wasm: add Boehm GC support

This adds support for `-gc=boehm` on `-target=wasip1` and `-target=wasm`
(in a browser or NodeJS). Notably it does *not* add Boehm GC support for
`-target=wasip2`, since that target doesn't have a real libc.

Author: aykevl

builder/bdwgc.go

@@ -14,7 +14,7 @@ var BoehmGC = Library{
 	name: "bdwgc",
 	cflags: func(target, headerPath string) []string {
 		libdir := filepath.Join(goenv.Get("TINYGOROOT"), "lib/bdwgc")
-		return []string{
+		flags := []string{
 			// use a modern environment
 			"-DUSE_MMAP",              // mmap is available
 			"-DUSE_MUNMAP",            // return memory to the OS using munmap
@@ -30,6 +30,7 @@ var BoehmGC = Library{
 			// Use a minimal environment.
 			"-DNO_MSGBOX_ON_ERROR", // don't call MessageBoxA on Windows
 			"-DDONT_USE_ATEXIT",
+			"-DNO_GETENV",
 
 			// Special flag to work around the lack of __data_start in ld.lld.
 			// TODO: try to fix this in LLVM/lld directly so we don't have to
@@ -49,17 +50,22 @@ var BoehmGC = Library{
 
 			"-I" + libdir + "/include",
 		}
+		if strings.HasPrefix(target, "wasm") {
+			// Save binary size on WebAssembly by disabling snprintf.
+			// This reduces the binary by around 18kB.
+			flags = append(flags, "-DGC_DISABLE_SNPRINTF=1")
+		}
+		return flags
 	},
 	needsLibc: true,
 	sourceDir: func() string {
 		return filepath.Join(goenv.Get("TINYGOROOT"), "lib/bdwgc")
 	},
-	librarySources: func(target string) ([]string, error) {
+	librarySources: func(target string, _ bool) ([]string, error) {
 		sources := []string{
 			"allchblk.c",
 			"alloc.c",
 			"blacklst.c",
-			"dbg_mlc.c",
 			"dyn_load.c",
 			"finalize.c",
 			"headers.c",

builder/builtins.go

@@ -226,7 +226,7 @@ var libCompilerRT = Library{
 		// Development build.
 		return filepath.Join(goenv.Get("TINYGOROOT"), "lib/compiler-rt-builtins")
 	},
-	librarySources: func(target string) ([]string, error) {
+	librarySources: func(target string, _ bool) ([]string, error) {
 		builtins := append([]string{}, genericBuiltins...) // copy genericBuiltins
 		switch compileopts.CanonicalArchName(target) {
 		case "arm":

builder/library.go

@@ -38,7 +38,7 @@ type Library struct {
 	sourceDir func() string
 
 	// The source files, relative to sourceDir.
-	librarySources func(target string) ([]string, error)
+	librarySources func(target string, libcNeedsMalloc bool) ([]string, error)
 
 	// The source code for the crt1.o file, relative to sourceDir.
 	crt1Source string
@@ -226,7 +226,7 @@ func (l *Library) load(config *compileopts.Config, tmpdir string) (job *compileJ
 
 	// Create jobs to compile all sources. These jobs are depended upon by the
 	// archive job above, so must be run first.
-	paths, err := l.librarySources(target)
+	paths, err := l.librarySources(target, config.LibcNeedsMalloc())
 	if err != nil {
 		return nil, nil, err
 	}

builder/mingw-w64.go

@@ -48,7 +48,7 @@ var libMinGW = Library{
 		}
 		return flags
 	},
-	librarySources: func(target string) ([]string, error) {
+	librarySources: func(target string, _ bool) ([]string, error) {
 		// These files are needed so that printf and the like are supported.
 		var sources []string
 		if strings.Split(target, "-")[0] == "i386" {

builder/musl.go

@@ -121,7 +121,7 @@ var libMusl = Library{
 		return cflags
 	},
 	sourceDir: func() string { return filepath.Join(goenv.Get("TINYGOROOT"), "lib/musl/src") },
-	librarySources: func(target string) ([]string, error) {
+	librarySources: func(target string, _ bool) ([]string, error) {
 		arch := compileopts.MuslArchitecture(target)
 		globs := []string{
 			"conf/*.c",

builder/picolibc.go

@@ -43,7 +43,7 @@ var libPicolibc = Library{
 		}
 	},
 	sourceDir: func() string { return filepath.Join(goenv.Get("TINYGOROOT"), "lib/picolibc/newlib") },
-	librarySources: func(target string) ([]string, error) {
+	librarySources: func(target string, _ bool) ([]string, error) {
 		sources := append([]string(nil), picolibcSources...)
 		if !strings.HasPrefix(target, "avr") {
 			// Small chips without long jumps can't compile many files (printf,

builder/wasilibc.go

@@ -120,7 +120,7 @@ var libWasiLibc = Library{
 		return nil
 	},
 	sourceDir: func() string { return filepath.Join(goenv.Get("TINYGOROOT"), "lib/wasi-libc") },
-	librarySources: func(target string) ([]string, error) {
+	librarySources: func(target string, libcNeedsMalloc bool) ([]string, error) {
 		type filePattern struct {
 			glob    string
 			exclude []string
@@ -169,6 +169,11 @@ var libWasiLibc = Library{
 			{glob: "libc-bottom-half/sources/*.c"},
 		}
 
+		// We're using the Boehm GC, so we need a heap implementation in the libc.
+		if libcNeedsMalloc {
+			globs = append(globs, filePattern{glob: "dlmalloc/src/dlmalloc.c"})
+		}
+
 		// See: LIBC_TOP_HALF_MUSL_SOURCES in the Makefile
 		sources := []string{
 			"libc-top-half/musl/src/misc/a64l.c",

builder/wasmbuiltins.go

@@ -41,7 +41,7 @@ var libWasmBuiltins = Library{
 		}
 	},
 	sourceDir: func() string { return filepath.Join(goenv.Get("TINYGOROOT"), "lib/wasi-libc") },
-	librarySources: func(target string) ([]string, error) {
+	librarySources: func(target string, _ bool) ([]string, error) {
 		return []string{
 			// memory builtins needed for llvm.memcpy.*, llvm.memmove.*, and
 			// llvm.memset.* LLVM intrinsics.

compileopts/config.go

@@ -133,7 +133,7 @@ func (c *Config) GC() string {
 // that can be traced by the garbage collector.
 func (c *Config) NeedsStackObjects() bool {
 	switch c.GC() {
-	case "conservative", "custom", "precise":
+	case "conservative", "custom", "precise", "boehm":
 		for _, tag := range c.BuildTags() {
 			if tag == "tinygo.wasm" {
 				return true
@@ -258,6 +258,15 @@ func MuslArchitecture(triple string) string {
 	return CanonicalArchName(triple)
 }
 
+// Returns true if the libc needs to include malloc, for the libcs where this
+// matters.
+func (c *Config) LibcNeedsMalloc() bool {
+	if c.GC() == "boehm" && c.Target.Libc == "wasi-libc" {
+		return true
+	}
+	return false
+}
+
 // LibraryPath returns the path to the library build directory. The path will be
 // a library path in the cache directory (which might not yet be built).
 func (c *Config) LibraryPath(name string) string {
@@ -281,9 +290,14 @@ func (c *Config) LibraryPath(name string) string {
 		archname += "-v" + strconv.Itoa(v)
 	}
 
+	options := ""
+	if c.LibcNeedsMalloc() {
+		options += "+malloc"
+	}
+
 	// No precompiled library found. Determine the path name that will be used
 	// in the build cache.
-	return filepath.Join(goenv.Get("GOCACHE"), name+"-"+archname)
+	return filepath.Join(goenv.Get("GOCACHE"), name+options+"-"+archname)
 }
 
 // DefaultBinaryExtension returns the default extension for binaries, such as

compileopts/target.go

@@ -266,6 +266,9 @@ func defaultTarget(options *Options) (*TargetSpec, error) {
 		DefaultStackSize: 1024 * 64, // 64kB
 		GDB:              []string{"gdb"},
 		PortReset:        "false",
+		ExtraFiles: []string{
+			"src/runtime/gc_boehm.c",
+		},
 	}
 
 	// Configure target based on GOARCH.