go/src/cmd/cgo/out.go

2054 lines
61 KiB
Go
Raw Normal View History

// Copyright 2009 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
package main
import (
"bytes"
"cmd/internal/pkgpath"
"debug/elf"
"debug/macho"
"debug/pe"
"fmt"
"go/ast"
"go/printer"
"go/token"
"internal/xcoff"
"io"
"os"
"os/exec"
"path/filepath"
"regexp"
"sort"
"strings"
"unicode"
)
var (
conf = printer.Config{Mode: printer.SourcePos, Tabwidth: 8}
noSourceConf = printer.Config{Tabwidth: 8}
)
// writeDefs creates output files to be compiled by gc and gcc.
func (p *Package) writeDefs() {
var fgo2, fc io.Writer
f := creat(*objDir + "_cgo_gotypes.go")
defer f.Close()
fgo2 = f
if *gccgo {
f := creat(*objDir + "_cgo_defun.c")
defer f.Close()
fc = f
}
fm := creat(*objDir + "_cgo_main.c")
var gccgoInit strings.Builder
if !*gccgo {
for _, arg := range p.LdFlags {
fmt.Fprintf(fgo2, "//go:cgo_ldflag %q\n", arg)
}
} else {
fflg := creat(*objDir + "_cgo_flags")
for _, arg := range p.LdFlags {
fmt.Fprintf(fflg, "_CGO_LDFLAGS=%s\n", arg)
}
fflg.Close()
}
// Write C main file for using gcc to resolve imports.
fmt.Fprintf(fm, "#include <stddef.h>\n") // For size_t below.
fmt.Fprintf(fm, "int main(int argc __attribute__((unused)), char **argv __attribute__((unused))) { return 0; }\n")
if *importRuntimeCgo {
fmt.Fprintf(fm, "void crosscall2(void(*fn)(void*) __attribute__((unused)), void *a __attribute__((unused)), int c __attribute__((unused)), size_t ctxt __attribute__((unused))) { }\n")
fmt.Fprintf(fm, "size_t _cgo_wait_runtime_init_done(void) { return 0; }\n")
fmt.Fprintf(fm, "void _cgo_release_context(size_t ctxt __attribute__((unused))) { }\n")
fmt.Fprintf(fm, "char* _cgo_topofstack(void) { return (char*)0; }\n")
} else {
// If we're not importing runtime/cgo, we *are* runtime/cgo,
// which provides these functions. We just need a prototype.
fmt.Fprintf(fm, "void crosscall2(void(*fn)(void*), void *a, int c, size_t ctxt);\n")
fmt.Fprintf(fm, "size_t _cgo_wait_runtime_init_done(void);\n")
fmt.Fprintf(fm, "void _cgo_release_context(size_t);\n")
}
fmt.Fprintf(fm, "void _cgo_allocate(void *a __attribute__((unused)), int c __attribute__((unused))) { }\n")
fmt.Fprintf(fm, "void _cgo_panic(void *a __attribute__((unused)), int c __attribute__((unused))) { }\n")
fmt.Fprintf(fm, "void _cgo_reginit(void) { }\n")
// Write second Go output: definitions of _C_xxx.
// In a separate file so that the import of "unsafe" does not
// pollute the original file.
fmt.Fprintf(fgo2, "// Code generated by cmd/cgo; DO NOT EDIT.\n\n")
fmt.Fprintf(fgo2, "package %s\n\n", p.PackageName)
fmt.Fprintf(fgo2, "import \"unsafe\"\n\n")
if *importSyscall {
fmt.Fprintf(fgo2, "import \"syscall\"\n\n")
}
if *importRuntimeCgo {
if !*gccgoDefineCgoIncomplete {
fmt.Fprintf(fgo2, "import _cgopackage \"runtime/cgo\"\n\n")
fmt.Fprintf(fgo2, "type _ _cgopackage.Incomplete\n") // prevent import-not-used error
} else {
fmt.Fprintf(fgo2, "//go:notinheap\n")
fmt.Fprintf(fgo2, "type _cgopackage_Incomplete struct{ _ struct{ _ struct{} } }\n")
}
}
if *importSyscall {
fmt.Fprintf(fgo2, "var _ syscall.Errno\n")
}
fmt.Fprintf(fgo2, "func _Cgo_ptr(ptr unsafe.Pointer) unsafe.Pointer { return ptr }\n\n")
if !*gccgo {
fmt.Fprintf(fgo2, "//go:linkname _Cgo_always_false runtime.cgoAlwaysFalse\n")
fmt.Fprintf(fgo2, "var _Cgo_always_false bool\n")
fmt.Fprintf(fgo2, "//go:linkname _Cgo_use runtime.cgoUse\n")
fmt.Fprintf(fgo2, "func _Cgo_use(interface{})\n")
fmt.Fprintf(fgo2, "//go:linkname _Cgo_keepalive runtime.cgoKeepAlive\n")
fmt.Fprintf(fgo2, "//go:noescape\n")
fmt.Fprintf(fgo2, "func _Cgo_keepalive(interface{})\n")
}
fmt.Fprintf(fgo2, "//go:linkname _Cgo_no_callback runtime.cgoNoCallback\n")
fmt.Fprintf(fgo2, "func _Cgo_no_callback(bool)\n")
typedefNames := make([]string, 0, len(typedef))
for name := range typedef {
if name == "_Ctype_void" {
// We provide an appropriate declaration for
// _Ctype_void below (#39877).
continue
}
typedefNames = append(typedefNames, name)
}
sort.Strings(typedefNames)
for _, name := range typedefNames {
def := typedef[name]
fmt.Fprintf(fgo2, "type %s ", name)
// We don't have source info for these types, so write them out without source info.
// Otherwise types would look like:
//
// type _Ctype_struct_cb struct {
// //line :1
// on_test *[0]byte
// //line :1
// }
//
// Which is not useful. Moreover we never override source info,
// so subsequent source code uses the same source info.
// Moreover, empty file name makes compile emit no source debug info at all.
var buf bytes.Buffer
noSourceConf.Fprint(&buf, fset, def.Go)
cmd/cgo: fix mangling of enum and union types Consider this test package: package p // enum E { E0 }; // union U { long x; }; // void f(enum E e, union U* up) {} import "C" func f() { C.f(C.enum_E(C.E0), (*C.union_U)(nil)) } In Go 1.14, cgo translated this to (omitting irrelevant details): type _Ctype_union_U [8]byte func f() { _Cfunc_f(uint32(_Ciconst_E0), (*[8]byte)(nil)) } func _Cfunc_f(p0 uint32, p1 *[8]byte) (r1 _Ctype_void) { ... } Notably, _Ctype_union_U was declared as a defined type, but uses were being rewritten into uses of the underlying type, which matched how _Cfunc_f was declared. After CL 230037, cgo started consistently rewriting "C.foo" type expressions as "_Ctype_foo", which caused it to start emitting: type _Ctype_enum_E uint32 type _Ctype_union_U [8]byte func f() { _Cfunc_f(_Ctype_enum_E(_Ciconst_E0), (*_Ctype_union_U)(nil)) } // _Cfunc_f unchanged Of course, this fails to type-check because _Ctype_enum_E and _Ctype_union_U are defined types. This CL changes cgo to emit: type _Ctype_enum_E = uint32 type _Ctype_union_U = [8]byte // f unchanged since CL 230037 // _Cfunc_f still unchanged It would probably be better to fix this in (*typeConv).loadType so that cgo generated code uses the _Ctype_foo aliases too. But as it wouldn't have any effect on actual compilation, it's not worth the risk of touching it at this point in the release cycle. Updates #39537. Fixes #40494. Change-Id: I88269660b40aeda80a9a9433777601a781b48ac0 Reviewed-on: https://go-review.googlesource.com/c/go/+/246057 Reviewed-by: Ian Lance Taylor <iant@golang.org>
2020-07-30 18:35:00 -07:00
if bytes.HasPrefix(buf.Bytes(), []byte("_Ctype_")) ||
strings.HasPrefix(name, "_Ctype_enum_") ||
strings.HasPrefix(name, "_Ctype_union_") {
// This typedef is of the form `typedef a b` and should be an alias.
fmt.Fprintf(fgo2, "= ")
}
fmt.Fprintf(fgo2, "%s", buf.Bytes())
fmt.Fprintf(fgo2, "\n\n")
}
if *gccgo {
fmt.Fprintf(fgo2, "type _Ctype_void byte\n")
} else {
fmt.Fprintf(fgo2, "type _Ctype_void [0]byte\n")
}
if *gccgo {
fmt.Fprint(fgo2, gccgoGoProlog)
fmt.Fprint(fc, p.cPrologGccgo())
} else {
fmt.Fprint(fgo2, goProlog)
}
if fc != nil {
fmt.Fprintf(fc, "#line 1 \"cgo-generated-wrappers\"\n")
}
if fm != nil {
fmt.Fprintf(fm, "#line 1 \"cgo-generated-wrappers\"\n")
}
gccgoSymbolPrefix := p.gccgoSymbolPrefix()
cVars := make(map[string]bool)
for _, key := range nameKeys(p.Name) {
n := p.Name[key]
if !n.IsVar() {
continue
}
if !cVars[n.C] {
if *gccgo {
fmt.Fprintf(fc, "extern byte *%s;\n", n.C)
} else {
// Force a reference to all symbols so that
// the external linker will add DT_NEEDED
// entries as needed on ELF systems.
// Treat function variables differently
// to avoid type conflict errors from LTO
// (Link Time Optimization).
if n.Kind == "fpvar" {
fmt.Fprintf(fm, "extern void %s();\n", n.C)
} else {
fmt.Fprintf(fm, "extern char %s[];\n", n.C)
fmt.Fprintf(fm, "void *_cgohack_%s = %s;\n\n", n.C, n.C)
}
fmt.Fprintf(fgo2, "//go:linkname __cgo_%s %s\n", n.C, n.C)
fmt.Fprintf(fgo2, "//go:cgo_import_static %s\n", n.C)
fmt.Fprintf(fgo2, "var __cgo_%s byte\n", n.C)
}
cVars[n.C] = true
}
var node ast.Node
if n.Kind == "var" {
node = &ast.StarExpr{X: n.Type.Go}
} else if n.Kind == "fpvar" {
node = n.Type.Go
} else {
panic(fmt.Errorf("invalid var kind %q", n.Kind))
}
if *gccgo {
fmt.Fprintf(fc, `extern void *%s __asm__("%s.%s");`, n.Mangle, gccgoSymbolPrefix, gccgoToSymbol(n.Mangle))
fmt.Fprintf(&gccgoInit, "\t%s = &%s;\n", n.Mangle, n.C)
fmt.Fprintf(fc, "\n")
}
fmt.Fprintf(fgo2, "var %s ", n.Mangle)
conf.Fprint(fgo2, fset, node)
if !*gccgo {
fmt.Fprintf(fgo2, " = (")
conf.Fprint(fgo2, fset, node)
fmt.Fprintf(fgo2, ")(unsafe.Pointer(&__cgo_%s))", n.C)
}
fmt.Fprintf(fgo2, "\n")
}
if *gccgo {
fmt.Fprintf(fc, "\n")
}
for _, key := range nameKeys(p.Name) {
n := p.Name[key]
if n.Const != "" {
fmt.Fprintf(fgo2, "const %s = %s\n", n.Mangle, n.Const)
}
}
fmt.Fprintf(fgo2, "\n")
callsMalloc := false
for _, key := range nameKeys(p.Name) {
n := p.Name[key]
if n.FuncType != nil {
p.writeDefsFunc(fgo2, n, &callsMalloc)
}
}
fgcc := creat(*objDir + "_cgo_export.c")
fgcch := creat(*objDir + "_cgo_export.h")
if *gccgo {
p.writeGccgoExports(fgo2, fm, fgcc, fgcch)
} else {
p.writeExports(fgo2, fm, fgcc, fgcch)
}
if callsMalloc && !*gccgo {
fmt.Fprint(fgo2, strings.Replace(cMallocDefGo, "PREFIX", cPrefix, -1))
fmt.Fprint(fgcc, strings.Replace(strings.Replace(cMallocDefC, "PREFIX", cPrefix, -1), "PACKED", p.packedAttribute(), -1))
}
if err := fgcc.Close(); err != nil {
fatalf("%s", err)
}
if err := fgcch.Close(); err != nil {
fatalf("%s", err)
}
if *exportHeader != "" && len(p.ExpFunc) > 0 {
fexp := creat(*exportHeader)
fgcch, err := os.Open(*objDir + "_cgo_export.h")
if err != nil {
fatalf("%s", err)
}
defer fgcch.Close()
_, err = io.Copy(fexp, fgcch)
if err != nil {
fatalf("%s", err)
}
if err = fexp.Close(); err != nil {
fatalf("%s", err)
}
}
init := gccgoInit.String()
if init != "" {
// The init function does nothing but simple
// assignments, so it won't use much stack space, so
// it's OK to not split the stack. Splitting the stack
// can run into a bug in clang (as of 2018-11-09):
// this is a leaf function, and when clang sees a leaf
// function it won't emit the split stack prologue for
// the function. However, if this function refers to a
// non-split-stack function, which will happen if the
// cgo code refers to a C function not compiled with
// -fsplit-stack, then the linker will think that it
// needs to adjust the split stack prologue, but there
// won't be one. Marking the function explicitly
// no_split_stack works around this problem by telling
// the linker that it's OK if there is no split stack
// prologue.
fmt.Fprintln(fc, "static void init(void) __attribute__ ((constructor, no_split_stack));")
fmt.Fprintln(fc, "static void init(void) {")
fmt.Fprint(fc, init)
fmt.Fprintln(fc, "}")
}
}
// elfImportedSymbols is like elf.File.ImportedSymbols, but it
// includes weak symbols.
//
// A bug in some versions of LLD (at least LLD 8) cause it to emit
// several pthreads symbols as weak, but we need to import those. See
// issue #31912 or https://bugs.llvm.org/show_bug.cgi?id=42442.
//
// When doing external linking, we hand everything off to the external
// linker, which will create its own dynamic symbol tables. For
// internal linking, this may turn weak imports into strong imports,
// which could cause dynamic linking to fail if a symbol really isn't
// defined. However, the standard library depends on everything it
// imports, and this is the primary use of dynamic symbol tables with
// internal linking.
func elfImportedSymbols(f *elf.File) []elf.ImportedSymbol {
syms, _ := f.DynamicSymbols()
var imports []elf.ImportedSymbol
for _, s := range syms {
if (elf.ST_BIND(s.Info) == elf.STB_GLOBAL || elf.ST_BIND(s.Info) == elf.STB_WEAK) && s.Section == elf.SHN_UNDEF {
imports = append(imports, elf.ImportedSymbol{
Name: s.Name,
Library: s.Library,
Version: s.Version,
})
}
}
return imports
}
func dynimport(obj string) {
stdout := os.Stdout
if *dynout != "" {
f, err := os.Create(*dynout)
if err != nil {
fatalf("%s", err)
}
defer func() {
if err = f.Close(); err != nil {
fatalf("error closing %s: %v", *dynout, err)
}
}()
stdout = f
}
fmt.Fprintf(stdout, "package %s\n", *dynpackage)
if f, err := elf.Open(obj); err == nil {
defer f.Close()
if *dynlinker {
// Emit the cgo_dynamic_linker line.
if sec := f.Section(".interp"); sec != nil {
if data, err := sec.Data(); err == nil && len(data) > 1 {
// skip trailing \0 in data
fmt.Fprintf(stdout, "//go:cgo_dynamic_linker %q\n", string(data[:len(data)-1]))
}
}
}
sym := elfImportedSymbols(f)
for _, s := range sym {
targ := s.Name
if s.Version != "" {
targ += "#" + s.Version
}
2020-11-06 09:38:38 -08:00
checkImportSymName(s.Name)
checkImportSymName(targ)
fmt.Fprintf(stdout, "//go:cgo_import_dynamic %s %s %q\n", s.Name, targ, s.Library)
}
lib, _ := f.ImportedLibraries()
for _, l := range lib {
fmt.Fprintf(stdout, "//go:cgo_import_dynamic _ _ %q\n", l)
}
return
}
if f, err := macho.Open(obj); err == nil {
defer f.Close()
sym, _ := f.ImportedSymbols()
for _, s := range sym {
s = strings.TrimPrefix(s, "_")
2020-11-06 09:38:38 -08:00
checkImportSymName(s)
fmt.Fprintf(stdout, "//go:cgo_import_dynamic %s %s %q\n", s, s, "")
}
lib, _ := f.ImportedLibraries()
for _, l := range lib {
fmt.Fprintf(stdout, "//go:cgo_import_dynamic _ _ %q\n", l)
}
return
}
if f, err := pe.Open(obj); err == nil {
defer f.Close()
sym, _ := f.ImportedSymbols()
for _, s := range sym {
ss := strings.Split(s, ":")
name := strings.Split(ss[0], "@")[0]
2020-11-06 09:38:38 -08:00
checkImportSymName(name)
checkImportSymName(ss[0])
fmt.Fprintf(stdout, "//go:cgo_import_dynamic %s %s %q\n", name, ss[0], strings.ToLower(ss[1]))
}
return
}
if f, err := xcoff.Open(obj); err == nil {
defer f.Close()
sym, err := f.ImportedSymbols()
if err != nil {
fatalf("cannot load imported symbols from XCOFF file %s: %v", obj, err)
}
for _, s := range sym {
if s.Name == "runtime_rt0_go" || s.Name == "_rt0_ppc64_aix_lib" {
// These symbols are imported by runtime/cgo but
// must not be added to _cgo_import.go as there are
// Go symbols.
continue
}
2020-11-06 09:38:38 -08:00
checkImportSymName(s.Name)
fmt.Fprintf(stdout, "//go:cgo_import_dynamic %s %s %q\n", s.Name, s.Name, s.Library)
}
lib, err := f.ImportedLibraries()
if err != nil {
fatalf("cannot load imported libraries from XCOFF file %s: %v", obj, err)
}
for _, l := range lib {
fmt.Fprintf(stdout, "//go:cgo_import_dynamic _ _ %q\n", l)
}
return
}
fatalf("cannot parse %s as ELF, Mach-O, PE or XCOFF", obj)
}
2020-11-06 09:38:38 -08:00
// checkImportSymName checks a symbol name we are going to emit as part
// of a //go:cgo_import_dynamic pragma. These names come from object
// files, so they may be corrupt. We are going to emit them unquoted,
// so while they don't need to be valid symbol names (and in some cases,
// involving symbol versions, they won't be) they must contain only
// graphic characters and must not contain Go comments.
func checkImportSymName(s string) {
for _, c := range s {
if !unicode.IsGraphic(c) || unicode.IsSpace(c) {
fatalf("dynamic symbol %q contains unsupported character", s)
}
}
if strings.Contains(s, "//") || strings.Contains(s, "/*") {
fatalf("dynamic symbol %q contains Go comment", s)
2020-11-06 09:38:38 -08:00
}
}
// Construct a gcc struct matching the gc argument frame.
// Assumes that in gcc, char is 1 byte, short 2 bytes, int 4 bytes, long long 8 bytes.
// These assumptions are checked by the gccProlog.
// Also assumes that gc convention is to word-align the
// input and output parameters.
func (p *Package) structType(n *Name) (string, int64) {
var buf strings.Builder
fmt.Fprint(&buf, "struct {\n")
off := int64(0)
for i, t := range n.FuncType.Params {
if off%t.Align != 0 {
pad := t.Align - off%t.Align
fmt.Fprintf(&buf, "\t\tchar __pad%d[%d];\n", off, pad)
off += pad
}
c := t.Typedef
if c == "" {
c = t.C.String()
}
fmt.Fprintf(&buf, "\t\t%s p%d;\n", c, i)
off += t.Size
}
if off%p.PtrSize != 0 {
pad := p.PtrSize - off%p.PtrSize
fmt.Fprintf(&buf, "\t\tchar __pad%d[%d];\n", off, pad)
off += pad
}
if t := n.FuncType.Result; t != nil {
if off%t.Align != 0 {
pad := t.Align - off%t.Align
fmt.Fprintf(&buf, "\t\tchar __pad%d[%d];\n", off, pad)
off += pad
}
fmt.Fprintf(&buf, "\t\t%s r;\n", t.C)
off += t.Size
}
if off%p.PtrSize != 0 {
pad := p.PtrSize - off%p.PtrSize
fmt.Fprintf(&buf, "\t\tchar __pad%d[%d];\n", off, pad)
off += pad
}
if off == 0 {
fmt.Fprintf(&buf, "\t\tchar unused;\n") // avoid empty struct
}
fmt.Fprintf(&buf, "\t}")
return buf.String(), off
}
func (p *Package) writeDefsFunc(fgo2 io.Writer, n *Name, callsMalloc *bool) {
name := n.Go
gtype := n.FuncType.Go
void := gtype.Results == nil || len(gtype.Results.List) == 0
if n.AddError {
// Add "error" to return type list.
// Type list is known to be 0 or 1 element - it's a C function.
err := &ast.Field{Type: ast.NewIdent("error")}
l := gtype.Results.List
if len(l) == 0 {
l = []*ast.Field{err}
} else {
l = []*ast.Field{l[0], err}
}
t := new(ast.FuncType)
*t = *gtype
t.Results = &ast.FieldList{List: l}
gtype = t
}
// Go func declaration.
d := &ast.FuncDecl{
Name: ast.NewIdent(n.Mangle),
Type: gtype,
}
// Builtins defined in the C prolog.
inProlog := builtinDefs[name] != ""
cname := fmt.Sprintf("_cgo%s%s", cPrefix, n.Mangle)
paramnames := []string(nil)
if d.Type.Params != nil {
for i, param := range d.Type.Params.List {
paramName := fmt.Sprintf("p%d", i)
param.Names = []*ast.Ident{ast.NewIdent(paramName)}
paramnames = append(paramnames, paramName)
}
}
if *gccgo {
// Gccgo style hooks.
fmt.Fprint(fgo2, "\n")
conf.Fprint(fgo2, fset, d)
fmt.Fprint(fgo2, " {\n")
if !inProlog {
fmt.Fprint(fgo2, "\tdefer syscall.CgocallDone()\n")
fmt.Fprint(fgo2, "\tsyscall.Cgocall()\n")
}
if n.AddError {
fmt.Fprint(fgo2, "\tsyscall.SetErrno(0)\n")
}
fmt.Fprint(fgo2, "\t")
if !void {
fmt.Fprint(fgo2, "r := ")
}
fmt.Fprintf(fgo2, "%s(%s)\n", cname, strings.Join(paramnames, ", "))
if n.AddError {
fmt.Fprint(fgo2, "\te := syscall.GetErrno()\n")
fmt.Fprint(fgo2, "\tif e != 0 {\n")
fmt.Fprint(fgo2, "\t\treturn ")
if !void {
fmt.Fprint(fgo2, "r, ")
}
fmt.Fprint(fgo2, "e\n")
fmt.Fprint(fgo2, "\t}\n")
fmt.Fprint(fgo2, "\treturn ")
if !void {
fmt.Fprint(fgo2, "r, ")
}
fmt.Fprint(fgo2, "nil\n")
} else if !void {
fmt.Fprint(fgo2, "\treturn r\n")
}
fmt.Fprint(fgo2, "}\n")
// declare the C function.
fmt.Fprintf(fgo2, "//extern %s\n", cname)
d.Name = ast.NewIdent(cname)
if n.AddError {
l := d.Type.Results.List
d.Type.Results.List = l[:len(l)-1]
}
conf.Fprint(fgo2, fset, d)
fmt.Fprint(fgo2, "\n")
return
}
if inProlog {
fmt.Fprint(fgo2, builtinDefs[name])
if strings.Contains(builtinDefs[name], "_cgo_cmalloc") {
*callsMalloc = true
}
return
}
// Wrapper calls into gcc, passing a pointer to the argument frame.
fmt.Fprintf(fgo2, "//go:cgo_import_static %s\n", cname)
fmt.Fprintf(fgo2, "//go:linkname __cgofn_%s %s\n", cname, cname)
fmt.Fprintf(fgo2, "var __cgofn_%s byte\n", cname)
fmt.Fprintf(fgo2, "var %s = unsafe.Pointer(&__cgofn_%s)\n", cname, cname)
nret := 0
if !void {
d.Type.Results.List[0].Names = []*ast.Ident{ast.NewIdent("r1")}
nret = 1
}
if n.AddError {
d.Type.Results.List[nret].Names = []*ast.Ident{ast.NewIdent("r2")}
}
fmt.Fprint(fgo2, "\n")
fmt.Fprint(fgo2, "//go:cgo_unsafe_args\n")
conf.Fprint(fgo2, fset, d)
fmt.Fprint(fgo2, " {\n")
// NOTE: Using uintptr to hide from escape analysis.
arg := "0"
if len(paramnames) > 0 {
arg = "uintptr(unsafe.Pointer(&p0))"
} else if !void {
arg = "uintptr(unsafe.Pointer(&r1))"
}
noCallback := p.noCallbacks[n.C]
if noCallback {
// disable cgocallback, will check it in runtime.
fmt.Fprintf(fgo2, "\t_Cgo_no_callback(true)\n")
}
prefix := ""
if n.AddError {
prefix = "errno := "
}
fmt.Fprintf(fgo2, "\t%s_cgo_runtime_cgocall(%s, %s)\n", prefix, cname, arg)
if n.AddError {
fmt.Fprintf(fgo2, "\tif errno != 0 { r2 = syscall.Errno(errno) }\n")
}
if noCallback {
fmt.Fprintf(fgo2, "\t_Cgo_no_callback(false)\n")
}
// Use _Cgo_keepalive instead of _Cgo_use when noescape & nocallback exist,
// so that the compiler won't force to escape them to heap.
// Instead, make the compiler keep them alive by using _Cgo_keepalive.
touchFunc := "_Cgo_use"
if p.noEscapes[n.C] && p.noCallbacks[n.C] {
touchFunc = "_Cgo_keepalive"
}
fmt.Fprintf(fgo2, "\tif _Cgo_always_false {\n")
if d.Type.Params != nil {
for _, name := range paramnames {
fmt.Fprintf(fgo2, "\t\t%s(%s)\n", touchFunc, name)
}
}
fmt.Fprintf(fgo2, "\t}\n")
fmt.Fprintf(fgo2, "\treturn\n")
fmt.Fprintf(fgo2, "}\n")
}
// writeOutput creates stubs for a specific source file to be compiled by gc
func (p *Package) writeOutput(f *File, srcfile string) {
base := srcfile
base = strings.TrimSuffix(base, ".go")
base = filepath.Base(base)
fgo1 := creat(*objDir + base + ".cgo1.go")
fgcc := creat(*objDir + base + ".cgo2.c")
p.GoFiles = append(p.GoFiles, base+".cgo1.go")
p.GccFiles = append(p.GccFiles, base+".cgo2.c")
// Write Go output: Go input with rewrites of C.xxx to _C_xxx.
fmt.Fprintf(fgo1, "// Code generated by cmd/cgo; DO NOT EDIT.\n\n")
if strings.ContainsAny(srcfile, "\r\n") {
// This should have been checked when the file path was first resolved,
// but we double check here just to be sure.
fatalf("internal error: writeOutput: srcfile contains unexpected newline character: %q", srcfile)
}
fmt.Fprintf(fgo1, "//line %s:1:1\n", srcfile)
fgo1.Write(f.Edit.Bytes())
// While we process the vars and funcs, also write gcc output.
// Gcc output starts with the preamble.
fmt.Fprintf(fgcc, "%s\n", builtinProlog)
fmt.Fprintf(fgcc, "%s\n", f.Preamble)
fmt.Fprintf(fgcc, "%s\n", gccProlog)
fmt.Fprintf(fgcc, "%s\n", tsanProlog)
fmt.Fprintf(fgcc, "%s\n", msanProlog)
for _, key := range nameKeys(f.Name) {
n := f.Name[key]
if n.FuncType != nil {
p.writeOutputFunc(fgcc, n)
}
}
fgo1.Close()
fgcc.Close()
}
// fixGo converts the internal Name.Go field into the name we should show
// to users in error messages. There's only one for now: on input we rewrite
// C.malloc into C._CMalloc, so change it back here.
func fixGo(name string) string {
if name == "_CMalloc" {
return "malloc"
}
return name
}
var isBuiltin = map[string]bool{
"_Cfunc_CString": true,
"_Cfunc_CBytes": true,
"_Cfunc_GoString": true,
"_Cfunc_GoStringN": true,
"_Cfunc_GoBytes": true,
"_Cfunc__CMalloc": true,
}
func (p *Package) writeOutputFunc(fgcc *os.File, n *Name) {
name := n.Mangle
if isBuiltin[name] || p.Written[name] {
// The builtins are already defined in the C prolog, and we don't
// want to duplicate function definitions we've already done.
return
}
p.Written[name] = true
if *gccgo {
p.writeGccgoOutputFunc(fgcc, n)
return
}
ctype, _ := p.structType(n)
// Gcc wrapper unpacks the C argument struct
// and calls the actual C function.
fmt.Fprintf(fgcc, "CGO_NO_SANITIZE_THREAD\n")
if n.AddError {
fmt.Fprintf(fgcc, "int\n")
} else {
fmt.Fprintf(fgcc, "void\n")
}
fmt.Fprintf(fgcc, "_cgo%s%s(void *v)\n", cPrefix, n.Mangle)
fmt.Fprintf(fgcc, "{\n")
if n.AddError {
fmt.Fprintf(fgcc, "\tint _cgo_errno;\n")
}
// We're trying to write a gcc struct that matches gc's layout.
// Use packed attribute to force no padding in this struct in case
// gcc has different packing requirements.
fmt.Fprintf(fgcc, "\t%s %v *_cgo_a = v;\n", ctype, p.packedAttribute())
if n.FuncType.Result != nil {
// Save the stack top for use below.
fmt.Fprintf(fgcc, "\tchar *_cgo_stktop = _cgo_topofstack();\n")
}
tr := n.FuncType.Result
if tr != nil {
fmt.Fprintf(fgcc, "\t__typeof__(_cgo_a->r) _cgo_r;\n")
}
fmt.Fprintf(fgcc, "\t_cgo_tsan_acquire();\n")
if n.AddError {
fmt.Fprintf(fgcc, "\terrno = 0;\n")
}
fmt.Fprintf(fgcc, "\t")
if tr != nil {
fmt.Fprintf(fgcc, "_cgo_r = ")
if c := tr.C.String(); c[len(c)-1] == '*' {
fmt.Fprint(fgcc, "(__typeof__(_cgo_a->r)) ")
}
}
if n.Kind == "macro" {
fmt.Fprintf(fgcc, "%s;\n", n.C)
} else {
fmt.Fprintf(fgcc, "%s(", n.C)
for i := range n.FuncType.Params {
if i > 0 {
fmt.Fprintf(fgcc, ", ")
}
fmt.Fprintf(fgcc, "_cgo_a->p%d", i)
}
fmt.Fprintf(fgcc, ");\n")
}
if n.AddError {
fmt.Fprintf(fgcc, "\t_cgo_errno = errno;\n")
}
fmt.Fprintf(fgcc, "\t_cgo_tsan_release();\n")
if n.FuncType.Result != nil {
// The cgo call may have caused a stack copy (via a callback).
// Adjust the return value pointer appropriately.
fmt.Fprintf(fgcc, "\t_cgo_a = (void*)((char*)_cgo_a + (_cgo_topofstack() - _cgo_stktop));\n")
// Save the return value.
fmt.Fprintf(fgcc, "\t_cgo_a->r = _cgo_r;\n")
// The return value is on the Go stack. If we are using msan,
// and if the C value is partially or completely uninitialized,
// the assignment will mark the Go stack as uninitialized.
// The Go compiler does not update msan for changes to the
// stack. It is possible that the stack will remain
// uninitialized, and then later be used in a way that is
// visible to msan, possibly leading to a false positive.
// Mark the stack space as written, to avoid this problem.
// See issue 26209.
fmt.Fprintf(fgcc, "\t_cgo_msan_write(&_cgo_a->r, sizeof(_cgo_a->r));\n")
}
if n.AddError {
fmt.Fprintf(fgcc, "\treturn _cgo_errno;\n")
}
fmt.Fprintf(fgcc, "}\n")
fmt.Fprintf(fgcc, "\n")
}
// Write out a wrapper for a function when using gccgo. This is a
// simple wrapper that just calls the real function. We only need a
// wrapper to support static functions in the prologue--without a
// wrapper, we can't refer to the function, since the reference is in
// a different file.
func (p *Package) writeGccgoOutputFunc(fgcc *os.File, n *Name) {
fmt.Fprintf(fgcc, "CGO_NO_SANITIZE_THREAD\n")
if t := n.FuncType.Result; t != nil {
fmt.Fprintf(fgcc, "%s\n", t.C.String())
} else {
fmt.Fprintf(fgcc, "void\n")
}
fmt.Fprintf(fgcc, "_cgo%s%s(", cPrefix, n.Mangle)
for i, t := range n.FuncType.Params {
if i > 0 {
fmt.Fprintf(fgcc, ", ")
}
c := t.Typedef
if c == "" {
c = t.C.String()
}
fmt.Fprintf(fgcc, "%s p%d", c, i)
}
fmt.Fprintf(fgcc, ")\n")
fmt.Fprintf(fgcc, "{\n")
if t := n.FuncType.Result; t != nil {
fmt.Fprintf(fgcc, "\t%s _cgo_r;\n", t.C.String())
}
fmt.Fprintf(fgcc, "\t_cgo_tsan_acquire();\n")
fmt.Fprintf(fgcc, "\t")
if t := n.FuncType.Result; t != nil {
fmt.Fprintf(fgcc, "_cgo_r = ")
// Cast to void* to avoid warnings due to omitted qualifiers.
if c := t.C.String(); c[len(c)-1] == '*' {
fmt.Fprintf(fgcc, "(void*)")
}
}
if n.Kind == "macro" {
fmt.Fprintf(fgcc, "%s;\n", n.C)
} else {
fmt.Fprintf(fgcc, "%s(", n.C)
for i := range n.FuncType.Params {
if i > 0 {
fmt.Fprintf(fgcc, ", ")
}
fmt.Fprintf(fgcc, "p%d", i)
}
fmt.Fprintf(fgcc, ");\n")
}
fmt.Fprintf(fgcc, "\t_cgo_tsan_release();\n")
if t := n.FuncType.Result; t != nil {
fmt.Fprintf(fgcc, "\treturn ")
// Cast to void* to avoid warnings due to omitted qualifiers
// and explicit incompatible struct types.
if c := t.C.String(); c[len(c)-1] == '*' {
fmt.Fprintf(fgcc, "(void*)")
}
fmt.Fprintf(fgcc, "_cgo_r;\n")
}
fmt.Fprintf(fgcc, "}\n")
fmt.Fprintf(fgcc, "\n")
}
// packedAttribute returns host compiler struct attribute that will be
// used to match gc's struct layout. For example, on 386 Windows,
// gcc wants to 8-align int64s, but gc does not.
// Use __gcc_struct__ to work around https://gcc.gnu.org/PR52991 on x86,
// and https://golang.org/issue/5603.
func (p *Package) packedAttribute() string {
s := "__attribute__((__packed__"
if !p.GccIsClang && (goarch == "amd64" || goarch == "386") {
s += ", __gcc_struct__"
}
return s + "))"
}
// exportParamName returns the value of param as it should be
// displayed in a c header file. If param contains any non-ASCII
// characters, this function will return the character p followed by
// the value of position; otherwise, this function will return the
// value of param.
func exportParamName(param string, position int) string {
if param == "" {
return fmt.Sprintf("p%d", position)
}
pname := param
for i := 0; i < len(param); i++ {
if param[i] > unicode.MaxASCII {
pname = fmt.Sprintf("p%d", position)
break
}
}
return pname
}
// Write out the various stubs we need to support functions exported
// from Go so that they are callable from C.
func (p *Package) writeExports(fgo2, fm, fgcc, fgcch io.Writer) {
p.writeExportHeader(fgcch)
fmt.Fprintf(fgcc, "/* Code generated by cmd/cgo; DO NOT EDIT. */\n\n")
fmt.Fprintf(fgcc, "#include <stdlib.h>\n")
fmt.Fprintf(fgcc, "#include \"_cgo_export.h\"\n\n")
// We use packed structs, but they are always aligned.
// The pragmas and address-of-packed-member are only recognized as
// warning groups in clang 4.0+, so ignore unknown pragmas first.
fmt.Fprintf(fgcc, "#pragma GCC diagnostic ignored \"-Wunknown-pragmas\"\n")
fmt.Fprintf(fgcc, "#pragma GCC diagnostic ignored \"-Wpragmas\"\n")
fmt.Fprintf(fgcc, "#pragma GCC diagnostic ignored \"-Waddress-of-packed-member\"\n")
fmt.Fprintf(fgcc, "#pragma GCC diagnostic ignored \"-Wunknown-warning-option\"\n")
fmt.Fprintf(fgcc, "#pragma GCC diagnostic ignored \"-Wunaligned-access\"\n")
fmt.Fprintf(fgcc, "extern void crosscall2(void (*fn)(void *), void *, int, size_t);\n")
fmt.Fprintf(fgcc, "extern size_t _cgo_wait_runtime_init_done(void);\n")
fmt.Fprintf(fgcc, "extern void _cgo_release_context(size_t);\n\n")
fmt.Fprintf(fgcc, "extern char* _cgo_topofstack(void);")
fmt.Fprintf(fgcc, "%s\n", tsanProlog)
fmt.Fprintf(fgcc, "%s\n", msanProlog)
for _, exp := range p.ExpFunc {
fn := exp.Func
runtime,cmd/cgo: simplify C -> Go call path This redesigns the way calls work from C to exported Go functions. It removes several steps from the call path, makes cmd/cgo no longer sensitive to the Go calling convention, and eliminates the use of reflectcall from cgo. In order to avoid generating a large amount of FFI glue between the C and Go ABIs, the cgo tool has long depended on generating a C function that marshals the arguments into a struct, and then the actual ABI switch happens in functions with fixed signatures that simply take a pointer to this struct. In a way, this CL simply pushes this idea further. Currently, the cgo tool generates this argument struct in the exact layout of the Go stack frame and depends on reflectcall to unpack it into the appropriate Go call (even though it's actually reflectcall'ing a function generated by cgo). In this CL, we decouple this struct from the Go stack layout. Instead, cgo generates a Go function that takes the struct, unpacks it, and calls the exported function. Since this generated function has a generic signature (like the rest of the call path), we don't need reflectcall and can instead depend on the Go compiler itself to implement the call to the exported Go function. One complication is that syscall.NewCallback on Windows, which converts a Go function into a C function pointer, depends on cgocallback's current dynamic calling approach since the signatures of the callbacks aren't known statically. For this specific case, we continue to depend on reflectcall. Really, the current approach makes some overly simplistic assumptions about translating the C ABI to the Go ABI. Now we're at least in a much better position to do a proper ABI translation. For comparison, the current cgo call path looks like: GoF (generated C function) -> crosscall2 (in cgo/asm_*.s) -> _cgoexp_GoF (generated Go function) -> cgocallback (in asm_*.s) -> cgocallback_gofunc (in asm_*.s) -> cgocallbackg (in cgocall.go) -> cgocallbackg1 (in cgocall.go) -> reflectcall (in asm_*.s) -> _cgoexpwrap_GoF (generated Go function) -> p.GoF Now the call path looks like: GoF (generated C function) -> crosscall2 (in cgo/asm_*.s) -> cgocallback (in asm_*.s) -> cgocallbackg (in cgocall.go) -> cgocallbackg1 (in cgocall.go) -> _cgoexp_GoF (generated Go function) -> p.GoF Notably: 1. We combine _cgoexp_GoF and _cgoexpwrap_GoF and move the combined operation to the end of the sequence. This combined function also handles reflectcall's previous role. 2. We combined cgocallback and cgocallback_gofunc since the only purpose of having both was to convert a raw PC into a Go function value. We instead construct the Go function value in cgocallbackg1. 3. cgocallbackg1 no longer reaches backwards through the stack to get the arguments to cgocallback_gofunc. Instead, we just pass the arguments down. 4. Currently, we need an explicit msanwrite to mark the results struct as written because reflectcall doesn't do this. Now, the results are written by regular Go assignments, so the Go compiler generates the necessary MSAN annotations. This also means we no longer need to track the size of the arguments frame. Updates #40724, since now we don't need to teach cgo about the register ABI or change how it uses reflectcall. Change-Id: I7840489a2597962aeb670e0c1798a16a7359c94f Reviewed-on: https://go-review.googlesource.com/c/go/+/258938 Trust: Austin Clements <austin@google.com> Run-TryBot: Austin Clements <austin@google.com> TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Cherry Zhang <cherryyz@google.com>
2020-10-01 17:22:38 -04:00
// Construct a struct that will be used to communicate
// arguments from C to Go. The C and Go definitions
// just have to agree. The gcc struct will be compiled
// with __attribute__((packed)) so all padding must be
// accounted for explicitly.
var ctype strings.Builder
const start = "struct {\n"
ctype.WriteString(start)
runtime,cmd/cgo: simplify C -> Go call path This redesigns the way calls work from C to exported Go functions. It removes several steps from the call path, makes cmd/cgo no longer sensitive to the Go calling convention, and eliminates the use of reflectcall from cgo. In order to avoid generating a large amount of FFI glue between the C and Go ABIs, the cgo tool has long depended on generating a C function that marshals the arguments into a struct, and then the actual ABI switch happens in functions with fixed signatures that simply take a pointer to this struct. In a way, this CL simply pushes this idea further. Currently, the cgo tool generates this argument struct in the exact layout of the Go stack frame and depends on reflectcall to unpack it into the appropriate Go call (even though it's actually reflectcall'ing a function generated by cgo). In this CL, we decouple this struct from the Go stack layout. Instead, cgo generates a Go function that takes the struct, unpacks it, and calls the exported function. Since this generated function has a generic signature (like the rest of the call path), we don't need reflectcall and can instead depend on the Go compiler itself to implement the call to the exported Go function. One complication is that syscall.NewCallback on Windows, which converts a Go function into a C function pointer, depends on cgocallback's current dynamic calling approach since the signatures of the callbacks aren't known statically. For this specific case, we continue to depend on reflectcall. Really, the current approach makes some overly simplistic assumptions about translating the C ABI to the Go ABI. Now we're at least in a much better position to do a proper ABI translation. For comparison, the current cgo call path looks like: GoF (generated C function) -> crosscall2 (in cgo/asm_*.s) -> _cgoexp_GoF (generated Go function) -> cgocallback (in asm_*.s) -> cgocallback_gofunc (in asm_*.s) -> cgocallbackg (in cgocall.go) -> cgocallbackg1 (in cgocall.go) -> reflectcall (in asm_*.s) -> _cgoexpwrap_GoF (generated Go function) -> p.GoF Now the call path looks like: GoF (generated C function) -> crosscall2 (in cgo/asm_*.s) -> cgocallback (in asm_*.s) -> cgocallbackg (in cgocall.go) -> cgocallbackg1 (in cgocall.go) -> _cgoexp_GoF (generated Go function) -> p.GoF Notably: 1. We combine _cgoexp_GoF and _cgoexpwrap_GoF and move the combined operation to the end of the sequence. This combined function also handles reflectcall's previous role. 2. We combined cgocallback and cgocallback_gofunc since the only purpose of having both was to convert a raw PC into a Go function value. We instead construct the Go function value in cgocallbackg1. 3. cgocallbackg1 no longer reaches backwards through the stack to get the arguments to cgocallback_gofunc. Instead, we just pass the arguments down. 4. Currently, we need an explicit msanwrite to mark the results struct as written because reflectcall doesn't do this. Now, the results are written by regular Go assignments, so the Go compiler generates the necessary MSAN annotations. This also means we no longer need to track the size of the arguments frame. Updates #40724, since now we don't need to teach cgo about the register ABI or change how it uses reflectcall. Change-Id: I7840489a2597962aeb670e0c1798a16a7359c94f Reviewed-on: https://go-review.googlesource.com/c/go/+/258938 Trust: Austin Clements <austin@google.com> Run-TryBot: Austin Clements <austin@google.com> TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Cherry Zhang <cherryyz@google.com>
2020-10-01 17:22:38 -04:00
gotype := new(bytes.Buffer)
fmt.Fprintf(gotype, "struct {\n")
off := int64(0)
npad := 0
runtime,cmd/cgo: simplify C -> Go call path This redesigns the way calls work from C to exported Go functions. It removes several steps from the call path, makes cmd/cgo no longer sensitive to the Go calling convention, and eliminates the use of reflectcall from cgo. In order to avoid generating a large amount of FFI glue between the C and Go ABIs, the cgo tool has long depended on generating a C function that marshals the arguments into a struct, and then the actual ABI switch happens in functions with fixed signatures that simply take a pointer to this struct. In a way, this CL simply pushes this idea further. Currently, the cgo tool generates this argument struct in the exact layout of the Go stack frame and depends on reflectcall to unpack it into the appropriate Go call (even though it's actually reflectcall'ing a function generated by cgo). In this CL, we decouple this struct from the Go stack layout. Instead, cgo generates a Go function that takes the struct, unpacks it, and calls the exported function. Since this generated function has a generic signature (like the rest of the call path), we don't need reflectcall and can instead depend on the Go compiler itself to implement the call to the exported Go function. One complication is that syscall.NewCallback on Windows, which converts a Go function into a C function pointer, depends on cgocallback's current dynamic calling approach since the signatures of the callbacks aren't known statically. For this specific case, we continue to depend on reflectcall. Really, the current approach makes some overly simplistic assumptions about translating the C ABI to the Go ABI. Now we're at least in a much better position to do a proper ABI translation. For comparison, the current cgo call path looks like: GoF (generated C function) -> crosscall2 (in cgo/asm_*.s) -> _cgoexp_GoF (generated Go function) -> cgocallback (in asm_*.s) -> cgocallback_gofunc (in asm_*.s) -> cgocallbackg (in cgocall.go) -> cgocallbackg1 (in cgocall.go) -> reflectcall (in asm_*.s) -> _cgoexpwrap_GoF (generated Go function) -> p.GoF Now the call path looks like: GoF (generated C function) -> crosscall2 (in cgo/asm_*.s) -> cgocallback (in asm_*.s) -> cgocallbackg (in cgocall.go) -> cgocallbackg1 (in cgocall.go) -> _cgoexp_GoF (generated Go function) -> p.GoF Notably: 1. We combine _cgoexp_GoF and _cgoexpwrap_GoF and move the combined operation to the end of the sequence. This combined function also handles reflectcall's previous role. 2. We combined cgocallback and cgocallback_gofunc since the only purpose of having both was to convert a raw PC into a Go function value. We instead construct the Go function value in cgocallbackg1. 3. cgocallbackg1 no longer reaches backwards through the stack to get the arguments to cgocallback_gofunc. Instead, we just pass the arguments down. 4. Currently, we need an explicit msanwrite to mark the results struct as written because reflectcall doesn't do this. Now, the results are written by regular Go assignments, so the Go compiler generates the necessary MSAN annotations. This also means we no longer need to track the size of the arguments frame. Updates #40724, since now we don't need to teach cgo about the register ABI or change how it uses reflectcall. Change-Id: I7840489a2597962aeb670e0c1798a16a7359c94f Reviewed-on: https://go-review.googlesource.com/c/go/+/258938 Trust: Austin Clements <austin@google.com> Run-TryBot: Austin Clements <austin@google.com> TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Cherry Zhang <cherryyz@google.com>
2020-10-01 17:22:38 -04:00
argField := func(typ ast.Expr, namePat string, args ...interface{}) {
name := fmt.Sprintf(namePat, args...)
t := p.cgoType(typ)
if off%t.Align != 0 {
pad := t.Align - off%t.Align
fmt.Fprintf(&ctype, "\t\tchar __pad%d[%d];\n", npad, pad)
runtime,cmd/cgo: simplify C -> Go call path This redesigns the way calls work from C to exported Go functions. It removes several steps from the call path, makes cmd/cgo no longer sensitive to the Go calling convention, and eliminates the use of reflectcall from cgo. In order to avoid generating a large amount of FFI glue between the C and Go ABIs, the cgo tool has long depended on generating a C function that marshals the arguments into a struct, and then the actual ABI switch happens in functions with fixed signatures that simply take a pointer to this struct. In a way, this CL simply pushes this idea further. Currently, the cgo tool generates this argument struct in the exact layout of the Go stack frame and depends on reflectcall to unpack it into the appropriate Go call (even though it's actually reflectcall'ing a function generated by cgo). In this CL, we decouple this struct from the Go stack layout. Instead, cgo generates a Go function that takes the struct, unpacks it, and calls the exported function. Since this generated function has a generic signature (like the rest of the call path), we don't need reflectcall and can instead depend on the Go compiler itself to implement the call to the exported Go function. One complication is that syscall.NewCallback on Windows, which converts a Go function into a C function pointer, depends on cgocallback's current dynamic calling approach since the signatures of the callbacks aren't known statically. For this specific case, we continue to depend on reflectcall. Really, the current approach makes some overly simplistic assumptions about translating the C ABI to the Go ABI. Now we're at least in a much better position to do a proper ABI translation. For comparison, the current cgo call path looks like: GoF (generated C function) -> crosscall2 (in cgo/asm_*.s) -> _cgoexp_GoF (generated Go function) -> cgocallback (in asm_*.s) -> cgocallback_gofunc (in asm_*.s) -> cgocallbackg (in cgocall.go) -> cgocallbackg1 (in cgocall.go) -> reflectcall (in asm_*.s) -> _cgoexpwrap_GoF (generated Go function) -> p.GoF Now the call path looks like: GoF (generated C function) -> crosscall2 (in cgo/asm_*.s) -> cgocallback (in asm_*.s) -> cgocallbackg (in cgocall.go) -> cgocallbackg1 (in cgocall.go) -> _cgoexp_GoF (generated Go function) -> p.GoF Notably: 1. We combine _cgoexp_GoF and _cgoexpwrap_GoF and move the combined operation to the end of the sequence. This combined function also handles reflectcall's previous role. 2. We combined cgocallback and cgocallback_gofunc since the only purpose of having both was to convert a raw PC into a Go function value. We instead construct the Go function value in cgocallbackg1. 3. cgocallbackg1 no longer reaches backwards through the stack to get the arguments to cgocallback_gofunc. Instead, we just pass the arguments down. 4. Currently, we need an explicit msanwrite to mark the results struct as written because reflectcall doesn't do this. Now, the results are written by regular Go assignments, so the Go compiler generates the necessary MSAN annotations. This also means we no longer need to track the size of the arguments frame. Updates #40724, since now we don't need to teach cgo about the register ABI or change how it uses reflectcall. Change-Id: I7840489a2597962aeb670e0c1798a16a7359c94f Reviewed-on: https://go-review.googlesource.com/c/go/+/258938 Trust: Austin Clements <austin@google.com> Run-TryBot: Austin Clements <austin@google.com> TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Cherry Zhang <cherryyz@google.com>
2020-10-01 17:22:38 -04:00
off += pad
npad++
}
fmt.Fprintf(&ctype, "\t\t%s %s;\n", t.C, name)
runtime,cmd/cgo: simplify C -> Go call path This redesigns the way calls work from C to exported Go functions. It removes several steps from the call path, makes cmd/cgo no longer sensitive to the Go calling convention, and eliminates the use of reflectcall from cgo. In order to avoid generating a large amount of FFI glue between the C and Go ABIs, the cgo tool has long depended on generating a C function that marshals the arguments into a struct, and then the actual ABI switch happens in functions with fixed signatures that simply take a pointer to this struct. In a way, this CL simply pushes this idea further. Currently, the cgo tool generates this argument struct in the exact layout of the Go stack frame and depends on reflectcall to unpack it into the appropriate Go call (even though it's actually reflectcall'ing a function generated by cgo). In this CL, we decouple this struct from the Go stack layout. Instead, cgo generates a Go function that takes the struct, unpacks it, and calls the exported function. Since this generated function has a generic signature (like the rest of the call path), we don't need reflectcall and can instead depend on the Go compiler itself to implement the call to the exported Go function. One complication is that syscall.NewCallback on Windows, which converts a Go function into a C function pointer, depends on cgocallback's current dynamic calling approach since the signatures of the callbacks aren't known statically. For this specific case, we continue to depend on reflectcall. Really, the current approach makes some overly simplistic assumptions about translating the C ABI to the Go ABI. Now we're at least in a much better position to do a proper ABI translation. For comparison, the current cgo call path looks like: GoF (generated C function) -> crosscall2 (in cgo/asm_*.s) -> _cgoexp_GoF (generated Go function) -> cgocallback (in asm_*.s) -> cgocallback_gofunc (in asm_*.s) -> cgocallbackg (in cgocall.go) -> cgocallbackg1 (in cgocall.go) -> reflectcall (in asm_*.s) -> _cgoexpwrap_GoF (generated Go function) -> p.GoF Now the call path looks like: GoF (generated C function) -> crosscall2 (in cgo/asm_*.s) -> cgocallback (in asm_*.s) -> cgocallbackg (in cgocall.go) -> cgocallbackg1 (in cgocall.go) -> _cgoexp_GoF (generated Go function) -> p.GoF Notably: 1. We combine _cgoexp_GoF and _cgoexpwrap_GoF and move the combined operation to the end of the sequence. This combined function also handles reflectcall's previous role. 2. We combined cgocallback and cgocallback_gofunc since the only purpose of having both was to convert a raw PC into a Go function value. We instead construct the Go function value in cgocallbackg1. 3. cgocallbackg1 no longer reaches backwards through the stack to get the arguments to cgocallback_gofunc. Instead, we just pass the arguments down. 4. Currently, we need an explicit msanwrite to mark the results struct as written because reflectcall doesn't do this. Now, the results are written by regular Go assignments, so the Go compiler generates the necessary MSAN annotations. This also means we no longer need to track the size of the arguments frame. Updates #40724, since now we don't need to teach cgo about the register ABI or change how it uses reflectcall. Change-Id: I7840489a2597962aeb670e0c1798a16a7359c94f Reviewed-on: https://go-review.googlesource.com/c/go/+/258938 Trust: Austin Clements <austin@google.com> Run-TryBot: Austin Clements <austin@google.com> TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Cherry Zhang <cherryyz@google.com>
2020-10-01 17:22:38 -04:00
fmt.Fprintf(gotype, "\t\t%s ", name)
noSourceConf.Fprint(gotype, fset, typ)
fmt.Fprintf(gotype, "\n")
off += t.Size
}
runtime,cmd/cgo: simplify C -> Go call path This redesigns the way calls work from C to exported Go functions. It removes several steps from the call path, makes cmd/cgo no longer sensitive to the Go calling convention, and eliminates the use of reflectcall from cgo. In order to avoid generating a large amount of FFI glue between the C and Go ABIs, the cgo tool has long depended on generating a C function that marshals the arguments into a struct, and then the actual ABI switch happens in functions with fixed signatures that simply take a pointer to this struct. In a way, this CL simply pushes this idea further. Currently, the cgo tool generates this argument struct in the exact layout of the Go stack frame and depends on reflectcall to unpack it into the appropriate Go call (even though it's actually reflectcall'ing a function generated by cgo). In this CL, we decouple this struct from the Go stack layout. Instead, cgo generates a Go function that takes the struct, unpacks it, and calls the exported function. Since this generated function has a generic signature (like the rest of the call path), we don't need reflectcall and can instead depend on the Go compiler itself to implement the call to the exported Go function. One complication is that syscall.NewCallback on Windows, which converts a Go function into a C function pointer, depends on cgocallback's current dynamic calling approach since the signatures of the callbacks aren't known statically. For this specific case, we continue to depend on reflectcall. Really, the current approach makes some overly simplistic assumptions about translating the C ABI to the Go ABI. Now we're at least in a much better position to do a proper ABI translation. For comparison, the current cgo call path looks like: GoF (generated C function) -> crosscall2 (in cgo/asm_*.s) -> _cgoexp_GoF (generated Go function) -> cgocallback (in asm_*.s) -> cgocallback_gofunc (in asm_*.s) -> cgocallbackg (in cgocall.go) -> cgocallbackg1 (in cgocall.go) -> reflectcall (in asm_*.s) -> _cgoexpwrap_GoF (generated Go function) -> p.GoF Now the call path looks like: GoF (generated C function) -> crosscall2 (in cgo/asm_*.s) -> cgocallback (in asm_*.s) -> cgocallbackg (in cgocall.go) -> cgocallbackg1 (in cgocall.go) -> _cgoexp_GoF (generated Go function) -> p.GoF Notably: 1. We combine _cgoexp_GoF and _cgoexpwrap_GoF and move the combined operation to the end of the sequence. This combined function also handles reflectcall's previous role. 2. We combined cgocallback and cgocallback_gofunc since the only purpose of having both was to convert a raw PC into a Go function value. We instead construct the Go function value in cgocallbackg1. 3. cgocallbackg1 no longer reaches backwards through the stack to get the arguments to cgocallback_gofunc. Instead, we just pass the arguments down. 4. Currently, we need an explicit msanwrite to mark the results struct as written because reflectcall doesn't do this. Now, the results are written by regular Go assignments, so the Go compiler generates the necessary MSAN annotations. This also means we no longer need to track the size of the arguments frame. Updates #40724, since now we don't need to teach cgo about the register ABI or change how it uses reflectcall. Change-Id: I7840489a2597962aeb670e0c1798a16a7359c94f Reviewed-on: https://go-review.googlesource.com/c/go/+/258938 Trust: Austin Clements <austin@google.com> Run-TryBot: Austin Clements <austin@google.com> TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Cherry Zhang <cherryyz@google.com>
2020-10-01 17:22:38 -04:00
if fn.Recv != nil {
argField(fn.Recv.List[0].Type, "recv")
}
fntype := fn.Type
forFieldList(fntype.Params,
func(i int, aname string, atype ast.Expr) {
runtime,cmd/cgo: simplify C -> Go call path This redesigns the way calls work from C to exported Go functions. It removes several steps from the call path, makes cmd/cgo no longer sensitive to the Go calling convention, and eliminates the use of reflectcall from cgo. In order to avoid generating a large amount of FFI glue between the C and Go ABIs, the cgo tool has long depended on generating a C function that marshals the arguments into a struct, and then the actual ABI switch happens in functions with fixed signatures that simply take a pointer to this struct. In a way, this CL simply pushes this idea further. Currently, the cgo tool generates this argument struct in the exact layout of the Go stack frame and depends on reflectcall to unpack it into the appropriate Go call (even though it's actually reflectcall'ing a function generated by cgo). In this CL, we decouple this struct from the Go stack layout. Instead, cgo generates a Go function that takes the struct, unpacks it, and calls the exported function. Since this generated function has a generic signature (like the rest of the call path), we don't need reflectcall and can instead depend on the Go compiler itself to implement the call to the exported Go function. One complication is that syscall.NewCallback on Windows, which converts a Go function into a C function pointer, depends on cgocallback's current dynamic calling approach since the signatures of the callbacks aren't known statically. For this specific case, we continue to depend on reflectcall. Really, the current approach makes some overly simplistic assumptions about translating the C ABI to the Go ABI. Now we're at least in a much better position to do a proper ABI translation. For comparison, the current cgo call path looks like: GoF (generated C function) -> crosscall2 (in cgo/asm_*.s) -> _cgoexp_GoF (generated Go function) -> cgocallback (in asm_*.s) -> cgocallback_gofunc (in asm_*.s) -> cgocallbackg (in cgocall.go) -> cgocallbackg1 (in cgocall.go) -> reflectcall (in asm_*.s) -> _cgoexpwrap_GoF (generated Go function) -> p.GoF Now the call path looks like: GoF (generated C function) -> crosscall2 (in cgo/asm_*.s) -> cgocallback (in asm_*.s) -> cgocallbackg (in cgocall.go) -> cgocallbackg1 (in cgocall.go) -> _cgoexp_GoF (generated Go function) -> p.GoF Notably: 1. We combine _cgoexp_GoF and _cgoexpwrap_GoF and move the combined operation to the end of the sequence. This combined function also handles reflectcall's previous role. 2. We combined cgocallback and cgocallback_gofunc since the only purpose of having both was to convert a raw PC into a Go function value. We instead construct the Go function value in cgocallbackg1. 3. cgocallbackg1 no longer reaches backwards through the stack to get the arguments to cgocallback_gofunc. Instead, we just pass the arguments down. 4. Currently, we need an explicit msanwrite to mark the results struct as written because reflectcall doesn't do this. Now, the results are written by regular Go assignments, so the Go compiler generates the necessary MSAN annotations. This also means we no longer need to track the size of the arguments frame. Updates #40724, since now we don't need to teach cgo about the register ABI or change how it uses reflectcall. Change-Id: I7840489a2597962aeb670e0c1798a16a7359c94f Reviewed-on: https://go-review.googlesource.com/c/go/+/258938 Trust: Austin Clements <austin@google.com> Run-TryBot: Austin Clements <austin@google.com> TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Cherry Zhang <cherryyz@google.com>
2020-10-01 17:22:38 -04:00
argField(atype, "p%d", i)
})
forFieldList(fntype.Results,
func(i int, aname string, atype ast.Expr) {
runtime,cmd/cgo: simplify C -> Go call path This redesigns the way calls work from C to exported Go functions. It removes several steps from the call path, makes cmd/cgo no longer sensitive to the Go calling convention, and eliminates the use of reflectcall from cgo. In order to avoid generating a large amount of FFI glue between the C and Go ABIs, the cgo tool has long depended on generating a C function that marshals the arguments into a struct, and then the actual ABI switch happens in functions with fixed signatures that simply take a pointer to this struct. In a way, this CL simply pushes this idea further. Currently, the cgo tool generates this argument struct in the exact layout of the Go stack frame and depends on reflectcall to unpack it into the appropriate Go call (even though it's actually reflectcall'ing a function generated by cgo). In this CL, we decouple this struct from the Go stack layout. Instead, cgo generates a Go function that takes the struct, unpacks it, and calls the exported function. Since this generated function has a generic signature (like the rest of the call path), we don't need reflectcall and can instead depend on the Go compiler itself to implement the call to the exported Go function. One complication is that syscall.NewCallback on Windows, which converts a Go function into a C function pointer, depends on cgocallback's current dynamic calling approach since the signatures of the callbacks aren't known statically. For this specific case, we continue to depend on reflectcall. Really, the current approach makes some overly simplistic assumptions about translating the C ABI to the Go ABI. Now we're at least in a much better position to do a proper ABI translation. For comparison, the current cgo call path looks like: GoF (generated C function) -> crosscall2 (in cgo/asm_*.s) -> _cgoexp_GoF (generated Go function) -> cgocallback (in asm_*.s) -> cgocallback_gofunc (in asm_*.s) -> cgocallbackg (in cgocall.go) -> cgocallbackg1 (in cgocall.go) -> reflectcall (in asm_*.s) -> _cgoexpwrap_GoF (generated Go function) -> p.GoF Now the call path looks like: GoF (generated C function) -> crosscall2 (in cgo/asm_*.s) -> cgocallback (in asm_*.s) -> cgocallbackg (in cgocall.go) -> cgocallbackg1 (in cgocall.go) -> _cgoexp_GoF (generated Go function) -> p.GoF Notably: 1. We combine _cgoexp_GoF and _cgoexpwrap_GoF and move the combined operation to the end of the sequence. This combined function also handles reflectcall's previous role. 2. We combined cgocallback and cgocallback_gofunc since the only purpose of having both was to convert a raw PC into a Go function value. We instead construct the Go function value in cgocallbackg1. 3. cgocallbackg1 no longer reaches backwards through the stack to get the arguments to cgocallback_gofunc. Instead, we just pass the arguments down. 4. Currently, we need an explicit msanwrite to mark the results struct as written because reflectcall doesn't do this. Now, the results are written by regular Go assignments, so the Go compiler generates the necessary MSAN annotations. This also means we no longer need to track the size of the arguments frame. Updates #40724, since now we don't need to teach cgo about the register ABI or change how it uses reflectcall. Change-Id: I7840489a2597962aeb670e0c1798a16a7359c94f Reviewed-on: https://go-review.googlesource.com/c/go/+/258938 Trust: Austin Clements <austin@google.com> Run-TryBot: Austin Clements <austin@google.com> TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Cherry Zhang <cherryyz@google.com>
2020-10-01 17:22:38 -04:00
argField(atype, "r%d", i)
})
if ctype.Len() == len(start) {
ctype.WriteString("\t\tchar unused;\n") // avoid empty struct
}
ctype.WriteString("\t}")
runtime,cmd/cgo: simplify C -> Go call path This redesigns the way calls work from C to exported Go functions. It removes several steps from the call path, makes cmd/cgo no longer sensitive to the Go calling convention, and eliminates the use of reflectcall from cgo. In order to avoid generating a large amount of FFI glue between the C and Go ABIs, the cgo tool has long depended on generating a C function that marshals the arguments into a struct, and then the actual ABI switch happens in functions with fixed signatures that simply take a pointer to this struct. In a way, this CL simply pushes this idea further. Currently, the cgo tool generates this argument struct in the exact layout of the Go stack frame and depends on reflectcall to unpack it into the appropriate Go call (even though it's actually reflectcall'ing a function generated by cgo). In this CL, we decouple this struct from the Go stack layout. Instead, cgo generates a Go function that takes the struct, unpacks it, and calls the exported function. Since this generated function has a generic signature (like the rest of the call path), we don't need reflectcall and can instead depend on the Go compiler itself to implement the call to the exported Go function. One complication is that syscall.NewCallback on Windows, which converts a Go function into a C function pointer, depends on cgocallback's current dynamic calling approach since the signatures of the callbacks aren't known statically. For this specific case, we continue to depend on reflectcall. Really, the current approach makes some overly simplistic assumptions about translating the C ABI to the Go ABI. Now we're at least in a much better position to do a proper ABI translation. For comparison, the current cgo call path looks like: GoF (generated C function) -> crosscall2 (in cgo/asm_*.s) -> _cgoexp_GoF (generated Go function) -> cgocallback (in asm_*.s) -> cgocallback_gofunc (in asm_*.s) -> cgocallbackg (in cgocall.go) -> cgocallbackg1 (in cgocall.go) -> reflectcall (in asm_*.s) -> _cgoexpwrap_GoF (generated Go function) -> p.GoF Now the call path looks like: GoF (generated C function) -> crosscall2 (in cgo/asm_*.s) -> cgocallback (in asm_*.s) -> cgocallbackg (in cgocall.go) -> cgocallbackg1 (in cgocall.go) -> _cgoexp_GoF (generated Go function) -> p.GoF Notably: 1. We combine _cgoexp_GoF and _cgoexpwrap_GoF and move the combined operation to the end of the sequence. This combined function also handles reflectcall's previous role. 2. We combined cgocallback and cgocallback_gofunc since the only purpose of having both was to convert a raw PC into a Go function value. We instead construct the Go function value in cgocallbackg1. 3. cgocallbackg1 no longer reaches backwards through the stack to get the arguments to cgocallback_gofunc. Instead, we just pass the arguments down. 4. Currently, we need an explicit msanwrite to mark the results struct as written because reflectcall doesn't do this. Now, the results are written by regular Go assignments, so the Go compiler generates the necessary MSAN annotations. This also means we no longer need to track the size of the arguments frame. Updates #40724, since now we don't need to teach cgo about the register ABI or change how it uses reflectcall. Change-Id: I7840489a2597962aeb670e0c1798a16a7359c94f Reviewed-on: https://go-review.googlesource.com/c/go/+/258938 Trust: Austin Clements <austin@google.com> Run-TryBot: Austin Clements <austin@google.com> TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Cherry Zhang <cherryyz@google.com>
2020-10-01 17:22:38 -04:00
fmt.Fprintf(gotype, "\t}")
// Get the return type of the wrapper function
// compiled by gcc.
gccResult := ""
if fntype.Results == nil || len(fntype.Results.List) == 0 {
gccResult = "void"
} else if len(fntype.Results.List) == 1 && len(fntype.Results.List[0].Names) <= 1 {
gccResult = p.cgoType(fntype.Results.List[0].Type).C.String()
} else {
fmt.Fprintf(fgcch, "\n/* Return type for %s */\n", exp.ExpName)
fmt.Fprintf(fgcch, "struct %s_return {\n", exp.ExpName)
forFieldList(fntype.Results,
func(i int, aname string, atype ast.Expr) {
fmt.Fprintf(fgcch, "\t%s r%d;", p.cgoType(atype).C, i)
if len(aname) > 0 {
fmt.Fprintf(fgcch, " /* %s */", aname)
}
fmt.Fprint(fgcch, "\n")
})
fmt.Fprintf(fgcch, "};\n")
gccResult = "struct " + exp.ExpName + "_return"
}
// Build the wrapper function compiled by gcc.
gccExport := ""
if goos == "windows" {
gccExport = "__declspec(dllexport) "
}
var s strings.Builder
fmt.Fprintf(&s, "%s%s %s(", gccExport, gccResult, exp.ExpName)
if fn.Recv != nil {
s.WriteString(p.cgoType(fn.Recv.List[0].Type).C.String())
s.WriteString(" recv")
}
if len(fntype.Params.List) > 0 {
forFieldList(fntype.Params,
func(i int, aname string, atype ast.Expr) {
if i > 0 || fn.Recv != nil {
s.WriteString(", ")
}
fmt.Fprintf(&s, "%s %s", p.cgoType(atype).C, exportParamName(aname, i))
})
} else {
s.WriteString("void")
}
s.WriteByte(')')
if len(exp.Doc) > 0 {
fmt.Fprintf(fgcch, "\n%s", exp.Doc)
if !strings.HasSuffix(exp.Doc, "\n") {
fmt.Fprint(fgcch, "\n")
}
}
fmt.Fprintf(fgcch, "extern %s;\n", s.String())
runtime,cmd/cgo: simplify C -> Go call path This redesigns the way calls work from C to exported Go functions. It removes several steps from the call path, makes cmd/cgo no longer sensitive to the Go calling convention, and eliminates the use of reflectcall from cgo. In order to avoid generating a large amount of FFI glue between the C and Go ABIs, the cgo tool has long depended on generating a C function that marshals the arguments into a struct, and then the actual ABI switch happens in functions with fixed signatures that simply take a pointer to this struct. In a way, this CL simply pushes this idea further. Currently, the cgo tool generates this argument struct in the exact layout of the Go stack frame and depends on reflectcall to unpack it into the appropriate Go call (even though it's actually reflectcall'ing a function generated by cgo). In this CL, we decouple this struct from the Go stack layout. Instead, cgo generates a Go function that takes the struct, unpacks it, and calls the exported function. Since this generated function has a generic signature (like the rest of the call path), we don't need reflectcall and can instead depend on the Go compiler itself to implement the call to the exported Go function. One complication is that syscall.NewCallback on Windows, which converts a Go function into a C function pointer, depends on cgocallback's current dynamic calling approach since the signatures of the callbacks aren't known statically. For this specific case, we continue to depend on reflectcall. Really, the current approach makes some overly simplistic assumptions about translating the C ABI to the Go ABI. Now we're at least in a much better position to do a proper ABI translation. For comparison, the current cgo call path looks like: GoF (generated C function) -> crosscall2 (in cgo/asm_*.s) -> _cgoexp_GoF (generated Go function) -> cgocallback (in asm_*.s) -> cgocallback_gofunc (in asm_*.s) -> cgocallbackg (in cgocall.go) -> cgocallbackg1 (in cgocall.go) -> reflectcall (in asm_*.s) -> _cgoexpwrap_GoF (generated Go function) -> p.GoF Now the call path looks like: GoF (generated C function) -> crosscall2 (in cgo/asm_*.s) -> cgocallback (in asm_*.s) -> cgocallbackg (in cgocall.go) -> cgocallbackg1 (in cgocall.go) -> _cgoexp_GoF (generated Go function) -> p.GoF Notably: 1. We combine _cgoexp_GoF and _cgoexpwrap_GoF and move the combined operation to the end of the sequence. This combined function also handles reflectcall's previous role. 2. We combined cgocallback and cgocallback_gofunc since the only purpose of having both was to convert a raw PC into a Go function value. We instead construct the Go function value in cgocallbackg1. 3. cgocallbackg1 no longer reaches backwards through the stack to get the arguments to cgocallback_gofunc. Instead, we just pass the arguments down. 4. Currently, we need an explicit msanwrite to mark the results struct as written because reflectcall doesn't do this. Now, the results are written by regular Go assignments, so the Go compiler generates the necessary MSAN annotations. This also means we no longer need to track the size of the arguments frame. Updates #40724, since now we don't need to teach cgo about the register ABI or change how it uses reflectcall. Change-Id: I7840489a2597962aeb670e0c1798a16a7359c94f Reviewed-on: https://go-review.googlesource.com/c/go/+/258938 Trust: Austin Clements <austin@google.com> Run-TryBot: Austin Clements <austin@google.com> TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Cherry Zhang <cherryyz@google.com>
2020-10-01 17:22:38 -04:00
fmt.Fprintf(fgcc, "extern void _cgoexp%s_%s(void *);\n", cPrefix, exp.ExpName)
fmt.Fprintf(fgcc, "\nCGO_NO_SANITIZE_THREAD")
fmt.Fprintf(fgcc, "\n%s\n", s.String())
fmt.Fprintf(fgcc, "{\n")
fmt.Fprintf(fgcc, "\tsize_t _cgo_ctxt = _cgo_wait_runtime_init_done();\n")
runtime,cmd/cgo: simplify C -> Go call path This redesigns the way calls work from C to exported Go functions. It removes several steps from the call path, makes cmd/cgo no longer sensitive to the Go calling convention, and eliminates the use of reflectcall from cgo. In order to avoid generating a large amount of FFI glue between the C and Go ABIs, the cgo tool has long depended on generating a C function that marshals the arguments into a struct, and then the actual ABI switch happens in functions with fixed signatures that simply take a pointer to this struct. In a way, this CL simply pushes this idea further. Currently, the cgo tool generates this argument struct in the exact layout of the Go stack frame and depends on reflectcall to unpack it into the appropriate Go call (even though it's actually reflectcall'ing a function generated by cgo). In this CL, we decouple this struct from the Go stack layout. Instead, cgo generates a Go function that takes the struct, unpacks it, and calls the exported function. Since this generated function has a generic signature (like the rest of the call path), we don't need reflectcall and can instead depend on the Go compiler itself to implement the call to the exported Go function. One complication is that syscall.NewCallback on Windows, which converts a Go function into a C function pointer, depends on cgocallback's current dynamic calling approach since the signatures of the callbacks aren't known statically. For this specific case, we continue to depend on reflectcall. Really, the current approach makes some overly simplistic assumptions about translating the C ABI to the Go ABI. Now we're at least in a much better position to do a proper ABI translation. For comparison, the current cgo call path looks like: GoF (generated C function) -> crosscall2 (in cgo/asm_*.s) -> _cgoexp_GoF (generated Go function) -> cgocallback (in asm_*.s) -> cgocallback_gofunc (in asm_*.s) -> cgocallbackg (in cgocall.go) -> cgocallbackg1 (in cgocall.go) -> reflectcall (in asm_*.s) -> _cgoexpwrap_GoF (generated Go function) -> p.GoF Now the call path looks like: GoF (generated C function) -> crosscall2 (in cgo/asm_*.s) -> cgocallback (in asm_*.s) -> cgocallbackg (in cgocall.go) -> cgocallbackg1 (in cgocall.go) -> _cgoexp_GoF (generated Go function) -> p.GoF Notably: 1. We combine _cgoexp_GoF and _cgoexpwrap_GoF and move the combined operation to the end of the sequence. This combined function also handles reflectcall's previous role. 2. We combined cgocallback and cgocallback_gofunc since the only purpose of having both was to convert a raw PC into a Go function value. We instead construct the Go function value in cgocallbackg1. 3. cgocallbackg1 no longer reaches backwards through the stack to get the arguments to cgocallback_gofunc. Instead, we just pass the arguments down. 4. Currently, we need an explicit msanwrite to mark the results struct as written because reflectcall doesn't do this. Now, the results are written by regular Go assignments, so the Go compiler generates the necessary MSAN annotations. This also means we no longer need to track the size of the arguments frame. Updates #40724, since now we don't need to teach cgo about the register ABI or change how it uses reflectcall. Change-Id: I7840489a2597962aeb670e0c1798a16a7359c94f Reviewed-on: https://go-review.googlesource.com/c/go/+/258938 Trust: Austin Clements <austin@google.com> Run-TryBot: Austin Clements <austin@google.com> TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Cherry Zhang <cherryyz@google.com>
2020-10-01 17:22:38 -04:00
// The results part of the argument structure must be
// initialized to 0 so the write barriers generated by
// the assignments to these fields in Go are safe.
//
// We use a local static variable to get the zeroed
// value of the argument type. This avoids including
// string.h for memset, and is also robust to C++
// types with constructors. Both GCC and LLVM optimize
// this into just zeroing _cgo_a.
fmt.Fprintf(fgcc, "\ttypedef %s %v _cgo_argtype;\n", ctype.String(), p.packedAttribute())
fmt.Fprintf(fgcc, "\tstatic _cgo_argtype _cgo_zero;\n")
fmt.Fprintf(fgcc, "\t_cgo_argtype _cgo_a = _cgo_zero;\n")
if gccResult != "void" && (len(fntype.Results.List) > 1 || len(fntype.Results.List[0].Names) > 1) {
fmt.Fprintf(fgcc, "\t%s r;\n", gccResult)
}
if fn.Recv != nil {
fmt.Fprintf(fgcc, "\t_cgo_a.recv = recv;\n")
}
forFieldList(fntype.Params,
func(i int, aname string, atype ast.Expr) {
fmt.Fprintf(fgcc, "\t_cgo_a.p%d = %s;\n", i, exportParamName(aname, i))
})
fmt.Fprintf(fgcc, "\t_cgo_tsan_release();\n")
fmt.Fprintf(fgcc, "\tcrosscall2(_cgoexp%s_%s, &_cgo_a, %d, _cgo_ctxt);\n", cPrefix, exp.ExpName, off)
fmt.Fprintf(fgcc, "\t_cgo_tsan_acquire();\n")
fmt.Fprintf(fgcc, "\t_cgo_release_context(_cgo_ctxt);\n")
if gccResult != "void" {
if len(fntype.Results.List) == 1 && len(fntype.Results.List[0].Names) <= 1 {
fmt.Fprintf(fgcc, "\treturn _cgo_a.r0;\n")
} else {
forFieldList(fntype.Results,
func(i int, aname string, atype ast.Expr) {
fmt.Fprintf(fgcc, "\tr.r%d = _cgo_a.r%d;\n", i, i)
})
fmt.Fprintf(fgcc, "\treturn r;\n")
}
}
fmt.Fprintf(fgcc, "}\n")
// In internal linking mode, the Go linker sees both
// the C wrapper written above and the Go wrapper it
// references. Hence, export the C wrapper (e.g., for
// if we're building a shared object). The Go linker
// will resolve the C wrapper's reference to the Go
// wrapper without a separate export.
cmd/cgo: put the real C function in the dynamic symbol table In the past, cgo generated Go code and C code. The C code was linked into a shared library. The Go code was built into an executable that dynamically linked against that shared library. C wrappers were exported from the shared library, and the Go code called them. It was all a long time ago, but in order to permit C code to call back into Go, somebody implemented #pragma dynexport (https://golang.org/cl/661043) to export a Go symbol into the dynamic symbol table. Then that same person added code to cgo to recognize //export comments (https://golang.org/cl/853042). The //export comments were implemented by generating C code, to be compiled by GCC, that would refer to C code, to be compiled by 6c, that would call the Go code. The GCC code would go into a shared library. The code compiled by 6c would be in the Go executable. The GCC code needed to refer to the 6c code, so the 6c function was marked with #pragma dynexport. The important point here is that #pragma dynexport was used to expose an internal detail of the implementation of an exported function, because at the time it was necessary. Moving forward to today, cgo no longer generates a shared library and 6c no longer exists. It's still true that we have a function compiled by GCC that refers to a wrapper function now written in Go. In the normal case today we are doing an external link, and we use a //go:cgo_export_static function to make the Go wrapper function visible to the C code under a known name. The #pragma dynexport statement has become a //go:cgo_export_dynamic comment on the Go code. That comment only takes effect when doing internal linking. The comment tells the linker to put the symbol in the dynamic symbol table. That still makes sense for the now unusual case of using internal linking with a shared library. However, all the changes to this code have carefully preserved the property that the //go:cgo_export_dynamic comment refers to an internal detail of the implementation of an exported function. That was necessary a long time ago, but no longer makes sense. This CL changes the code to put the actual C-callable function into the dynamic symbol table. I considered dropping the comment entirely, but it turns out that there is even a test for this, so I preserved it. Change-Id: I66a7958e366e5974363099bfaa6ba862ca327849 Reviewed-on: https://go-review.googlesource.com/17061 Run-TryBot: Ian Lance Taylor <iant@golang.org> Reviewed-by: Minux Ma <minux@golang.org>
2015-11-19 10:23:20 -08:00
fmt.Fprintf(fgo2, "//go:cgo_export_dynamic %s\n", exp.ExpName)
// cgo_export_static refers to a symbol by its linker
// name, so set the linker name of the Go wrapper.
fmt.Fprintf(fgo2, "//go:linkname _cgoexp%s_%s _cgoexp%s_%s\n", cPrefix, exp.ExpName, cPrefix, exp.ExpName)
// In external linking mode, the Go linker sees the Go
// wrapper, but not the C wrapper. For this case,
// export the Go wrapper so the host linker can
// resolve the reference from the C wrapper to the Go
// wrapper.
fmt.Fprintf(fgo2, "//go:cgo_export_static _cgoexp%s_%s\n", cPrefix, exp.ExpName)
// Build the wrapper function compiled by cmd/compile.
// This unpacks the argument struct above and calls the Go function.
runtime,cmd/cgo: simplify C -> Go call path This redesigns the way calls work from C to exported Go functions. It removes several steps from the call path, makes cmd/cgo no longer sensitive to the Go calling convention, and eliminates the use of reflectcall from cgo. In order to avoid generating a large amount of FFI glue between the C and Go ABIs, the cgo tool has long depended on generating a C function that marshals the arguments into a struct, and then the actual ABI switch happens in functions with fixed signatures that simply take a pointer to this struct. In a way, this CL simply pushes this idea further. Currently, the cgo tool generates this argument struct in the exact layout of the Go stack frame and depends on reflectcall to unpack it into the appropriate Go call (even though it's actually reflectcall'ing a function generated by cgo). In this CL, we decouple this struct from the Go stack layout. Instead, cgo generates a Go function that takes the struct, unpacks it, and calls the exported function. Since this generated function has a generic signature (like the rest of the call path), we don't need reflectcall and can instead depend on the Go compiler itself to implement the call to the exported Go function. One complication is that syscall.NewCallback on Windows, which converts a Go function into a C function pointer, depends on cgocallback's current dynamic calling approach since the signatures of the callbacks aren't known statically. For this specific case, we continue to depend on reflectcall. Really, the current approach makes some overly simplistic assumptions about translating the C ABI to the Go ABI. Now we're at least in a much better position to do a proper ABI translation. For comparison, the current cgo call path looks like: GoF (generated C function) -> crosscall2 (in cgo/asm_*.s) -> _cgoexp_GoF (generated Go function) -> cgocallback (in asm_*.s) -> cgocallback_gofunc (in asm_*.s) -> cgocallbackg (in cgocall.go) -> cgocallbackg1 (in cgocall.go) -> reflectcall (in asm_*.s) -> _cgoexpwrap_GoF (generated Go function) -> p.GoF Now the call path looks like: GoF (generated C function) -> crosscall2 (in cgo/asm_*.s) -> cgocallback (in asm_*.s) -> cgocallbackg (in cgocall.go) -> cgocallbackg1 (in cgocall.go) -> _cgoexp_GoF (generated Go function) -> p.GoF Notably: 1. We combine _cgoexp_GoF and _cgoexpwrap_GoF and move the combined operation to the end of the sequence. This combined function also handles reflectcall's previous role. 2. We combined cgocallback and cgocallback_gofunc since the only purpose of having both was to convert a raw PC into a Go function value. We instead construct the Go function value in cgocallbackg1. 3. cgocallbackg1 no longer reaches backwards through the stack to get the arguments to cgocallback_gofunc. Instead, we just pass the arguments down. 4. Currently, we need an explicit msanwrite to mark the results struct as written because reflectcall doesn't do this. Now, the results are written by regular Go assignments, so the Go compiler generates the necessary MSAN annotations. This also means we no longer need to track the size of the arguments frame. Updates #40724, since now we don't need to teach cgo about the register ABI or change how it uses reflectcall. Change-Id: I7840489a2597962aeb670e0c1798a16a7359c94f Reviewed-on: https://go-review.googlesource.com/c/go/+/258938 Trust: Austin Clements <austin@google.com> Run-TryBot: Austin Clements <austin@google.com> TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Cherry Zhang <cherryyz@google.com>
2020-10-01 17:22:38 -04:00
fmt.Fprintf(fgo2, "func _cgoexp%s_%s(a *%s) {\n", cPrefix, exp.ExpName, gotype)
fmt.Fprintf(fm, "void _cgoexp%s_%s(void* p __attribute__((unused))){}\n", cPrefix, exp.ExpName)
fmt.Fprintf(fgo2, "\t")
if gccResult != "void" {
runtime,cmd/cgo: simplify C -> Go call path This redesigns the way calls work from C to exported Go functions. It removes several steps from the call path, makes cmd/cgo no longer sensitive to the Go calling convention, and eliminates the use of reflectcall from cgo. In order to avoid generating a large amount of FFI glue between the C and Go ABIs, the cgo tool has long depended on generating a C function that marshals the arguments into a struct, and then the actual ABI switch happens in functions with fixed signatures that simply take a pointer to this struct. In a way, this CL simply pushes this idea further. Currently, the cgo tool generates this argument struct in the exact layout of the Go stack frame and depends on reflectcall to unpack it into the appropriate Go call (even though it's actually reflectcall'ing a function generated by cgo). In this CL, we decouple this struct from the Go stack layout. Instead, cgo generates a Go function that takes the struct, unpacks it, and calls the exported function. Since this generated function has a generic signature (like the rest of the call path), we don't need reflectcall and can instead depend on the Go compiler itself to implement the call to the exported Go function. One complication is that syscall.NewCallback on Windows, which converts a Go function into a C function pointer, depends on cgocallback's current dynamic calling approach since the signatures of the callbacks aren't known statically. For this specific case, we continue to depend on reflectcall. Really, the current approach makes some overly simplistic assumptions about translating the C ABI to the Go ABI. Now we're at least in a much better position to do a proper ABI translation. For comparison, the current cgo call path looks like: GoF (generated C function) -> crosscall2 (in cgo/asm_*.s) -> _cgoexp_GoF (generated Go function) -> cgocallback (in asm_*.s) -> cgocallback_gofunc (in asm_*.s) -> cgocallbackg (in cgocall.go) -> cgocallbackg1 (in cgocall.go) -> reflectcall (in asm_*.s) -> _cgoexpwrap_GoF (generated Go function) -> p.GoF Now the call path looks like: GoF (generated C function) -> crosscall2 (in cgo/asm_*.s) -> cgocallback (in asm_*.s) -> cgocallbackg (in cgocall.go) -> cgocallbackg1 (in cgocall.go) -> _cgoexp_GoF (generated Go function) -> p.GoF Notably: 1. We combine _cgoexp_GoF and _cgoexpwrap_GoF and move the combined operation to the end of the sequence. This combined function also handles reflectcall's previous role. 2. We combined cgocallback and cgocallback_gofunc since the only purpose of having both was to convert a raw PC into a Go function value. We instead construct the Go function value in cgocallbackg1. 3. cgocallbackg1 no longer reaches backwards through the stack to get the arguments to cgocallback_gofunc. Instead, we just pass the arguments down. 4. Currently, we need an explicit msanwrite to mark the results struct as written because reflectcall doesn't do this. Now, the results are written by regular Go assignments, so the Go compiler generates the necessary MSAN annotations. This also means we no longer need to track the size of the arguments frame. Updates #40724, since now we don't need to teach cgo about the register ABI or change how it uses reflectcall. Change-Id: I7840489a2597962aeb670e0c1798a16a7359c94f Reviewed-on: https://go-review.googlesource.com/c/go/+/258938 Trust: Austin Clements <austin@google.com> Run-TryBot: Austin Clements <austin@google.com> TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Cherry Zhang <cherryyz@google.com>
2020-10-01 17:22:38 -04:00
// Write results back to frame.
forFieldList(fntype.Results,
func(i int, aname string, atype ast.Expr) {
if i > 0 {
runtime,cmd/cgo: simplify C -> Go call path This redesigns the way calls work from C to exported Go functions. It removes several steps from the call path, makes cmd/cgo no longer sensitive to the Go calling convention, and eliminates the use of reflectcall from cgo. In order to avoid generating a large amount of FFI glue between the C and Go ABIs, the cgo tool has long depended on generating a C function that marshals the arguments into a struct, and then the actual ABI switch happens in functions with fixed signatures that simply take a pointer to this struct. In a way, this CL simply pushes this idea further. Currently, the cgo tool generates this argument struct in the exact layout of the Go stack frame and depends on reflectcall to unpack it into the appropriate Go call (even though it's actually reflectcall'ing a function generated by cgo). In this CL, we decouple this struct from the Go stack layout. Instead, cgo generates a Go function that takes the struct, unpacks it, and calls the exported function. Since this generated function has a generic signature (like the rest of the call path), we don't need reflectcall and can instead depend on the Go compiler itself to implement the call to the exported Go function. One complication is that syscall.NewCallback on Windows, which converts a Go function into a C function pointer, depends on cgocallback's current dynamic calling approach since the signatures of the callbacks aren't known statically. For this specific case, we continue to depend on reflectcall. Really, the current approach makes some overly simplistic assumptions about translating the C ABI to the Go ABI. Now we're at least in a much better position to do a proper ABI translation. For comparison, the current cgo call path looks like: GoF (generated C function) -> crosscall2 (in cgo/asm_*.s) -> _cgoexp_GoF (generated Go function) -> cgocallback (in asm_*.s) -> cgocallback_gofunc (in asm_*.s) -> cgocallbackg (in cgocall.go) -> cgocallbackg1 (in cgocall.go) -> reflectcall (in asm_*.s) -> _cgoexpwrap_GoF (generated Go function) -> p.GoF Now the call path looks like: GoF (generated C function) -> crosscall2 (in cgo/asm_*.s) -> cgocallback (in asm_*.s) -> cgocallbackg (in cgocall.go) -> cgocallbackg1 (in cgocall.go) -> _cgoexp_GoF (generated Go function) -> p.GoF Notably: 1. We combine _cgoexp_GoF and _cgoexpwrap_GoF and move the combined operation to the end of the sequence. This combined function also handles reflectcall's previous role. 2. We combined cgocallback and cgocallback_gofunc since the only purpose of having both was to convert a raw PC into a Go function value. We instead construct the Go function value in cgocallbackg1. 3. cgocallbackg1 no longer reaches backwards through the stack to get the arguments to cgocallback_gofunc. Instead, we just pass the arguments down. 4. Currently, we need an explicit msanwrite to mark the results struct as written because reflectcall doesn't do this. Now, the results are written by regular Go assignments, so the Go compiler generates the necessary MSAN annotations. This also means we no longer need to track the size of the arguments frame. Updates #40724, since now we don't need to teach cgo about the register ABI or change how it uses reflectcall. Change-Id: I7840489a2597962aeb670e0c1798a16a7359c94f Reviewed-on: https://go-review.googlesource.com/c/go/+/258938 Trust: Austin Clements <austin@google.com> Run-TryBot: Austin Clements <austin@google.com> TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Cherry Zhang <cherryyz@google.com>
2020-10-01 17:22:38 -04:00
fmt.Fprintf(fgo2, ", ")
}
runtime,cmd/cgo: simplify C -> Go call path This redesigns the way calls work from C to exported Go functions. It removes several steps from the call path, makes cmd/cgo no longer sensitive to the Go calling convention, and eliminates the use of reflectcall from cgo. In order to avoid generating a large amount of FFI glue between the C and Go ABIs, the cgo tool has long depended on generating a C function that marshals the arguments into a struct, and then the actual ABI switch happens in functions with fixed signatures that simply take a pointer to this struct. In a way, this CL simply pushes this idea further. Currently, the cgo tool generates this argument struct in the exact layout of the Go stack frame and depends on reflectcall to unpack it into the appropriate Go call (even though it's actually reflectcall'ing a function generated by cgo). In this CL, we decouple this struct from the Go stack layout. Instead, cgo generates a Go function that takes the struct, unpacks it, and calls the exported function. Since this generated function has a generic signature (like the rest of the call path), we don't need reflectcall and can instead depend on the Go compiler itself to implement the call to the exported Go function. One complication is that syscall.NewCallback on Windows, which converts a Go function into a C function pointer, depends on cgocallback's current dynamic calling approach since the signatures of the callbacks aren't known statically. For this specific case, we continue to depend on reflectcall. Really, the current approach makes some overly simplistic assumptions about translating the C ABI to the Go ABI. Now we're at least in a much better position to do a proper ABI translation. For comparison, the current cgo call path looks like: GoF (generated C function) -> crosscall2 (in cgo/asm_*.s) -> _cgoexp_GoF (generated Go function) -> cgocallback (in asm_*.s) -> cgocallback_gofunc (in asm_*.s) -> cgocallbackg (in cgocall.go) -> cgocallbackg1 (in cgocall.go) -> reflectcall (in asm_*.s) -> _cgoexpwrap_GoF (generated Go function) -> p.GoF Now the call path looks like: GoF (generated C function) -> crosscall2 (in cgo/asm_*.s) -> cgocallback (in asm_*.s) -> cgocallbackg (in cgocall.go) -> cgocallbackg1 (in cgocall.go) -> _cgoexp_GoF (generated Go function) -> p.GoF Notably: 1. We combine _cgoexp_GoF and _cgoexpwrap_GoF and move the combined operation to the end of the sequence. This combined function also handles reflectcall's previous role. 2. We combined cgocallback and cgocallback_gofunc since the only purpose of having both was to convert a raw PC into a Go function value. We instead construct the Go function value in cgocallbackg1. 3. cgocallbackg1 no longer reaches backwards through the stack to get the arguments to cgocallback_gofunc. Instead, we just pass the arguments down. 4. Currently, we need an explicit msanwrite to mark the results struct as written because reflectcall doesn't do this. Now, the results are written by regular Go assignments, so the Go compiler generates the necessary MSAN annotations. This also means we no longer need to track the size of the arguments frame. Updates #40724, since now we don't need to teach cgo about the register ABI or change how it uses reflectcall. Change-Id: I7840489a2597962aeb670e0c1798a16a7359c94f Reviewed-on: https://go-review.googlesource.com/c/go/+/258938 Trust: Austin Clements <austin@google.com> Run-TryBot: Austin Clements <austin@google.com> TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Cherry Zhang <cherryyz@google.com>
2020-10-01 17:22:38 -04:00
fmt.Fprintf(fgo2, "a.r%d", i)
})
runtime,cmd/cgo: simplify C -> Go call path This redesigns the way calls work from C to exported Go functions. It removes several steps from the call path, makes cmd/cgo no longer sensitive to the Go calling convention, and eliminates the use of reflectcall from cgo. In order to avoid generating a large amount of FFI glue between the C and Go ABIs, the cgo tool has long depended on generating a C function that marshals the arguments into a struct, and then the actual ABI switch happens in functions with fixed signatures that simply take a pointer to this struct. In a way, this CL simply pushes this idea further. Currently, the cgo tool generates this argument struct in the exact layout of the Go stack frame and depends on reflectcall to unpack it into the appropriate Go call (even though it's actually reflectcall'ing a function generated by cgo). In this CL, we decouple this struct from the Go stack layout. Instead, cgo generates a Go function that takes the struct, unpacks it, and calls the exported function. Since this generated function has a generic signature (like the rest of the call path), we don't need reflectcall and can instead depend on the Go compiler itself to implement the call to the exported Go function. One complication is that syscall.NewCallback on Windows, which converts a Go function into a C function pointer, depends on cgocallback's current dynamic calling approach since the signatures of the callbacks aren't known statically. For this specific case, we continue to depend on reflectcall. Really, the current approach makes some overly simplistic assumptions about translating the C ABI to the Go ABI. Now we're at least in a much better position to do a proper ABI translation. For comparison, the current cgo call path looks like: GoF (generated C function) -> crosscall2 (in cgo/asm_*.s) -> _cgoexp_GoF (generated Go function) -> cgocallback (in asm_*.s) -> cgocallback_gofunc (in asm_*.s) -> cgocallbackg (in cgocall.go) -> cgocallbackg1 (in cgocall.go) -> reflectcall (in asm_*.s) -> _cgoexpwrap_GoF (generated Go function) -> p.GoF Now the call path looks like: GoF (generated C function) -> crosscall2 (in cgo/asm_*.s) -> cgocallback (in asm_*.s) -> cgocallbackg (in cgocall.go) -> cgocallbackg1 (in cgocall.go) -> _cgoexp_GoF (generated Go function) -> p.GoF Notably: 1. We combine _cgoexp_GoF and _cgoexpwrap_GoF and move the combined operation to the end of the sequence. This combined function also handles reflectcall's previous role. 2. We combined cgocallback and cgocallback_gofunc since the only purpose of having both was to convert a raw PC into a Go function value. We instead construct the Go function value in cgocallbackg1. 3. cgocallbackg1 no longer reaches backwards through the stack to get the arguments to cgocallback_gofunc. Instead, we just pass the arguments down. 4. Currently, we need an explicit msanwrite to mark the results struct as written because reflectcall doesn't do this. Now, the results are written by regular Go assignments, so the Go compiler generates the necessary MSAN annotations. This also means we no longer need to track the size of the arguments frame. Updates #40724, since now we don't need to teach cgo about the register ABI or change how it uses reflectcall. Change-Id: I7840489a2597962aeb670e0c1798a16a7359c94f Reviewed-on: https://go-review.googlesource.com/c/go/+/258938 Trust: Austin Clements <austin@google.com> Run-TryBot: Austin Clements <austin@google.com> TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Cherry Zhang <cherryyz@google.com>
2020-10-01 17:22:38 -04:00
fmt.Fprintf(fgo2, " = ")
}
if fn.Recv != nil {
runtime,cmd/cgo: simplify C -> Go call path This redesigns the way calls work from C to exported Go functions. It removes several steps from the call path, makes cmd/cgo no longer sensitive to the Go calling convention, and eliminates the use of reflectcall from cgo. In order to avoid generating a large amount of FFI glue between the C and Go ABIs, the cgo tool has long depended on generating a C function that marshals the arguments into a struct, and then the actual ABI switch happens in functions with fixed signatures that simply take a pointer to this struct. In a way, this CL simply pushes this idea further. Currently, the cgo tool generates this argument struct in the exact layout of the Go stack frame and depends on reflectcall to unpack it into the appropriate Go call (even though it's actually reflectcall'ing a function generated by cgo). In this CL, we decouple this struct from the Go stack layout. Instead, cgo generates a Go function that takes the struct, unpacks it, and calls the exported function. Since this generated function has a generic signature (like the rest of the call path), we don't need reflectcall and can instead depend on the Go compiler itself to implement the call to the exported Go function. One complication is that syscall.NewCallback on Windows, which converts a Go function into a C function pointer, depends on cgocallback's current dynamic calling approach since the signatures of the callbacks aren't known statically. For this specific case, we continue to depend on reflectcall. Really, the current approach makes some overly simplistic assumptions about translating the C ABI to the Go ABI. Now we're at least in a much better position to do a proper ABI translation. For comparison, the current cgo call path looks like: GoF (generated C function) -> crosscall2 (in cgo/asm_*.s) -> _cgoexp_GoF (generated Go function) -> cgocallback (in asm_*.s) -> cgocallback_gofunc (in asm_*.s) -> cgocallbackg (in cgocall.go) -> cgocallbackg1 (in cgocall.go) -> reflectcall (in asm_*.s) -> _cgoexpwrap_GoF (generated Go function) -> p.GoF Now the call path looks like: GoF (generated C function) -> crosscall2 (in cgo/asm_*.s) -> cgocallback (in asm_*.s) -> cgocallbackg (in cgocall.go) -> cgocallbackg1 (in cgocall.go) -> _cgoexp_GoF (generated Go function) -> p.GoF Notably: 1. We combine _cgoexp_GoF and _cgoexpwrap_GoF and move the combined operation to the end of the sequence. This combined function also handles reflectcall's previous role. 2. We combined cgocallback and cgocallback_gofunc since the only purpose of having both was to convert a raw PC into a Go function value. We instead construct the Go function value in cgocallbackg1. 3. cgocallbackg1 no longer reaches backwards through the stack to get the arguments to cgocallback_gofunc. Instead, we just pass the arguments down. 4. Currently, we need an explicit msanwrite to mark the results struct as written because reflectcall doesn't do this. Now, the results are written by regular Go assignments, so the Go compiler generates the necessary MSAN annotations. This also means we no longer need to track the size of the arguments frame. Updates #40724, since now we don't need to teach cgo about the register ABI or change how it uses reflectcall. Change-Id: I7840489a2597962aeb670e0c1798a16a7359c94f Reviewed-on: https://go-review.googlesource.com/c/go/+/258938 Trust: Austin Clements <austin@google.com> Run-TryBot: Austin Clements <austin@google.com> TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Cherry Zhang <cherryyz@google.com>
2020-10-01 17:22:38 -04:00
fmt.Fprintf(fgo2, "a.recv.")
}
fmt.Fprintf(fgo2, "%s(", exp.Func.Name)
forFieldList(fntype.Params,
func(i int, aname string, atype ast.Expr) {
if i > 0 {
fmt.Fprint(fgo2, ", ")
}
runtime,cmd/cgo: simplify C -> Go call path This redesigns the way calls work from C to exported Go functions. It removes several steps from the call path, makes cmd/cgo no longer sensitive to the Go calling convention, and eliminates the use of reflectcall from cgo. In order to avoid generating a large amount of FFI glue between the C and Go ABIs, the cgo tool has long depended on generating a C function that marshals the arguments into a struct, and then the actual ABI switch happens in functions with fixed signatures that simply take a pointer to this struct. In a way, this CL simply pushes this idea further. Currently, the cgo tool generates this argument struct in the exact layout of the Go stack frame and depends on reflectcall to unpack it into the appropriate Go call (even though it's actually reflectcall'ing a function generated by cgo). In this CL, we decouple this struct from the Go stack layout. Instead, cgo generates a Go function that takes the struct, unpacks it, and calls the exported function. Since this generated function has a generic signature (like the rest of the call path), we don't need reflectcall and can instead depend on the Go compiler itself to implement the call to the exported Go function. One complication is that syscall.NewCallback on Windows, which converts a Go function into a C function pointer, depends on cgocallback's current dynamic calling approach since the signatures of the callbacks aren't known statically. For this specific case, we continue to depend on reflectcall. Really, the current approach makes some overly simplistic assumptions about translating the C ABI to the Go ABI. Now we're at least in a much better position to do a proper ABI translation. For comparison, the current cgo call path looks like: GoF (generated C function) -> crosscall2 (in cgo/asm_*.s) -> _cgoexp_GoF (generated Go function) -> cgocallback (in asm_*.s) -> cgocallback_gofunc (in asm_*.s) -> cgocallbackg (in cgocall.go) -> cgocallbackg1 (in cgocall.go) -> reflectcall (in asm_*.s) -> _cgoexpwrap_GoF (generated Go function) -> p.GoF Now the call path looks like: GoF (generated C function) -> crosscall2 (in cgo/asm_*.s) -> cgocallback (in asm_*.s) -> cgocallbackg (in cgocall.go) -> cgocallbackg1 (in cgocall.go) -> _cgoexp_GoF (generated Go function) -> p.GoF Notably: 1. We combine _cgoexp_GoF and _cgoexpwrap_GoF and move the combined operation to the end of the sequence. This combined function also handles reflectcall's previous role. 2. We combined cgocallback and cgocallback_gofunc since the only purpose of having both was to convert a raw PC into a Go function value. We instead construct the Go function value in cgocallbackg1. 3. cgocallbackg1 no longer reaches backwards through the stack to get the arguments to cgocallback_gofunc. Instead, we just pass the arguments down. 4. Currently, we need an explicit msanwrite to mark the results struct as written because reflectcall doesn't do this. Now, the results are written by regular Go assignments, so the Go compiler generates the necessary MSAN annotations. This also means we no longer need to track the size of the arguments frame. Updates #40724, since now we don't need to teach cgo about the register ABI or change how it uses reflectcall. Change-Id: I7840489a2597962aeb670e0c1798a16a7359c94f Reviewed-on: https://go-review.googlesource.com/c/go/+/258938 Trust: Austin Clements <austin@google.com> Run-TryBot: Austin Clements <austin@google.com> TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Cherry Zhang <cherryyz@google.com>
2020-10-01 17:22:38 -04:00
fmt.Fprintf(fgo2, "a.p%d", i)
})
fmt.Fprint(fgo2, ")\n")
runtime,cmd/cgo: simplify C -> Go call path This redesigns the way calls work from C to exported Go functions. It removes several steps from the call path, makes cmd/cgo no longer sensitive to the Go calling convention, and eliminates the use of reflectcall from cgo. In order to avoid generating a large amount of FFI glue between the C and Go ABIs, the cgo tool has long depended on generating a C function that marshals the arguments into a struct, and then the actual ABI switch happens in functions with fixed signatures that simply take a pointer to this struct. In a way, this CL simply pushes this idea further. Currently, the cgo tool generates this argument struct in the exact layout of the Go stack frame and depends on reflectcall to unpack it into the appropriate Go call (even though it's actually reflectcall'ing a function generated by cgo). In this CL, we decouple this struct from the Go stack layout. Instead, cgo generates a Go function that takes the struct, unpacks it, and calls the exported function. Since this generated function has a generic signature (like the rest of the call path), we don't need reflectcall and can instead depend on the Go compiler itself to implement the call to the exported Go function. One complication is that syscall.NewCallback on Windows, which converts a Go function into a C function pointer, depends on cgocallback's current dynamic calling approach since the signatures of the callbacks aren't known statically. For this specific case, we continue to depend on reflectcall. Really, the current approach makes some overly simplistic assumptions about translating the C ABI to the Go ABI. Now we're at least in a much better position to do a proper ABI translation. For comparison, the current cgo call path looks like: GoF (generated C function) -> crosscall2 (in cgo/asm_*.s) -> _cgoexp_GoF (generated Go function) -> cgocallback (in asm_*.s) -> cgocallback_gofunc (in asm_*.s) -> cgocallbackg (in cgocall.go) -> cgocallbackg1 (in cgocall.go) -> reflectcall (in asm_*.s) -> _cgoexpwrap_GoF (generated Go function) -> p.GoF Now the call path looks like: GoF (generated C function) -> crosscall2 (in cgo/asm_*.s) -> cgocallback (in asm_*.s) -> cgocallbackg (in cgocall.go) -> cgocallbackg1 (in cgocall.go) -> _cgoexp_GoF (generated Go function) -> p.GoF Notably: 1. We combine _cgoexp_GoF and _cgoexpwrap_GoF and move the combined operation to the end of the sequence. This combined function also handles reflectcall's previous role. 2. We combined cgocallback and cgocallback_gofunc since the only purpose of having both was to convert a raw PC into a Go function value. We instead construct the Go function value in cgocallbackg1. 3. cgocallbackg1 no longer reaches backwards through the stack to get the arguments to cgocallback_gofunc. Instead, we just pass the arguments down. 4. Currently, we need an explicit msanwrite to mark the results struct as written because reflectcall doesn't do this. Now, the results are written by regular Go assignments, so the Go compiler generates the necessary MSAN annotations. This also means we no longer need to track the size of the arguments frame. Updates #40724, since now we don't need to teach cgo about the register ABI or change how it uses reflectcall. Change-Id: I7840489a2597962aeb670e0c1798a16a7359c94f Reviewed-on: https://go-review.googlesource.com/c/go/+/258938 Trust: Austin Clements <austin@google.com> Run-TryBot: Austin Clements <austin@google.com> TryBot-Result: Go Bot <gobot@golang.org> Reviewed-by: Cherry Zhang <cherryyz@google.com>
2020-10-01 17:22:38 -04:00
if gccResult != "void" {
// Verify that any results don't contain any
// Go pointers.
forFieldList(fntype.Results,
func(i int, aname string, atype ast.Expr) {
if !p.hasPointer(nil, atype, false) {
return
}
fmt.Fprintf(fgo2, "\t_cgoCheckResult(a.r%d)\n", i)
})
}
fmt.Fprint(fgo2, "}\n")
}
fmt.Fprintf(fgcch, "%s", gccExportHeaderEpilog)
}
// Write out the C header allowing C code to call exported gccgo functions.
func (p *Package) writeGccgoExports(fgo2, fm, fgcc, fgcch io.Writer) {
gccgoSymbolPrefix := p.gccgoSymbolPrefix()
p.writeExportHeader(fgcch)
fmt.Fprintf(fgcc, "/* Code generated by cmd/cgo; DO NOT EDIT. */\n\n")
fmt.Fprintf(fgcc, "#include \"_cgo_export.h\"\n")
fmt.Fprintf(fgcc, "%s\n", gccgoExportFileProlog)
fmt.Fprintf(fgcc, "%s\n", tsanProlog)
fmt.Fprintf(fgcc, "%s\n", msanProlog)
for _, exp := range p.ExpFunc {
fn := exp.Func
fntype := fn.Type
cdeclBuf := new(strings.Builder)
resultCount := 0
forFieldList(fntype.Results,
func(i int, aname string, atype ast.Expr) { resultCount++ })
switch resultCount {
case 0:
fmt.Fprintf(cdeclBuf, "void")
case 1:
forFieldList(fntype.Results,
func(i int, aname string, atype ast.Expr) {
t := p.cgoType(atype)
fmt.Fprintf(cdeclBuf, "%s", t.C)
})
default:
// Declare a result struct.
fmt.Fprintf(fgcch, "\n/* Return type for %s */\n", exp.ExpName)
fmt.Fprintf(fgcch, "struct %s_return {\n", exp.ExpName)
forFieldList(fntype.Results,
func(i int, aname string, atype ast.Expr) {
t := p.cgoType(atype)
fmt.Fprintf(fgcch, "\t%s r%d;", t.C, i)
if len(aname) > 0 {
fmt.Fprintf(fgcch, " /* %s */", aname)
}
fmt.Fprint(fgcch, "\n")
})
fmt.Fprintf(fgcch, "};\n")
fmt.Fprintf(cdeclBuf, "struct %s_return", exp.ExpName)
}
cRet := cdeclBuf.String()
cdeclBuf = new(strings.Builder)
fmt.Fprintf(cdeclBuf, "(")
if fn.Recv != nil {
fmt.Fprintf(cdeclBuf, "%s recv", p.cgoType(fn.Recv.List[0].Type).C.String())
}
// Function parameters.
forFieldList(fntype.Params,
func(i int, aname string, atype ast.Expr) {
if i > 0 || fn.Recv != nil {
fmt.Fprintf(cdeclBuf, ", ")
}
t := p.cgoType(atype)
fmt.Fprintf(cdeclBuf, "%s p%d", t.C, i)
})
fmt.Fprintf(cdeclBuf, ")")
cParams := cdeclBuf.String()
if len(exp.Doc) > 0 {
fmt.Fprintf(fgcch, "\n%s", exp.Doc)
}
fmt.Fprintf(fgcch, "extern %s %s%s;\n", cRet, exp.ExpName, cParams)
// We need to use a name that will be exported by the
// Go code; otherwise gccgo will make it static and we
// will not be able to link against it from the C
// code.
goName := "Cgoexp_" + exp.ExpName
fmt.Fprintf(fgcc, `extern %s %s %s __asm__("%s.%s");`, cRet, goName, cParams, gccgoSymbolPrefix, gccgoToSymbol(goName))
fmt.Fprint(fgcc, "\n")
fmt.Fprint(fgcc, "\nCGO_NO_SANITIZE_THREAD\n")
fmt.Fprintf(fgcc, "%s %s %s {\n", cRet, exp.ExpName, cParams)
if resultCount > 0 {
fmt.Fprintf(fgcc, "\t%s r;\n", cRet)
}
fmt.Fprintf(fgcc, "\tif(_cgo_wait_runtime_init_done)\n")
fmt.Fprintf(fgcc, "\t\t_cgo_wait_runtime_init_done();\n")
fmt.Fprintf(fgcc, "\t_cgo_tsan_release();\n")
fmt.Fprint(fgcc, "\t")
if resultCount > 0 {
fmt.Fprint(fgcc, "r = ")
}
fmt.Fprintf(fgcc, "%s(", goName)
if fn.Recv != nil {
fmt.Fprint(fgcc, "recv")
}
forFieldList(fntype.Params,
func(i int, aname string, atype ast.Expr) {
if i > 0 || fn.Recv != nil {
fmt.Fprintf(fgcc, ", ")
}
fmt.Fprintf(fgcc, "p%d", i)
})
fmt.Fprint(fgcc, ");\n")
fmt.Fprintf(fgcc, "\t_cgo_tsan_acquire();\n")
if resultCount > 0 {
fmt.Fprint(fgcc, "\treturn r;\n")
}
fmt.Fprint(fgcc, "}\n")
// Dummy declaration for _cgo_main.c
fmt.Fprintf(fm, `char %s[1] __asm__("%s.%s");`, goName, gccgoSymbolPrefix, gccgoToSymbol(goName))
fmt.Fprint(fm, "\n")
// For gccgo we use a wrapper function in Go, in order
// to call CgocallBack and CgocallBackDone.
// This code uses printer.Fprint, not conf.Fprint,
// because we don't want //line comments in the middle
// of the function types.
fmt.Fprint(fgo2, "\n")
fmt.Fprintf(fgo2, "func %s(", goName)
if fn.Recv != nil {
fmt.Fprint(fgo2, "recv ")
printer.Fprint(fgo2, fset, fn.Recv.List[0].Type)
}
forFieldList(fntype.Params,
func(i int, aname string, atype ast.Expr) {
if i > 0 || fn.Recv != nil {
fmt.Fprintf(fgo2, ", ")
}
fmt.Fprintf(fgo2, "p%d ", i)
printer.Fprint(fgo2, fset, atype)
})
fmt.Fprintf(fgo2, ")")
if resultCount > 0 {
fmt.Fprintf(fgo2, " (")
forFieldList(fntype.Results,
func(i int, aname string, atype ast.Expr) {
if i > 0 {
fmt.Fprint(fgo2, ", ")
}
printer.Fprint(fgo2, fset, atype)
})
fmt.Fprint(fgo2, ")")
}
fmt.Fprint(fgo2, " {\n")
fmt.Fprint(fgo2, "\tsyscall.CgocallBack()\n")
fmt.Fprint(fgo2, "\tdefer syscall.CgocallBackDone()\n")
fmt.Fprint(fgo2, "\t")
if resultCount > 0 {
fmt.Fprint(fgo2, "return ")
}
if fn.Recv != nil {
fmt.Fprint(fgo2, "recv.")
}
fmt.Fprintf(fgo2, "%s(", exp.Func.Name)
forFieldList(fntype.Params,
func(i int, aname string, atype ast.Expr) {
if i > 0 {
fmt.Fprint(fgo2, ", ")
}
fmt.Fprintf(fgo2, "p%d", i)
})
fmt.Fprint(fgo2, ")\n")
fmt.Fprint(fgo2, "}\n")
}
fmt.Fprintf(fgcch, "%s", gccExportHeaderEpilog)
}
// writeExportHeader writes out the start of the _cgo_export.h file.
func (p *Package) writeExportHeader(fgcch io.Writer) {
fmt.Fprintf(fgcch, "/* Code generated by cmd/cgo; DO NOT EDIT. */\n\n")
pkg := *importPath
if pkg == "" {
pkg = p.PackagePath
}
fmt.Fprintf(fgcch, "/* package %s */\n\n", pkg)
fmt.Fprintf(fgcch, "%s\n", builtinExportProlog)
// Remove absolute paths from #line comments in the preamble.
// They aren't useful for people using the header file,
// and they mean that the header files change based on the
// exact location of GOPATH.
re := regexp.MustCompile(`(?m)^(#line\s+\d+\s+")[^"]*[/\\]([^"]*")`)
preamble := re.ReplaceAllString(p.Preamble, "$1$2")
fmt.Fprintf(fgcch, "/* Start of preamble from import \"C\" comments. */\n\n")
fmt.Fprintf(fgcch, "%s\n", preamble)
fmt.Fprintf(fgcch, "\n/* End of preamble from import \"C\" comments. */\n\n")
fmt.Fprintf(fgcch, "%s\n", p.gccExportHeaderProlog())
}
// gccgoToSymbol converts a name to a mangled symbol for gccgo.
func gccgoToSymbol(ppath string) string {
if gccgoMangler == nil {
var err error
cmd := os.Getenv("GCCGO")
if cmd == "" {
cmd, err = exec.LookPath("gccgo")
if err != nil {
fatalf("unable to locate gccgo: %v", err)
}
}
gccgoMangler, err = pkgpath.ToSymbolFunc(cmd, *objDir)
if err != nil {
fatalf("%v", err)
}
}
return gccgoMangler(ppath)
}
// Return the package prefix when using gccgo.
func (p *Package) gccgoSymbolPrefix() string {
if !*gccgo {
return ""
}
if *gccgopkgpath != "" {
return gccgoToSymbol(*gccgopkgpath)
}
if *gccgoprefix == "" && p.PackageName == "main" {
return "main"
}
prefix := gccgoToSymbol(*gccgoprefix)
if prefix == "" {
prefix = "go"
}
return prefix + "." + p.PackageName
}
// Call a function for each entry in an ast.FieldList, passing the
// index into the list, the name if any, and the type.
func forFieldList(fl *ast.FieldList, fn func(int, string, ast.Expr)) {
if fl == nil {
return
}
i := 0
for _, r := range fl.List {
if r.Names == nil {
fn(i, "", r.Type)
i++
} else {
for _, n := range r.Names {
fn(i, n.Name, r.Type)
i++
}
}
}
}
func c(repr string, args ...interface{}) *TypeRepr {
return &TypeRepr{repr, args}
}
// Map predeclared Go types to Type.
var goTypes = map[string]*Type{
"bool": {Size: 1, Align: 1, C: c("GoUint8")},
"byte": {Size: 1, Align: 1, C: c("GoUint8")},
"int": {Size: 0, Align: 0, C: c("GoInt")},
"uint": {Size: 0, Align: 0, C: c("GoUint")},
"rune": {Size: 4, Align: 4, C: c("GoInt32")},
"int8": {Size: 1, Align: 1, C: c("GoInt8")},
"uint8": {Size: 1, Align: 1, C: c("GoUint8")},
"int16": {Size: 2, Align: 2, C: c("GoInt16")},
"uint16": {Size: 2, Align: 2, C: c("GoUint16")},
"int32": {Size: 4, Align: 4, C: c("GoInt32")},
"uint32": {Size: 4, Align: 4, C: c("GoUint32")},
"int64": {Size: 8, Align: 8, C: c("GoInt64")},
"uint64": {Size: 8, Align: 8, C: c("GoUint64")},
"float32": {Size: 4, Align: 4, C: c("GoFloat32")},
"float64": {Size: 8, Align: 8, C: c("GoFloat64")},
"complex64": {Size: 8, Align: 4, C: c("GoComplex64")},
"complex128": {Size: 16, Align: 8, C: c("GoComplex128")},
}
// Map an ast type to a Type.
func (p *Package) cgoType(e ast.Expr) *Type {
return p.doCgoType(e, make(map[ast.Expr]bool))
}
// Map an ast type to a Type, avoiding cycles.
func (p *Package) doCgoType(e ast.Expr, m map[ast.Expr]bool) *Type {
if m[e] {
fatalf("%s: invalid recursive type", fset.Position(e.Pos()))
}
m[e] = true
switch t := e.(type) {
case *ast.StarExpr:
x := p.doCgoType(t.X, m)
return &Type{Size: p.PtrSize, Align: p.PtrSize, C: c("%s*", x.C)}
case *ast.ArrayType:
if t.Len == nil {
// Slice: pointer, len, cap.
return &Type{Size: p.PtrSize * 3, Align: p.PtrSize, C: c("GoSlice")}
}
// Non-slice array types are not supported.
case *ast.StructType:
// Not supported.
case *ast.FuncType:
return &Type{Size: p.PtrSize, Align: p.PtrSize, C: c("void*")}
case *ast.InterfaceType:
return &Type{Size: 2 * p.PtrSize, Align: p.PtrSize, C: c("GoInterface")}
case *ast.MapType:
return &Type{Size: p.PtrSize, Align: p.PtrSize, C: c("GoMap")}
case *ast.ChanType:
return &Type{Size: p.PtrSize, Align: p.PtrSize, C: c("GoChan")}
case *ast.Ident:
cmd/go: cgo export header to be compatible with MSVC complex types After CL 379474 has landed, the only remaining cgo export header incompatibility with MSVC is the use of the _Complex macro, which is not supported in MSVC even when it is part of the ISO C99 standard (1). Since MSVC 2015 (2), complex math are supported via _Fcomplex and _Dcomplex, which are equivalent to float _Complex and double _Complex. As MSVC and C complex types have the same memory layout, we should be able to typedef GoComplex64 and GoComplex128 to the appropriate type in MSVC. It is important to note that this CL is not adding MSVC support to cgo. C compilers should still be GCC-compatible. This CL is about allowing to include, without further modifications, a DLL export header generated by cgo, normally using Mingw-W64 compiler, into a MSVC project. This was already possible if the export header changes introduced in this CL were done outside cgo, either manually or in a post-build script. Fixes #36233 1: https://docs.microsoft.com/en-us/cpp/c-runtime-library/complex-math-support 2: https://docs.microsoft.com/en-us/cpp/overview/visual-cpp-language-conformance?c-standard-library-features-1 Change-Id: Iad8f26984b115c728e3b73f3a8334ade7a11cfa1 Reviewed-on: https://go-review.googlesource.com/c/go/+/397134 Reviewed-by: Ian Lance Taylor <iant@golang.org> Trust: Cherry Mui <cherryyz@google.com> Run-TryBot: Cherry Mui <cherryyz@google.com> Auto-Submit: Cherry Mui <cherryyz@google.com> TryBot-Result: Gopher Robot <gobot@golang.org>
2022-03-31 14:15:22 +02:00
goTypesFixup := func(r *Type) *Type {
if r.Size == 0 { // int or uint
rr := new(Type)
*rr = *r
rr.Size = p.IntSize
rr.Align = p.IntSize
r = rr
}
if r.Align > p.PtrSize {
r.Align = p.PtrSize
}
return r
}
// Look up the type in the top level declarations.
// TODO: Handle types defined within a function.
for _, d := range p.Decl {
gd, ok := d.(*ast.GenDecl)
if !ok || gd.Tok != token.TYPE {
continue
}
for _, spec := range gd.Specs {
ts, ok := spec.(*ast.TypeSpec)
if !ok {
continue
}
if ts.Name.Name == t.Name {
// Give a better error than the one
// above if we detect a recursive type.
if m[ts.Type] {
fatalf("%s: invalid recursive type: %s refers to itself", fset.Position(e.Pos()), t.Name)
}
return p.doCgoType(ts.Type, m)
}
}
}
if def := typedef[t.Name]; def != nil {
cmd/go: cgo export header to be compatible with MSVC complex types After CL 379474 has landed, the only remaining cgo export header incompatibility with MSVC is the use of the _Complex macro, which is not supported in MSVC even when it is part of the ISO C99 standard (1). Since MSVC 2015 (2), complex math are supported via _Fcomplex and _Dcomplex, which are equivalent to float _Complex and double _Complex. As MSVC and C complex types have the same memory layout, we should be able to typedef GoComplex64 and GoComplex128 to the appropriate type in MSVC. It is important to note that this CL is not adding MSVC support to cgo. C compilers should still be GCC-compatible. This CL is about allowing to include, without further modifications, a DLL export header generated by cgo, normally using Mingw-W64 compiler, into a MSVC project. This was already possible if the export header changes introduced in this CL were done outside cgo, either manually or in a post-build script. Fixes #36233 1: https://docs.microsoft.com/en-us/cpp/c-runtime-library/complex-math-support 2: https://docs.microsoft.com/en-us/cpp/overview/visual-cpp-language-conformance?c-standard-library-features-1 Change-Id: Iad8f26984b115c728e3b73f3a8334ade7a11cfa1 Reviewed-on: https://go-review.googlesource.com/c/go/+/397134 Reviewed-by: Ian Lance Taylor <iant@golang.org> Trust: Cherry Mui <cherryyz@google.com> Run-TryBot: Cherry Mui <cherryyz@google.com> Auto-Submit: Cherry Mui <cherryyz@google.com> TryBot-Result: Gopher Robot <gobot@golang.org>
2022-03-31 14:15:22 +02:00
if defgo, ok := def.Go.(*ast.Ident); ok {
switch defgo.Name {
case "complex64", "complex128":
// MSVC does not support the _Complex keyword
// nor the complex macro.
// Use GoComplex64 and GoComplex128 instead,
// which are typedef-ed to a compatible type.
// See go.dev/issues/36233.
return goTypesFixup(goTypes[defgo.Name])
}
}
return def
}
if t.Name == "uintptr" {
return &Type{Size: p.PtrSize, Align: p.PtrSize, C: c("GoUintptr")}
}
if t.Name == "string" {
// The string data is 1 pointer + 1 (pointer-sized) int.
return &Type{Size: 2 * p.PtrSize, Align: p.PtrSize, C: c("GoString")}
}
if t.Name == "error" {
return &Type{Size: 2 * p.PtrSize, Align: p.PtrSize, C: c("GoInterface")}
}
if r, ok := goTypes[t.Name]; ok {
cmd/go: cgo export header to be compatible with MSVC complex types After CL 379474 has landed, the only remaining cgo export header incompatibility with MSVC is the use of the _Complex macro, which is not supported in MSVC even when it is part of the ISO C99 standard (1). Since MSVC 2015 (2), complex math are supported via _Fcomplex and _Dcomplex, which are equivalent to float _Complex and double _Complex. As MSVC and C complex types have the same memory layout, we should be able to typedef GoComplex64 and GoComplex128 to the appropriate type in MSVC. It is important to note that this CL is not adding MSVC support to cgo. C compilers should still be GCC-compatible. This CL is about allowing to include, without further modifications, a DLL export header generated by cgo, normally using Mingw-W64 compiler, into a MSVC project. This was already possible if the export header changes introduced in this CL were done outside cgo, either manually or in a post-build script. Fixes #36233 1: https://docs.microsoft.com/en-us/cpp/c-runtime-library/complex-math-support 2: https://docs.microsoft.com/en-us/cpp/overview/visual-cpp-language-conformance?c-standard-library-features-1 Change-Id: Iad8f26984b115c728e3b73f3a8334ade7a11cfa1 Reviewed-on: https://go-review.googlesource.com/c/go/+/397134 Reviewed-by: Ian Lance Taylor <iant@golang.org> Trust: Cherry Mui <cherryyz@google.com> Run-TryBot: Cherry Mui <cherryyz@google.com> Auto-Submit: Cherry Mui <cherryyz@google.com> TryBot-Result: Gopher Robot <gobot@golang.org>
2022-03-31 14:15:22 +02:00
return goTypesFixup(r)
}
error_(e.Pos(), "unrecognized Go type %s", t.Name)
return &Type{Size: 4, Align: 4, C: c("int")}
case *ast.SelectorExpr:
id, ok := t.X.(*ast.Ident)
if ok && id.Name == "unsafe" && t.Sel.Name == "Pointer" {
return &Type{Size: p.PtrSize, Align: p.PtrSize, C: c("void*")}
}
}
error_(e.Pos(), "Go type not supported in export: %s", gofmt(e))
return &Type{Size: 4, Align: 4, C: c("int")}
}
const gccProlog = `
#line 1 "cgo-gcc-prolog"
/*
If x and y are not equal, the type will be invalid
(have a negative array count) and an inscrutable error will come
out of the compiler and hopefully mention "name".
*/
#define __cgo_compile_assert_eq(x, y, name) typedef char name[(x-y)*(x-y)*-2UL+1UL];
/* Check at compile time that the sizes we use match our expectations. */
#define __cgo_size_assert(t, n) __cgo_compile_assert_eq(sizeof(t), (size_t)n, _cgo_sizeof_##t##_is_not_##n)
__cgo_size_assert(char, 1)
__cgo_size_assert(short, 2)
__cgo_size_assert(int, 4)
typedef long long __cgo_long_long;
__cgo_size_assert(__cgo_long_long, 8)
__cgo_size_assert(float, 4)
__cgo_size_assert(double, 8)
extern char* _cgo_topofstack(void);
/*
We use packed structs, but they are always aligned.
The pragmas and address-of-packed-member are only recognized as warning
groups in clang 4.0+, so ignore unknown pragmas first.
*/
#pragma GCC diagnostic ignored "-Wunknown-pragmas"
#pragma GCC diagnostic ignored "-Wpragmas"
#pragma GCC diagnostic ignored "-Waddress-of-packed-member"
#pragma GCC diagnostic ignored "-Wunknown-warning-option"
#pragma GCC diagnostic ignored "-Wunaligned-access"
#include <errno.h>
#include <string.h>
`
// Prologue defining TSAN functions in C.
const noTsanProlog = `
#define CGO_NO_SANITIZE_THREAD
#define _cgo_tsan_acquire()
#define _cgo_tsan_release()
`
// This must match the TSAN code in runtime/cgo/libcgo.h.
// This is used when the code is built with the C/C++ Thread SANitizer,
// which is not the same as the Go race detector.
// __tsan_acquire tells TSAN that we are acquiring a lock on a variable,
// in this case _cgo_sync. __tsan_release releases the lock.
// (There is no actual lock, we are just telling TSAN that there is.)
//
// When we call from Go to C we call _cgo_tsan_acquire.
// When the C function returns we call _cgo_tsan_release.
// Similarly, when C calls back into Go we call _cgo_tsan_release
// and then call _cgo_tsan_acquire when we return to C.
// These calls tell TSAN that there is a serialization point at the C call.
//
// This is necessary because TSAN, which is a C/C++ tool, can not see
// the synchronization in the Go code. Without these calls, when
// multiple goroutines call into C code, TSAN does not understand
// that the calls are properly synchronized on the Go side.
//
// To be clear, if the calls are not properly synchronized on the Go side,
// we will be hiding races. But when using TSAN on mixed Go C/C++ code
// it is more important to avoid false positives, which reduce confidence
// in the tool, than to avoid false negatives.
const yesTsanProlog = `
#line 1 "cgo-tsan-prolog"
#define CGO_NO_SANITIZE_THREAD __attribute__ ((no_sanitize_thread))
long long _cgo_sync __attribute__ ((common));
extern void __tsan_acquire(void*);
extern void __tsan_release(void*);
__attribute__ ((unused))
static void _cgo_tsan_acquire() {
__tsan_acquire(&_cgo_sync);
}
__attribute__ ((unused))
static void _cgo_tsan_release() {
__tsan_release(&_cgo_sync);
}
`
// Set to yesTsanProlog if we see -fsanitize=thread in the flags for gcc.
var tsanProlog = noTsanProlog
// noMsanProlog is a prologue defining an MSAN function in C.
// This is used when not compiling with -fsanitize=memory.
const noMsanProlog = `
#define _cgo_msan_write(addr, sz)
`
// yesMsanProlog is a prologue defining an MSAN function in C.
// This is used when compiling with -fsanitize=memory.
// See the comment above where _cgo_msan_write is called.
const yesMsanProlog = `
extern void __msan_unpoison(const volatile void *, size_t);
#define _cgo_msan_write(addr, sz) __msan_unpoison((addr), (sz))
`
// msanProlog is set to yesMsanProlog if we see -fsanitize=memory in the flags
// for the C compiler.
var msanProlog = noMsanProlog
const builtinProlog = `
#line 1 "cgo-builtin-prolog"
#include <stddef.h>
/* Define intgo when compiling with GCC. */
typedef ptrdiff_t intgo;
#define GO_CGO_GOSTRING_TYPEDEF
typedef struct { const char *p; intgo n; } _GoString_;
typedef struct { char *p; intgo n; intgo c; } _GoBytes_;
_GoString_ GoString(char *p);
_GoString_ GoStringN(char *p, int l);
_GoBytes_ GoBytes(void *p, int n);
char *CString(_GoString_);
void *CBytes(_GoBytes_);
void *_CMalloc(size_t);
__attribute__ ((unused))
static size_t _GoStringLen(_GoString_ s) { return (size_t)s.n; }
__attribute__ ((unused))
static const char *_GoStringPtr(_GoString_ s) { return s.p; }
`
const goProlog = `
//go:linkname _cgo_runtime_cgocall runtime.cgocall
func _cgo_runtime_cgocall(unsafe.Pointer, uintptr) int32
//go:linkname _cgoCheckPointer runtime.cgoCheckPointer
//go:noescape
func _cgoCheckPointer(interface{}, interface{})
//go:linkname _cgoCheckResult runtime.cgoCheckResult
//go:noescape
func _cgoCheckResult(interface{})
`
const gccgoGoProlog = `
func _cgoCheckPointer(interface{}, interface{})
func _cgoCheckResult(interface{})
`
const goStringDef = `
//go:linkname _cgo_runtime_gostring runtime.gostring
func _cgo_runtime_gostring(*_Ctype_char) string
// GoString converts the C string p into a Go string.
func _Cfunc_GoString(p *_Ctype_char) string {
return _cgo_runtime_gostring(p)
}
`
const goStringNDef = `
//go:linkname _cgo_runtime_gostringn runtime.gostringn
func _cgo_runtime_gostringn(*_Ctype_char, int) string
// GoStringN converts the C data p with explicit length l to a Go string.
func _Cfunc_GoStringN(p *_Ctype_char, l _Ctype_int) string {
return _cgo_runtime_gostringn(p, int(l))
}
`
const goBytesDef = `
//go:linkname _cgo_runtime_gobytes runtime.gobytes
func _cgo_runtime_gobytes(unsafe.Pointer, int) []byte
// GoBytes converts the C data p with explicit length l to a Go []byte.
func _Cfunc_GoBytes(p unsafe.Pointer, l _Ctype_int) []byte {
return _cgo_runtime_gobytes(p, int(l))
}
`
const cStringDef = `
// CString converts the Go string s to a C string.
//
// The C string is allocated in the C heap using malloc.
// It is the caller's responsibility to arrange for it to be
// freed, such as by calling C.free (be sure to include stdlib.h
// if C.free is needed).
func _Cfunc_CString(s string) *_Ctype_char {
if len(s)+1 <= 0 {
panic("string too large")
}
p := _cgo_cmalloc(uint64(len(s)+1))
sliceHeader := struct {
p unsafe.Pointer
len int
cap int
}{p, len(s)+1, len(s)+1}
b := *(*[]byte)(unsafe.Pointer(&sliceHeader))
copy(b, s)
b[len(s)] = 0
return (*_Ctype_char)(p)
}
`
const cBytesDef = `
// CBytes converts the Go []byte slice b to a C array.
//
// The C array is allocated in the C heap using malloc.
// It is the caller's responsibility to arrange for it to be
// freed, such as by calling C.free (be sure to include stdlib.h
// if C.free is needed).
func _Cfunc_CBytes(b []byte) unsafe.Pointer {
p := _cgo_cmalloc(uint64(len(b)))
sliceHeader := struct {
p unsafe.Pointer
len int
cap int
}{p, len(b), len(b)}
s := *(*[]byte)(unsafe.Pointer(&sliceHeader))
copy(s, b)
return p
}
`
const cMallocDef = `
func _Cfunc__CMalloc(n _Ctype_size_t) unsafe.Pointer {
return _cgo_cmalloc(uint64(n))
}
`
var builtinDefs = map[string]string{
"GoString": goStringDef,
"GoStringN": goStringNDef,
"GoBytes": goBytesDef,
"CString": cStringDef,
"CBytes": cBytesDef,
"_CMalloc": cMallocDef,
}
// Definitions for C.malloc in Go and in C. We define it ourselves
// since we call it from functions we define, such as C.CString.
// Also, we have historically ensured that C.malloc does not return
// nil even for an allocation of 0.
const cMallocDefGo = `
//go:cgo_import_static _cgoPREFIX_Cfunc__Cmalloc
//go:linkname __cgofn__cgoPREFIX_Cfunc__Cmalloc _cgoPREFIX_Cfunc__Cmalloc
var __cgofn__cgoPREFIX_Cfunc__Cmalloc byte
var _cgoPREFIX_Cfunc__Cmalloc = unsafe.Pointer(&__cgofn__cgoPREFIX_Cfunc__Cmalloc)
//go:linkname runtime_throw runtime.throw
func runtime_throw(string)
//go:cgo_unsafe_args
func _cgo_cmalloc(p0 uint64) (r1 unsafe.Pointer) {
_cgo_runtime_cgocall(_cgoPREFIX_Cfunc__Cmalloc, uintptr(unsafe.Pointer(&p0)))
if r1 == nil {
runtime_throw("runtime: C malloc failed")
}
return
}
`
// cMallocDefC defines the C version of C.malloc for the gc compiler.
// It is defined here because C.CString and friends need a definition.
// We define it by hand, rather than simply inventing a reference to
// C.malloc, because <stdlib.h> may not have been included.
// This is approximately what writeOutputFunc would generate, but
// skips the cgo_topofstack code (which is only needed if the C code
// calls back into Go). This also avoids returning nil for an
// allocation of 0 bytes.
const cMallocDefC = `
CGO_NO_SANITIZE_THREAD
void _cgoPREFIX_Cfunc__Cmalloc(void *v) {
struct {
unsigned long long p0;
void *r1;
} PACKED *a = v;
void *ret;
_cgo_tsan_acquire();
ret = malloc(a->p0);
if (ret == 0 && a->p0 == 0) {
ret = malloc(1);
}
a->r1 = ret;
_cgo_tsan_release();
}
`
func (p *Package) cPrologGccgo() string {
r := strings.NewReplacer(
"PREFIX", cPrefix,
"GCCGOSYMBOLPREF", p.gccgoSymbolPrefix(),
"_cgoCheckPointer", gccgoToSymbol("_cgoCheckPointer"),
"_cgoCheckResult", gccgoToSymbol("_cgoCheckResult"))
return r.Replace(cPrologGccgo)
}
const cPrologGccgo = `
#line 1 "cgo-c-prolog-gccgo"
#include <stdint.h>
#include <stdlib.h>
#include <string.h>
typedef unsigned char byte;
typedef intptr_t intgo;
struct __go_string {
const unsigned char *__data;
intgo __length;
};
typedef struct __go_open_array {
void* __values;
intgo __count;
intgo __capacity;
} Slice;
struct __go_string __go_byte_array_to_string(const void* p, intgo len);
struct __go_open_array __go_string_to_byte_array (struct __go_string str);
extern void runtime_throw(const char *);
const char *_cgoPREFIX_Cfunc_CString(struct __go_string s) {
char *p = malloc(s.__length+1);
if(p == NULL)
runtime_throw("runtime: C malloc failed");
memmove(p, s.__data, s.__length);
p[s.__length] = 0;
return p;
}
void *_cgoPREFIX_Cfunc_CBytes(struct __go_open_array b) {
char *p = malloc(b.__count);
if(p == NULL)
runtime_throw("runtime: C malloc failed");
memmove(p, b.__values, b.__count);
return p;
}
struct __go_string _cgoPREFIX_Cfunc_GoString(char *p) {
intgo len = (p != NULL) ? strlen(p) : 0;
return __go_byte_array_to_string(p, len);
}
struct __go_string _cgoPREFIX_Cfunc_GoStringN(char *p, int32_t n) {
return __go_byte_array_to_string(p, n);
}
Slice _cgoPREFIX_Cfunc_GoBytes(char *p, int32_t n) {
struct __go_string s = { (const unsigned char *)p, n };
return __go_string_to_byte_array(s);
}
void *_cgoPREFIX_Cfunc__CMalloc(size_t n) {
void *p = malloc(n);
if(p == NULL && n == 0)
p = malloc(1);
if(p == NULL)
runtime_throw("runtime: C malloc failed");
return p;
}
struct __go_type_descriptor;
typedef struct __go_empty_interface {
const struct __go_type_descriptor *__type_descriptor;
void *__object;
} Eface;
extern void runtimeCgoCheckPointer(Eface, Eface)
__asm__("runtime.cgoCheckPointer")
__attribute__((weak));
extern void localCgoCheckPointer(Eface, Eface)
__asm__("GCCGOSYMBOLPREF._cgoCheckPointer");
void localCgoCheckPointer(Eface ptr, Eface arg) {
if(runtimeCgoCheckPointer) {
runtimeCgoCheckPointer(ptr, arg);
}
}
extern void runtimeCgoCheckResult(Eface)
__asm__("runtime.cgoCheckResult")
__attribute__((weak));
extern void localCgoCheckResult(Eface)
__asm__("GCCGOSYMBOLPREF._cgoCheckResult");
void localCgoCheckResult(Eface val) {
if(runtimeCgoCheckResult) {
runtimeCgoCheckResult(val);
}
}
`
// builtinExportProlog is a shorter version of builtinProlog,
// to be put into the _cgo_export.h file.
// For historical reasons we can't use builtinProlog in _cgo_export.h,
// because _cgo_export.h defines GoString as a struct while builtinProlog
// defines it as a function. We don't change this to avoid unnecessarily
// breaking existing code.
// The test of GO_CGO_GOSTRING_TYPEDEF avoids a duplicate definition
// error if a Go file with a cgo comment #include's the export header
// generated by a different package.
const builtinExportProlog = `
#line 1 "cgo-builtin-export-prolog"
#include <stddef.h>
#ifndef GO_CGO_EXPORT_PROLOGUE_H
#define GO_CGO_EXPORT_PROLOGUE_H
#ifndef GO_CGO_GOSTRING_TYPEDEF
typedef struct { const char *p; ptrdiff_t n; } _GoString_;
extern size_t _GoStringLen(_GoString_ s);
extern const char *_GoStringPtr(_GoString_ s);
#endif
#endif
`
func (p *Package) gccExportHeaderProlog() string {
return strings.Replace(gccExportHeaderProlog, "GOINTBITS", fmt.Sprint(8*p.IntSize), -1)
}
// gccExportHeaderProlog is written to the exported header, after the
// import "C" comment preamble but before the generated declarations
// of exported functions. This permits the generated declarations to
// use the type names that appear in goTypes, above.
//
// The test of GO_CGO_GOSTRING_TYPEDEF avoids a duplicate definition
// error if a Go file with a cgo comment #include's the export header
// generated by a different package. Unfortunately GoString means two
// different things: in this prolog it means a C name for the Go type,
// while in the prolog written into the start of the C code generated
// from a cgo-using Go file it means the C.GoString function. There is
// no way to resolve this conflict, but it also doesn't make much
// difference, as Go code never wants to refer to the latter meaning.
const gccExportHeaderProlog = `
/* Start of boilerplate cgo prologue. */
#line 1 "cgo-gcc-export-header-prolog"
#ifndef GO_CGO_PROLOGUE_H
#define GO_CGO_PROLOGUE_H
typedef signed char GoInt8;
typedef unsigned char GoUint8;
typedef short GoInt16;
typedef unsigned short GoUint16;
typedef int GoInt32;
typedef unsigned int GoUint32;
typedef long long GoInt64;
typedef unsigned long long GoUint64;
typedef GoIntGOINTBITS GoInt;
typedef GoUintGOINTBITS GoUint;
typedef size_t GoUintptr;
typedef float GoFloat32;
typedef double GoFloat64;
cmd/go: cgo export header to be compatible with MSVC complex types After CL 379474 has landed, the only remaining cgo export header incompatibility with MSVC is the use of the _Complex macro, which is not supported in MSVC even when it is part of the ISO C99 standard (1). Since MSVC 2015 (2), complex math are supported via _Fcomplex and _Dcomplex, which are equivalent to float _Complex and double _Complex. As MSVC and C complex types have the same memory layout, we should be able to typedef GoComplex64 and GoComplex128 to the appropriate type in MSVC. It is important to note that this CL is not adding MSVC support to cgo. C compilers should still be GCC-compatible. This CL is about allowing to include, without further modifications, a DLL export header generated by cgo, normally using Mingw-W64 compiler, into a MSVC project. This was already possible if the export header changes introduced in this CL were done outside cgo, either manually or in a post-build script. Fixes #36233 1: https://docs.microsoft.com/en-us/cpp/c-runtime-library/complex-math-support 2: https://docs.microsoft.com/en-us/cpp/overview/visual-cpp-language-conformance?c-standard-library-features-1 Change-Id: Iad8f26984b115c728e3b73f3a8334ade7a11cfa1 Reviewed-on: https://go-review.googlesource.com/c/go/+/397134 Reviewed-by: Ian Lance Taylor <iant@golang.org> Trust: Cherry Mui <cherryyz@google.com> Run-TryBot: Cherry Mui <cherryyz@google.com> Auto-Submit: Cherry Mui <cherryyz@google.com> TryBot-Result: Gopher Robot <gobot@golang.org>
2022-03-31 14:15:22 +02:00
#ifdef _MSC_VER
#if !defined(__cplusplus) || _MSVC_LANG <= 201402L
cmd/go: cgo export header to be compatible with MSVC complex types After CL 379474 has landed, the only remaining cgo export header incompatibility with MSVC is the use of the _Complex macro, which is not supported in MSVC even when it is part of the ISO C99 standard (1). Since MSVC 2015 (2), complex math are supported via _Fcomplex and _Dcomplex, which are equivalent to float _Complex and double _Complex. As MSVC and C complex types have the same memory layout, we should be able to typedef GoComplex64 and GoComplex128 to the appropriate type in MSVC. It is important to note that this CL is not adding MSVC support to cgo. C compilers should still be GCC-compatible. This CL is about allowing to include, without further modifications, a DLL export header generated by cgo, normally using Mingw-W64 compiler, into a MSVC project. This was already possible if the export header changes introduced in this CL were done outside cgo, either manually or in a post-build script. Fixes #36233 1: https://docs.microsoft.com/en-us/cpp/c-runtime-library/complex-math-support 2: https://docs.microsoft.com/en-us/cpp/overview/visual-cpp-language-conformance?c-standard-library-features-1 Change-Id: Iad8f26984b115c728e3b73f3a8334ade7a11cfa1 Reviewed-on: https://go-review.googlesource.com/c/go/+/397134 Reviewed-by: Ian Lance Taylor <iant@golang.org> Trust: Cherry Mui <cherryyz@google.com> Run-TryBot: Cherry Mui <cherryyz@google.com> Auto-Submit: Cherry Mui <cherryyz@google.com> TryBot-Result: Gopher Robot <gobot@golang.org>
2022-03-31 14:15:22 +02:00
#include <complex.h>
typedef _Fcomplex GoComplex64;
typedef _Dcomplex GoComplex128;
#else
#include <complex>
typedef std::complex<float> GoComplex64;
typedef std::complex<double> GoComplex128;
#endif
#else
typedef float _Complex GoComplex64;
typedef double _Complex GoComplex128;
cmd/go: cgo export header to be compatible with MSVC complex types After CL 379474 has landed, the only remaining cgo export header incompatibility with MSVC is the use of the _Complex macro, which is not supported in MSVC even when it is part of the ISO C99 standard (1). Since MSVC 2015 (2), complex math are supported via _Fcomplex and _Dcomplex, which are equivalent to float _Complex and double _Complex. As MSVC and C complex types have the same memory layout, we should be able to typedef GoComplex64 and GoComplex128 to the appropriate type in MSVC. It is important to note that this CL is not adding MSVC support to cgo. C compilers should still be GCC-compatible. This CL is about allowing to include, without further modifications, a DLL export header generated by cgo, normally using Mingw-W64 compiler, into a MSVC project. This was already possible if the export header changes introduced in this CL were done outside cgo, either manually or in a post-build script. Fixes #36233 1: https://docs.microsoft.com/en-us/cpp/c-runtime-library/complex-math-support 2: https://docs.microsoft.com/en-us/cpp/overview/visual-cpp-language-conformance?c-standard-library-features-1 Change-Id: Iad8f26984b115c728e3b73f3a8334ade7a11cfa1 Reviewed-on: https://go-review.googlesource.com/c/go/+/397134 Reviewed-by: Ian Lance Taylor <iant@golang.org> Trust: Cherry Mui <cherryyz@google.com> Run-TryBot: Cherry Mui <cherryyz@google.com> Auto-Submit: Cherry Mui <cherryyz@google.com> TryBot-Result: Gopher Robot <gobot@golang.org>
2022-03-31 14:15:22 +02:00
#endif
/*
static assertion to make sure the file is being used on architecture
at least with matching size of GoInt.
*/
typedef char _check_for_GOINTBITS_bit_pointer_matching_GoInt[sizeof(void*)==GOINTBITS/8 ? 1:-1];
#ifndef GO_CGO_GOSTRING_TYPEDEF
typedef _GoString_ GoString;
#endif
typedef void *GoMap;
typedef void *GoChan;
typedef struct { void *t; void *v; } GoInterface;
typedef struct { void *data; GoInt len; GoInt cap; } GoSlice;
#endif
/* End of boilerplate cgo prologue. */
#ifdef __cplusplus
extern "C" {
#endif
`
// gccExportHeaderEpilog goes at the end of the generated header file.
const gccExportHeaderEpilog = `
#ifdef __cplusplus
}
#endif
`
// gccgoExportFileProlog is written to the _cgo_export.c file when
// using gccgo.
// We use weak declarations, and test the addresses, so that this code
// works with older versions of gccgo.
const gccgoExportFileProlog = `
#line 1 "cgo-gccgo-export-file-prolog"
extern _Bool runtime_iscgo __attribute__ ((weak));
static void GoInit(void) __attribute__ ((constructor));
static void GoInit(void) {
if(&runtime_iscgo)
runtime_iscgo = 1;
}
extern size_t _cgo_wait_runtime_init_done(void) __attribute__ ((weak));
`