add new build system

runner/compiler.go

@@ -0,0 +1,198 @@
+package runner
+
+import (
+	"fmt"
+	"runtime"
+	"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 != "" {
+		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 containsString(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
+}
+
+func containsString(xs []string, x string) bool {
+	for _, v := range xs {
+		if v == x {
+			return true
+		}
+	}
+	return false
+}

runner/compiler_test.go

@@ -0,0 +1,293 @@
+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 i := 0; i < 20; i++ {
+		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")
+	want := []string{"cl", "/nologo", "/std:c11", "/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
+}

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
 }

runner/result_test.go

@@ -0,0 +1,57 @@
+package runner
+
+import "testing"
+
+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)
+	}
+}

runner/runner.go

@@ -21,6 +21,8 @@ type Config struct {
 	WorkDir    string
 	BinaryName string
 	Wrapper    string
+
+	TargetBuild string
 }
 
 type Runner struct {
@@ -63,41 +65,271 @@ 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
+}
+
+// runLegacyBuild handles the classic `build "shell-string"` form. It produces
+// a single BuildRun named "default" so that downstream consumers always see
+// the same shape.
+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
+// runStructuredBuilds handles the new DSL form with one or more named builds.
+// Each build is resolved against the current OS and toolchain, compiled via
+// the structured translator, and then exercised against every group.
+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 {
+			// User asked for a different build. Don't include this one
+			// in the result at all — discovery via --list-builds is the
+			// caller's responsibility.
+			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
+}
+
+// runGroups exercises every group/test in the suite against the currently
+// selected binary (r.binary) and records the outcome into run.
+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
+	}
+}
+
+// fillBuildError populates a BuildRun with one failing synthetic test per
+// group when the build itself failed. This keeps the reported totals at 0
+// and matches the legacy behaviour.
+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
+}
+
+// legacyBuild runs the free-form shell build command. Kept under the old
+// name so reviewers can diff against the previous implementation easily.
+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
+}
+
+// compileStructured compiles one structured build via the translator and
+// returns the build log plus the absolute path to the produced binary.
+// The compiler is invoked via exec.Command directly — no shell involved.
+func (r *Runner) compileStructured(name string, cfg dsl.BuildConfig, tc Toolchain) (string, string, error) {
+	// Expand source globs against the work dir.
+	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
+
+	// Decide output path: <workdir>/build/<name>/<output>[.exe].
+	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
+}
+
+// expandSources expands each glob in patterns against workDir and returns
+// a slice of paths relative to workDir. Globs that match no files cause an
+// error — silent zero matches are almost always a typo.
+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 {
+			// Fall back to treating it as a literal path.
+			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 +384,6 @@ 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)
@@ -392,21 +590,23 @@ func (r *Runner) runTest(t *dsl.Test) *TestResult {
 		tr.addFailure("%s", f)
 	}
 
-	for name, expected := range t.OutFiles {
-		path := filepath.Join(tmpDir, name)
-		content, err := os.ReadFile(path)
-		if err != nil {
-			tr.addFailure("output file %q not found: %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)
-		}
+	if len(tr.Failures) > 0 {
+		tr.Status = StatusFail
 	}
 
-	if tr.Status == StatusPass && len(tr.Failures) > 0 {
-		tr.Status = StatusFail
+	for name, expected := range t.OutFiles {
+		actualPath := filepath.Join(tmpDir, name)
+		data, err := os.ReadFile(actualPath)
+		if err != nil {
+			tr.Status = StatusFail
+			tr.addFailure("output file %q missing: %v", name, err)
+			continue
+		}
+		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)
+		}
 	}
 
 	return tr