {
+		if got := c.s.String(); got != c.want {
+			t.Errorf("Status(%d).String() = %q, want %q", c.s, got, c.want)
+		}
+	}
+}
+
+func TestAddFailureAppends(t *testing.T) {
+	r := &TestResult{}
+	r.addFailure("first %s", "msg")
+	r.addFailure("second %d", 2)
+	if len(r.Failures) != 2 {
+		t.Fatalf("Failures len = %d, want 2", len(r.Failures))
+	}
+	if r.Failures[0] != "first msg" {
+		t.Errorf("Failures[0] = %q", r.Failures[0])
+	}
+	if !strings.Contains(r.Failures[1], "second 2") {
+		t.Errorf("Failures[1] = %q", r.Failures[1])
+	}
+}
+
+func TestAggregateScoreEmpty(t *testing.T) {
+	r := &SuiteResult{}
+	if got := r.AggregateScore(); got != 0 {
+		t.Errorf("empty aggregate = %v, want 0", got)
+	}
+}
+
+func TestAggregateScoreSingleBuild(t *testing.T) {
+	r := &SuiteResult{
+		Builds: []*BuildRun{{Name: "release", TotalScore: 0.75}},
+	}
+	if got := r.AggregateScore(); got != 0.75 {
+		t.Errorf("single build aggregate = %v, want 0.75", got)
+	}
+}
+
+func TestAggregateScoreTakesMinimum(t *testing.T) {
+	r := &SuiteResult{
+		Builds: []*BuildRun{
+			{Name: "release", TotalScore: 1.0},
+			{Name: "debug", TotalScore: 0.9},
+			{Name: "sanitized", TotalScore: 0.95},
+		},
+	}
+	if got := r.AggregateScore(); got != 0.9 {
+		t.Errorf("aggregate = %v, want 0.9 (minimum)", got)
+	}
+}
+
+func TestAggregateScoreIgnoresSkipped(t *testing.T) {
+	r := &SuiteResult{
+		Builds: []*BuildRun{
+			{Name: "release", TotalScore: 1.0},
+			{Name: "sanitized", Skipped: true, SkipReason: "platforms=linux"},
+			{Name: "debug", TotalScore: 0.8},
+		},
+	}
+	if got := r.AggregateScore(); got != 0.8 {
+		t.Errorf("aggregate with skipped = %v, want 0.8", got)
+	}
+}
+
+func TestAggregateScoreAllSkipped(t *testing.T) {
+	r := &SuiteResult{
+		Builds: []*BuildRun{
+			{Name: "a", Skipped: true},
+			{Name: "b", Skipped: true},
+		},
+	}
+	if got := r.AggregateScore(); got != 1.0 {
+		t.Errorf("all-skipped aggregate = %v, want 1.0", got)
+	}
+}
--- a/runner/runner.go
+++ b/runner/runner.go
@@ -21,6 +21,8 @@ type Config struct {
 	WorkDir    string
 	BinaryName string
 	Wrapper    string
+
+	TargetBuild string
 }

 type Runner struct {
@@ -63,41 +65,246 @@ func (r *Runner) Run() *SuiteResult {

 	result := &SuiteResult{}

-	buildLog, err := r.build()
-	result.BuildLog = buildLog
-	if err != nil {
-		for _, g := range r.file.Groups {
-			gr := &GroupResult{
-				Name:   g.Name,
-				Weight: g.Weight,
-				Score:  0,
-			}
+	if len(r.file.Builds) == 0 {
+		run := r.runLegacyBuild()
+		result.Builds = append(result.Builds, run)
+	} else {
+		result.Builds = r.runStructuredBuilds()
+	}

-			total := len(g.Tests)
-			if g.Pattern != nil {
-				total = -1
-			}
-			gr.Total = total
-			for _, t := range g.Tests {
-				gr.Tests = append(gr.Tests, &TestResult{
-					Name:   t.Name,
-					Status: StatusBuildError,
-				})
-			}
-			result.Groups = append(result.Groups, gr)
-		}
-		return result
+	result.TotalScore = result.AggregateScore()
+	return result
+}
+
+func (r *Runner) runLegacyBuild() *BuildRun {
+	run := &BuildRun{Name: "default"}
+
+	if r.cfg.TargetBuild != "" && r.cfg.TargetBuild != "default" {
+		run.Skipped = true
+		run.SkipReason = fmt.Sprintf("--build=%q selected, but this suite has no structured builds", r.cfg.TargetBuild)
+		return run
+	}
+
+	buildLog, err := r.legacyBuild()
+	run.BuildLog = buildLog
+	if err != nil {
+		r.fillBuildError(run)
+		return run
 	}

 	r.binary = resolveBinary(r.cfg.WorkDir, filepath.Base(r.binary))
+	r.runGroups(run)
+	return run
+}

-	for _, g := range r.file.Groups {
-		gr := r.runGroup(g)
-		result.Groups = append(result.Groups, gr)
-		result.TotalScore += gr.Score
+func (r *Runner) resolveRuntimeToolchain() (Toolchain, string) {
+	goos := runtime.GOOS
+	wanted := os.Getenv("JUDGE_TOOLCHAIN")
+	if wanted == "" {
+		wanted = os.Getenv("JUDGE_CC")
+	}
+	for _, spec := range r.file.Toolchains {
+		if spec.Name == wanted {
+			return ResolveToolchainSpec(spec), goos
+		}
+	}
+	return ResolveToolchain(wanted), goos
+}
+
+func (r *Runner) runStructuredBuilds() []*BuildRun {
+	tc, goos := r.resolveRuntimeToolchain()
+	var runs []*BuildRun
+
+	for _, b := range r.file.Builds {
+		run := &BuildRun{Name: b.Name, Toolchain: tc.Name}
+
+		if r.cfg.TargetBuild != "" && r.cfg.TargetBuild != b.Name {
+			continue
+		}
+
+		effective := b.Resolve(r.file.BuildDefaults, goos)
+		if !effective.AppliesTo(goos, tc.Name) {
+			run.Skipped = true
+			run.SkipReason = fmt.Sprintf("not applicable to %s/%s (platforms=%v, compilers=%v)", goos, tc.Name, effective.Platforms, effective.Compilers)
+			runs = append(runs, run)
+			continue
+		}
+
+		log, binaryPath, err := r.compileStructured(b.Name, effective, tc)
+		run.BuildLog = log
+		if err != nil {
+			run.Groups = r.synthesizeBuildError()
+			run.TotalScore = 0
+			runs = append(runs, run)
+			continue
+		}
+
+		prevBinary := r.binary
+		prevWrapper := r.cfg.Wrapper
+		r.binary = binaryPath
+		if r.cfg.Wrapper == "" && effective.Wrapper != "" {
+			r.cfg.Wrapper = effective.Wrapper
+		}
+		r.runGroups(run)
+		r.binary = prevBinary
+		r.cfg.Wrapper = prevWrapper
+
+		runs = append(runs, run)
 	}

-	return result
+	return runs
+}
+
+func (r *Runner) runGroups(run *BuildRun) {
+	for _, g := range r.file.Groups {
+		gr := r.runGroup(g)
+		run.Groups = append(run.Groups, gr)
+		run.TotalScore += gr.Score
+	}
+}
+
+func (r *Runner) fillBuildError(run *BuildRun) {
+	run.Groups = r.synthesizeBuildError()
+}
+
+func (r *Runner) synthesizeBuildError() []*GroupResult {
+	var out []*GroupResult
+	for _, g := range r.file.Groups {
+		gr := &GroupResult{
+			Name:   g.Name,
+			Weight: g.Weight,
+			Score:  0,
+		}
+		total := len(g.Tests)
+		if g.Pattern != nil {
+			total = -1
+		}
+		gr.Total = total
+		for _, t := range g.Tests {
+			gr.Tests = append(gr.Tests, &TestResult{
+				Name:   t.Name,
+				Status: StatusBuildError,
+			})
+		}
+		out = append(out, gr)
+	}
+	return out
+}
+
+func (r *Runner) legacyBuild() (string, error) {
+	buildCmd := r.buildCommand()
+
+	sources, err := r.findSources()
+	if err != nil {
+		return "", err
+	}
+	if sources != "" {
+		buildCmd = strings.ReplaceAll(buildCmd, "$SOURCES", sources)
+	}
+
+	ctx := context.Background()
+	if r.file.Timeout > 0 {
+		var cancel context.CancelFunc
+		ctx, cancel = context.WithTimeout(ctx, r.file.Timeout)
+		defer cancel()
+	}
+
+	cmd := shellCommand(ctx, buildCmd)
+	cmd.Dir = r.cfg.WorkDir
+	setProcessGroup(cmd)
+	cmd.Env = os.Environ()
+
+	var out bytes.Buffer
+	cmd.Stdout = &out
+	cmd.Stderr = &out
+
+	if err := cmd.Run(); err != nil {
+		killProcessGroup(cmd)
+		return out.String(), fmt.Errorf("build failed: %w\n%s", err, out.String())
+	}
+	return out.String(), nil
+}
+
+func (r *Runner) compileStructured(name string, cfg dsl.BuildConfig, tc Toolchain) (string, string, error) {
+	sources, err := expandSources(r.cfg.WorkDir, cfg.Sources)
+	if err != nil {
+		return "", "", err
+	}
+	if len(sources) == 0 {
+		return "", "", fmt.Errorf("build %q: no sources", name)
+	}
+	cfg.Sources = sources
+
+	outputName := cfg.Output
+	if outputName == "" {
+		outputName = "solution"
+	}
+	if runtime.GOOS == "windows" && !strings.HasSuffix(strings.ToLower(outputName), ".exe") {
+		outputName += ".exe"
+	}
+	buildDir := filepath.Join(r.cfg.WorkDir, "build", name)
+	if err := os.MkdirAll(buildDir, 0755); err != nil {
+		return "", "", fmt.Errorf("mkdir %s: %w", buildDir, err)
+	}
+	outputPath := filepath.Join(buildDir, outputName)
+
+	argv, err := Compile(cfg, tc, outputPath)
+	if err != nil {
+		return "", "", err
+	}
+
+	ctx := context.Background()
+	if r.file.Timeout > 0 {
+		var cancel context.CancelFunc
+		ctx, cancel = context.WithTimeout(ctx, r.file.Timeout)
+		defer cancel()
+	}
+
+	cmd := exec.CommandContext(ctx, argv[0], argv[1:]...)
+	cmd.Dir = r.cfg.WorkDir
+	setProcessGroup(cmd)
+	cmd.Env = os.Environ()
+
+	var out bytes.Buffer
+	cmd.Stdout = &out
+	cmd.Stderr = &out
+
+	logPrefix := fmt.Sprintf("$ %s\n", strings.Join(argv, " "))
+	if err := cmd.Run(); err != nil {
+		killProcessGroup(cmd)
+		return logPrefix + out.String(), "", fmt.Errorf("build %q failed: %w\n%s", name, err, out.String())
+	}
+	return logPrefix + out.String(), outputPath, nil
+}
+
+func expandSources(workDir string, patterns []string) ([]string, error) {
+	var out []string
+	seen := map[string]bool{}
+	for _, pat := range patterns {
+		matches, err := filepath.Glob(filepath.Join(workDir, pat))
+		if err != nil {
+			return nil, fmt.Errorf("glob %q: %w", pat, err)
+		}
+		if len(matches) == 0 {
+			if _, statErr := os.Stat(filepath.Join(workDir, pat)); statErr == nil {
+				matches = []string{filepath.Join(workDir, pat)}
+			} else {
+				return nil, fmt.Errorf("source glob %q matched no files", pat)
+			}
+		}
+		for _, m := range matches {
+			rel, err := filepath.Rel(workDir, m)
+			if err != nil {
+				rel = m
+			}
+			rel = filepath.ToSlash(rel)
+			if !seen[rel] {
+				seen[rel] = true
+				out = append(out, rel)
+			}
+		}
+	}
+	return out, nil
 }

 func (r *Runner) buildCommand() string {
@@ -152,40 +359,6 @@ func (r *Runner) findSources() (string, error) {
 	return strings.Join(files, " "), nil
 }

-func (r *Runner) build() (string, error) {
-	buildCmd := r.buildCommand()
-
-	sources, err := r.findSources()
-	if err != nil {
-		return "", err
-	}
-	if sources != "" {
-		buildCmd = strings.ReplaceAll(buildCmd, "$SOURCES", sources)
-	}
-
-	ctx := context.Background()
-	if r.file.Timeout > 0 {
-		var cancel context.CancelFunc
-		ctx, cancel = context.WithTimeout(ctx, r.file.Timeout)
-		defer cancel()
-	}
-
-	cmd := shellCommand(ctx, buildCmd)
-	cmd.Dir = r.cfg.WorkDir
-	setProcessGroup(cmd)
-	cmd.Env = os.Environ()
-
-	var out bytes.Buffer
-	cmd.Stdout = &out
-	cmd.Stderr = &out
-
-	if err := cmd.Run(); err != nil {
-		killProcessGroup(cmd)
-		return out.String(), fmt.Errorf("build failed: %w\n%s", err, out.String())
-	}
-	return out.String(), nil
-}
-
 func shellCommand(ctx context.Context, cmdline string) *exec.Cmd {
 	if runtime.GOOS == "windows" {
 		return exec.CommandContext(ctx, "cmd", "/C", cmdline)
@@ -385,30 +558,32 @@ func (r *Runner) runTest(t *dsl.Test) *TestResult {
 		tr.addFailure("exit code: expected %d, got %d", *t.ExitCode, actualCode)
 	}

-	for _, f := range applyMatcher("stdout", t.Stdout, tr.ActualStdout) {
+	for _, f := range t.Stdout.Match("stdout", tr.ActualStdout) {
 		tr.addFailure("%s", f)
 	}
-	for _, f := range applyMatcher("stderr", t.Stderr, tr.ActualStderr) {
+	for _, f := range t.Stderr.Match("stderr", tr.ActualStderr) {
 		tr.addFailure("%s", f)
 	}

+	if len(tr.Failures) > 0 {
+		tr.Status = StatusFail
+	}
+
 	for name, expected := range t.OutFiles {
-		path := filepath.Join(tmpDir, name)
-		content, err := os.ReadFile(path)
+		actualPath := filepath.Join(tmpDir, name)
+		data, err := os.ReadFile(actualPath)
 		if err != nil {
-			tr.addFailure("output file %q not found: %v", name, err)
+			tr.Status = StatusFail
+			tr.addFailure("output file %q missing: %v", name, err)
 			continue
 		}
-		actual := normalizeOutput(string(content), r.file)
-		for _, f := range applyMatcher(fmt.Sprintf("file(%s)", name), dsl.ExactMatcher{Value: expected}, actual) {
-			tr.addFailure("%s", f)
+		actual := normalizeOutput(string(data), r.file)
+		if actual != expected {
+			tr.Status = StatusFail
+			tr.addFailure("output file %q mismatch\n  expected: %q\n  actual:   %q", name, expected, actual)
 		}
 	}

-	if tr.Status == StatusPass && len(tr.Failures) > 0 {
-		tr.Status = StatusFail
-	}
-
 	return tr
 }
Author	SHA1	Message	Date
Mikhail Kornilovich	cbc000e872	fix ci All checks were successful go-test / go test (push) Successful in 14s Details Release / Build & publish (push) Successful in 1m8s Details	2026-04-18 09:58:03 +03:00
Mikhail Kornilovich	c023831222	add to release flow darwin/amd64 Some checks failed go-test / go test (push) Successful in 21s Details Release / Build & publish (push) Failing after 28s Details	2026-04-18 09:55:18 +03:00
Mikhail Kornilovich	7f14f8c236	test: expand dsl/runner coverage and add go-test CI workflow All checks were successful go-test / go test (push) Successful in 56s Details - Add dsl/matcher_test.go covering ExactMatcher, ContainsMatcher, RegexMatcher, NumericEpsMatcher, AnyOrderMatcher and NoMatcher — previously 0% — including epsilon, count mismatch, unparsable numbers, invalid regex, and splitLines edge cases. - Add dsl/ast_test.go for the new Test.SetInputFile / SetStdin / SetOutputFile / SetStdout helpers and Pattern.IsDirMode. - Add dsl/build_string_test.go covering BuildProfile.String, WarningLevel.String and BuildConfig.MergeFrom (wrapper, defines into nil map, defines override existing, nil src). - Add dsl/merge_test.go driving mergeFiles to 100%: legacy build fields, duplicate toolchain/build/group from include, binary / sources / normalize_crlf / trim_trailing_ws propagation, local overrides of timeout and memory_limit. - Add dsl/parser_features_test.go for parseTest / parseGroup happy paths that were missing: file/outFile, env, wrapper, per-test timeout/memory overrides, non-zero exitCode, scoring partial / all_or_none and unknown scoring. - Add dsl/parser_errors_test.go, a 54-case table-driven test that hits every `expect(...)` error branch in parseGroup and parseTest (missing LPAREN/RPAREN/LBRACE/RBRACE/ASSIGN, wrong token types on weight/timeout/memory_limit/scoring/env/wrapper/file/outFile, and unclosed blocks). - Add dsl/parser_misc_test.go covering parsePattern dir-mode with args, unknown pattern field, non-ident in pattern, top-level binary / sources / normalize_crlf / trim_trailing_ws / bare-int memory_limit, parseBool invalid ident and non-ident, matcher without an operator, validateBuilds legacy+structured conflict. - Add dsl/build_parser_test.go covering every BuildConfig field (sources, includes, sanitize, link, extra, platforms, compilers, defines), OS overrides on named builds, nested / duplicate OS override errors, unknown build / profile / warnings / platform, missing = on assign-string and assign-string-list, define(...) error cases, and parseToolchainsBlock (duplicate name, missing platforms, bad name token, unknown field, unknown compiler class, binary and class propagation). - Add dsl/lexer_test.go for Token.String, TokenType.String UNKNOWN branch, line comments, unterminated string and heredoc, unknown escape sequence, escape decoding, unexpected character, every K/M/G/KiB/MiB/GiB size suffix, ms/s/m duration suffixes, negative integer lexing and float literals. - Extend runner/result_test.go with Status.String and TestResult.addFailure (both previously 0%). - Add .gitea/workflows/go-test.yml running `go vet` and `go test -race -coverprofile=coverage.out ./...` on push, pull_request and manual dispatch, uploading coverage.out as an artifact. Coverage: dsl 60.5% -> 85%+, runner 29.0% -> 30.5%.	2026-04-16 00:59:09 +03:00
Mikhail Kornilovich	5e0effc6fe	refactor: extract expander helpers and Test stdio/file setters - Add globWithAffixes in runner/expander.go that wraps filepath.Glob with an empty-match check and returns the computed prefix/suffix from splitGlob, collapsing three near-identical lookup blocks in expandGlobPattern into single calls. - Extract per-case Test construction from buildTests into a new buildTest helper so the loop body is a single call and the read / assemble / arg-template logic lives in one place. - Add Test.SetInputFile, Test.SetStdin, Test.SetOutputFile and Test.SetStdout methods on dsl.Test to encapsulate the stdin-vs- InFiles and stdout-vs-OutFiles wiring that buildTest previously did inline. - Adopt the `for range N` loop form in the determinism check in runner/compiler_test.go. - Switch the MSVC release test to expect /std:c17 since MSVC does not ship a c11 mode (worth surfacing a warning about this later).	2026-04-16 00:28:30 +03:00
Mikhail Kornilovich	c85c65ed49	refactor: modernize stdlib usage and move matchers into dsl All checks were successful build-dsl-smoke / Build judge (push) Successful in 16s Details build-dsl-smoke / debug / clang / linux (push) Successful in 6s Details build-dsl-smoke / debug / gcc / linux (push) Successful in 7s Details build-dsl-smoke / release / clang / linux (push) Successful in 8s Details build-dsl-smoke / release / gcc / linux (push) Successful in 9s Details build-dsl-smoke / sanitized / gcc / linux (push) Successful in 6s Details build-dsl-smoke / sanitized / clang / linux (push) Successful in 9s Details build-dsl-smoke / debug-valgrind / gcc / linux (push) Successful in 15s Details build-dsl-smoke / debug / clang / windows (push) Successful in 15s Details build-dsl-smoke / debug / msvc / windows (push) Successful in 19s Details build-dsl-smoke / release / clang / windows (push) Successful in 18s Details build-dsl-smoke / release / msvc / windows (push) Successful in 18s Details build-dsl-smoke / SUMMARY (push) Successful in 4s Details - Move Matcher types and matching logic from runner/matcher.go into the dsl package as methods on the Matcher types. Runner now calls t.Stdout.Match(label, actual) instead of type-switching via a package-level applyMatcher helper. - Replace custom contains/containsString helpers with slices.Contains in dsl/build.go and runner/compiler.go. - Use maps.Copy instead of manual map copy in BuildConfig.MergeFrom. - Adopt strings.SplitSeq, strings.CutPrefix and the `for range N` loop form in runner/limiter_linux.go. - Ignore example/imdb build artifact.	2026-04-15 21:24:11 +03:00
Mikhail Kornilovich	52202abb53	fix compile standard for msvc All checks were successful build-dsl-smoke / Build judge (push) Successful in 13s Details build-dsl-smoke / debug / clang / linux (push) Successful in 6s Details build-dsl-smoke / debug / gcc / linux (push) Successful in 9s Details build-dsl-smoke / release / clang / linux (push) Successful in 9s Details build-dsl-smoke / release / gcc / linux (push) Successful in 9s Details build-dsl-smoke / sanitized / clang / linux (push) Successful in 10s Details build-dsl-smoke / sanitized / gcc / linux (push) Successful in 9s Details build-dsl-smoke / debug / clang / windows (push) Successful in 18s Details build-dsl-smoke / debug-valgrind / gcc / linux (push) Successful in 18s Details build-dsl-smoke / debug / msvc / windows (push) Successful in 20s Details build-dsl-smoke / release / clang / windows (push) Successful in 16s Details build-dsl-smoke / release / msvc / windows (push) Successful in 17s Details build-dsl-smoke / SUMMARY (push) Successful in 4s Details Release / Build & publish (push) Successful in 49s Details	2026-04-15 11:02:05 +03:00
Mikhail Kornilovich	ea2ca49f0f	fix jdg All checks were successful build-dsl-smoke / Build judge (push) Successful in 13s Details build-dsl-smoke / debug / clang / linux (push) Successful in 6s Details build-dsl-smoke / debug / gcc / linux (push) Successful in 7s Details build-dsl-smoke / release / clang / linux (push) Successful in 10s Details build-dsl-smoke / release / gcc / linux (push) Successful in 9s Details build-dsl-smoke / sanitized / clang / linux (push) Successful in 9s Details build-dsl-smoke / sanitized / gcc / linux (push) Successful in 7s Details build-dsl-smoke / debug-valgrind / gcc / linux (push) Successful in 16s Details Release / Build & publish (push) Successful in 44s Details build-dsl-smoke / debug / clang / windows (push) Successful in 34s Details build-dsl-smoke / release / clang / windows (push) Successful in 31s Details build-dsl-smoke / debug / msvc / windows (push) Successful in 36s Details build-dsl-smoke / release / msvc / windows (push) Successful in 33s Details build-dsl-smoke / SUMMARY (push) Successful in 5s Details	2026-04-13 13:13:20 +03:00
Mikhail Kornilovich	7ec3a43c7a	1. New build system All checks were successful build-dsl-smoke / Build judge (push) Successful in 12s Details build-dsl-smoke / debug / clang / linux (push) Successful in 6s Details build-dsl-smoke / debug / gcc / linux (push) Successful in 8s Details build-dsl-smoke / release / clang / linux (push) Successful in 8s Details build-dsl-smoke / release / gcc / linux (push) Successful in 6s Details build-dsl-smoke / sanitized / clang / linux (push) Successful in 8s Details build-dsl-smoke / sanitized / gcc / linux (push) Successful in 7s Details build-dsl-smoke / debug / clang / windows (push) Successful in 13s Details build-dsl-smoke / debug-valgrind / gcc / linux (push) Successful in 14s Details build-dsl-smoke / release / clang / windows (push) Successful in 16s Details build-dsl-smoke / debug / msvc / windows (push) Successful in 18s Details build-dsl-smoke / release / msvc / windows (push) Successful in 17s Details build-dsl-smoke / SUMMARY (push) Successful in 4s Details Release / Build & publish (push) Successful in 48s Details Reviewed-on: #1	2026-04-12 07:59:38 +00:00