1
0
mirror of https://github.com/golang/go synced 2024-11-18 04:14:49 -07:00
go/test/run.go

2079 lines
58 KiB
Go
Raw Normal View History

// skip
// Copyright 2012 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.
// Run runs tests in the test directory.
package main
import (
"bytes"
"encoding/json"
"errors"
"flag"
"fmt"
"go/build"
"go/build/constraint"
"hash/fnv"
"io"
"io/fs"
"io/ioutil"
"log"
"os"
"os/exec"
"path"
"path/filepath"
"regexp"
"runtime"
"sort"
"strconv"
"strings"
"time"
"unicode"
)
var (
verbose = flag.Bool("v", false, "verbose. if set, parallelism is set to 1.")
keep = flag.Bool("k", false, "keep. keep temporary directory.")
numParallel = flag.Int("n", runtime.NumCPU(), "number of parallel tests to run")
summary = flag.Bool("summary", false, "show summary of results")
allCodegen = flag.Bool("all_codegen", defaultAllCodeGen(), "run all goos/goarch for codegen")
showSkips = flag.Bool("show_skips", false, "show skipped tests")
runSkips = flag.Bool("run_skips", false, "run skipped tests (ignore skip and build tags)")
linkshared = flag.Bool("linkshared", false, "")
updateErrors = flag.Bool("update_errors", false, "update error messages in test file based on compiler output")
runoutputLimit = flag.Int("l", defaultRunOutputLimit(), "number of parallel runoutput tests to run")
force = flag.Bool("f", false, "ignore expected-failure test lists")
shard = flag.Int("shard", 0, "shard index to run. Only applicable if -shards is non-zero.")
shards = flag.Int("shards", 0, "number of shards. If 0, all tests are run. This is used by the continuous build.")
)
type envVars struct {
GOOS string
GOARCH string
GOEXPERIMENT string
CGO_ENABLED string
}
var env = func() (res envVars) {
cmd := exec.Command("go", "env", "-json")
stdout, err := cmd.StdoutPipe()
if err != nil {
log.Fatal("StdoutPipe:", err)
}
if err := cmd.Start(); err != nil {
log.Fatal("Start:", err)
}
if err := json.NewDecoder(stdout).Decode(&res); err != nil {
log.Fatal("Decode:", err)
}
if err := cmd.Wait(); err != nil {
log.Fatal("Wait:", err)
}
return
}()
// TODO(mdempsky): This will give false negatives if the unified
// experiment is enabled by default, but presumably at that point we
// won't need to disable tests for it anymore anyway.
var unifiedEnabled = strings.Contains(","+env.GOEXPERIMENT+",", ",unified,")
// defaultAllCodeGen returns the default value of the -all_codegen
// flag. By default, we prefer to be fast (returning false), except on
// the linux-amd64 builder that's already very fast, so we get more
// test coverage on trybots. See https://golang.org/issue/34297.
func defaultAllCodeGen() bool {
return os.Getenv("GO_BUILDER_NAME") == "linux-amd64"
}
var (
goos = env.GOOS
goarch = env.GOARCH
cgoEnabled, _ = strconv.ParseBool(env.CGO_ENABLED)
// dirs are the directories to look for *.go files in.
// TODO(bradfitz): just use all directories?
dirs = []string{".", "ken", "chan", "interface", "syntax", "dwarf", "fixedbugs", "codegen", "runtime", "abi", "typeparam", "typeparam/mdempsky"}
// ratec controls the max number of tests running at a time.
ratec chan bool
// toRun is the channel of tests to run.
// It is nil until the first test is started.
toRun chan *test
// rungatec controls the max number of runoutput tests
// executed in parallel as they can each consume a lot of memory.
rungatec chan bool
)
// maxTests is an upper bound on the total number of tests.
// It is used as a channel buffer size to make sure sends don't block.
const maxTests = 5000
func main() {
flag.Parse()
findExecCmd()
// Disable parallelism if printing or if using a simulator.
if *verbose || len(findExecCmd()) > 0 {
*numParallel = 1
*runoutputLimit = 1
}
ratec = make(chan bool, *numParallel)
rungatec = make(chan bool, *runoutputLimit)
var tests []*test
if flag.NArg() > 0 {
for _, arg := range flag.Args() {
if arg == "-" || arg == "--" {
// Permit running:
// $ go run run.go - env.go
// $ go run run.go -- env.go
// $ go run run.go - ./fixedbugs
// $ go run run.go -- ./fixedbugs
continue
}
if fi, err := os.Stat(arg); err == nil && fi.IsDir() {
for _, baseGoFile := range goFiles(arg) {
tests = append(tests, startTest(arg, baseGoFile))
}
} else if strings.HasSuffix(arg, ".go") {
dir, file := filepath.Split(arg)
tests = append(tests, startTest(dir, file))
} else {
log.Fatalf("can't yet deal with non-directory and non-go file %q", arg)
}
}
} else {
for _, dir := range dirs {
for _, baseGoFile := range goFiles(dir) {
tests = append(tests, startTest(dir, baseGoFile))
}
}
}
failed := false
resCount := map[string]int{}
for _, test := range tests {
<-test.donec
status := "ok "
errStr := ""
if e, isSkip := test.err.(skipError); isSkip {
test.err = nil
errStr = "unexpected skip for " + path.Join(test.dir, test.gofile) + ": " + string(e)
status = "FAIL"
}
if test.err != nil {
errStr = test.err.Error()
if test.expectFail {
errStr += " (expected)"
} else {
status = "FAIL"
}
} else if test.expectFail {
status = "FAIL"
errStr = "unexpected success"
}
if status == "FAIL" {
failed = true
}
resCount[status]++
dt := fmt.Sprintf("%.3fs", test.dt.Seconds())
if status == "FAIL" {
cmd/compile: require -p flag The -p flag specifies the import path of the package being compiled. This CL makes it required when invoking the compiler and adjusts tests that invoke the compiler directly to conform to this new requirement. The go command already passes the flag, so it is unmodified in this CL. It is expected that any other Go build systems also already pass -p, or else they will need to arrange to do so before updating to Go 1.19. Of particular note, Bazel already does for rules with an importpath= attribute, which includes all Gazelle-generated rules. There is more cleanup possible now in cmd/compile, cmd/link, and other consumers of Go object files, but that is left to future CLs. Additional historical background follows but can be ignored. Long ago, before the go command, or modules, or any kind of versioning, symbols in Go archive files were named using just the package name, so that for example func F in math/rand and func F in crypto/rand would both be the object file symbol 'rand.F'. This led to collisions even in small source trees, which made certain packages unusable in the presence of other packages and generally was a problem for Go's goal of scaling to very large source trees. Fixing this problem required changing from package names to import paths in symbol names, which was mostly straightforward. One wrinkle, though, is that the compiler did not know the import path of the package being compiled; it only knew the package name. At the time, there was no go command, just Makefiles that people had invoking 6g (now “go tool compile”) and then copying the resulting object file to an importable location. That is, everyone had a custom build setup for Go, because there was no standard one. So it was not particularly attractive to change how the compiler was invoked, since that would break approximately every Go user at the time. Instead, we arranged for the compiler to emit, and other tools reading object files to recognize, a special import path (the empty string, it turned out) denoting “the import path of this object file”. This worked well enough at the time and maintained complete command-line compatibility with existing Go usage. The changes implementing this transition can be found by searching the Git history for “package global name space”, which is what they eliminated. In particular, CL 190076 (a6736fa4), CL 186263 (758f2bc5), CL 193080 (1cecac81), CL 194053 (19126320), and CL 194071 (531e6b77) did the bulk of this transformation in January 2010. Later, in September 2011, we added the -p flag to the compiler for diagnostic purposes. The problem was that it was easy to create import cycles, especially in tests, and these could not be diagnosed until link time. You'd really want the compiler to diagnose these, for example if the compilation of package sort noticed it was importing a package that itself imported "sort". But the compilation of package sort didn't know its own import path, and so it could not tell whether it had found itself as a transitive dependency. Adding the -p flag solved this problem, and its use was optional, since the linker would still diagnose the import cycle in builds that had not updated to start passing -p. This was CL 4972057 (1e480cd1). There was still no go command at this point, but when we introduced the go command we made it pass -p, which it has for many years at this point. Over time, parts of the compiler began to depend on the presence of the -p flag for various reasonable purposes. For example: In CL 6497074 (041fc8bf; Oct 2012), the race detector used -p to detect packages that should not have race annotations, such as runtime/race and sync/atomic. In CL 13367052 (7276c02b; Sep 2013), a bug fix used -p to detect the compilation of package reflect. In CL 30539 (8aadcc55; Oct 2016), the compiler started using -p to identify package math, to be able to intrinsify calls to Sqrt inside that package. In CL 61019 (9daee931; Sep 2017), CL 71430 (2c1d2e06; Oct 2017), and later related CLs, the compiler started using the -p value when creating various DWARF debugging information. In CL 174657 (cc5eaf93; May 2019), the compiler started writing symbols without the magic empty string whenever -p was used, to reduce the amount of work required in the linker. In CL 179861 (dde7c770; Jun 2019), the compiler made the second argument to //go:linkname optional when -p is used, because in that case the compiler can derive an appropriate default. There are more examples. Today it is impossible to compile the Go standard library without using -p, and DWARF debug information is incomplete without using -p. All known Go build systems pass -p. In particular, the go command does, which is what nearly all Go developers invoke to build Go code. And Bazel does, for go_library rules that set the importpath attribute, which is all rules generated by Gazelle. Gccgo has an equivalent of -p and has required its use in order to disambiguate packages with the same name but different import paths since 2010. On top of all this, various parts of code generation for generics are made more complicated by needing to cope with the case where -p is not specified, even though it's essentially always specified. In summary, the current state is: - Use of the -p flag with cmd/compile is required for building the standard library, and for complete DWARF information, and to enable certain linker speedups. - The go command and Bazel, which we expect account for just about 100% of Go builds, both invoke cmd/compile with -p. - The code in cmd/compile to support builds without -p is complex and has become more complex with generics, but it is almost always dead code and therefore not worth maintaining. - Gccgo already requires its equivalent of -p in any build where two packages have the same name. All this supports the change in this CL, which makes -p required and adjusts tests that invoke cmd/compile to add -p appropriately. Future CLs will be able to remove all the code dealing with the possibility of -p not having been specified. Change-Id: I6b95b9d4cffe59c7bac82eb273ef6c4a67bb0e43 Reviewed-on: https://go-review.googlesource.com/c/go/+/391014 Trust: Russ Cox <rsc@golang.org> Run-TryBot: Russ Cox <rsc@golang.org> TryBot-Result: Gopher Robot <gobot@golang.org> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2022-03-08 16:16:35 -07:00
fmt.Printf("# go run run.go -- %s\n%s\nFAIL\t%s\t%s\n",
path.Join(test.dir, test.gofile),
errStr, test.goFileName(), dt)
continue
}
if !*verbose {
continue
}
fmt.Printf("%s\t%s\t%s\n", status, test.goFileName(), dt)
}
if *summary {
for k, v := range resCount {
fmt.Printf("%5d %s\n", v, k)
}
}
if failed {
os.Exit(1)
}
}
// goTool reports the path of the go tool to use to run the tests.
// If possible, use the same Go used to run run.go, otherwise
// fallback to the go version found in the PATH.
func goTool() string {
var exeSuffix string
if runtime.GOOS == "windows" {
exeSuffix = ".exe"
}
path := filepath.Join(runtime.GOROOT(), "bin", "go"+exeSuffix)
if _, err := os.Stat(path); err == nil {
return path
}
// Just run "go" from PATH
return "go"
}
func shardMatch(name string) bool {
if *shards == 0 {
return true
}
h := fnv.New32()
io.WriteString(h, name)
return int(h.Sum32()%uint32(*shards)) == *shard
}
func goFiles(dir string) []string {
f, err := os.Open(dir)
if err != nil {
log.Fatal(err)
}
dirnames, err := f.Readdirnames(-1)
f.Close()
if err != nil {
log.Fatal(err)
}
names := []string{}
for _, name := range dirnames {
if !strings.HasPrefix(name, ".") && strings.HasSuffix(name, ".go") && shardMatch(name) {
names = append(names, name)
}
}
sort.Strings(names)
return names
}
type runCmd func(...string) ([]byte, error)
func compileFile(runcmd runCmd, longname string, flags []string) (out []byte, err error) {
cmd/compile: require -p flag The -p flag specifies the import path of the package being compiled. This CL makes it required when invoking the compiler and adjusts tests that invoke the compiler directly to conform to this new requirement. The go command already passes the flag, so it is unmodified in this CL. It is expected that any other Go build systems also already pass -p, or else they will need to arrange to do so before updating to Go 1.19. Of particular note, Bazel already does for rules with an importpath= attribute, which includes all Gazelle-generated rules. There is more cleanup possible now in cmd/compile, cmd/link, and other consumers of Go object files, but that is left to future CLs. Additional historical background follows but can be ignored. Long ago, before the go command, or modules, or any kind of versioning, symbols in Go archive files were named using just the package name, so that for example func F in math/rand and func F in crypto/rand would both be the object file symbol 'rand.F'. This led to collisions even in small source trees, which made certain packages unusable in the presence of other packages and generally was a problem for Go's goal of scaling to very large source trees. Fixing this problem required changing from package names to import paths in symbol names, which was mostly straightforward. One wrinkle, though, is that the compiler did not know the import path of the package being compiled; it only knew the package name. At the time, there was no go command, just Makefiles that people had invoking 6g (now “go tool compile”) and then copying the resulting object file to an importable location. That is, everyone had a custom build setup for Go, because there was no standard one. So it was not particularly attractive to change how the compiler was invoked, since that would break approximately every Go user at the time. Instead, we arranged for the compiler to emit, and other tools reading object files to recognize, a special import path (the empty string, it turned out) denoting “the import path of this object file”. This worked well enough at the time and maintained complete command-line compatibility with existing Go usage. The changes implementing this transition can be found by searching the Git history for “package global name space”, which is what they eliminated. In particular, CL 190076 (a6736fa4), CL 186263 (758f2bc5), CL 193080 (1cecac81), CL 194053 (19126320), and CL 194071 (531e6b77) did the bulk of this transformation in January 2010. Later, in September 2011, we added the -p flag to the compiler for diagnostic purposes. The problem was that it was easy to create import cycles, especially in tests, and these could not be diagnosed until link time. You'd really want the compiler to diagnose these, for example if the compilation of package sort noticed it was importing a package that itself imported "sort". But the compilation of package sort didn't know its own import path, and so it could not tell whether it had found itself as a transitive dependency. Adding the -p flag solved this problem, and its use was optional, since the linker would still diagnose the import cycle in builds that had not updated to start passing -p. This was CL 4972057 (1e480cd1). There was still no go command at this point, but when we introduced the go command we made it pass -p, which it has for many years at this point. Over time, parts of the compiler began to depend on the presence of the -p flag for various reasonable purposes. For example: In CL 6497074 (041fc8bf; Oct 2012), the race detector used -p to detect packages that should not have race annotations, such as runtime/race and sync/atomic. In CL 13367052 (7276c02b; Sep 2013), a bug fix used -p to detect the compilation of package reflect. In CL 30539 (8aadcc55; Oct 2016), the compiler started using -p to identify package math, to be able to intrinsify calls to Sqrt inside that package. In CL 61019 (9daee931; Sep 2017), CL 71430 (2c1d2e06; Oct 2017), and later related CLs, the compiler started using the -p value when creating various DWARF debugging information. In CL 174657 (cc5eaf93; May 2019), the compiler started writing symbols without the magic empty string whenever -p was used, to reduce the amount of work required in the linker. In CL 179861 (dde7c770; Jun 2019), the compiler made the second argument to //go:linkname optional when -p is used, because in that case the compiler can derive an appropriate default. There are more examples. Today it is impossible to compile the Go standard library without using -p, and DWARF debug information is incomplete without using -p. All known Go build systems pass -p. In particular, the go command does, which is what nearly all Go developers invoke to build Go code. And Bazel does, for go_library rules that set the importpath attribute, which is all rules generated by Gazelle. Gccgo has an equivalent of -p and has required its use in order to disambiguate packages with the same name but different import paths since 2010. On top of all this, various parts of code generation for generics are made more complicated by needing to cope with the case where -p is not specified, even though it's essentially always specified. In summary, the current state is: - Use of the -p flag with cmd/compile is required for building the standard library, and for complete DWARF information, and to enable certain linker speedups. - The go command and Bazel, which we expect account for just about 100% of Go builds, both invoke cmd/compile with -p. - The code in cmd/compile to support builds without -p is complex and has become more complex with generics, but it is almost always dead code and therefore not worth maintaining. - Gccgo already requires its equivalent of -p in any build where two packages have the same name. All this supports the change in this CL, which makes -p required and adjusts tests that invoke cmd/compile to add -p appropriately. Future CLs will be able to remove all the code dealing with the possibility of -p not having been specified. Change-Id: I6b95b9d4cffe59c7bac82eb273ef6c4a67bb0e43 Reviewed-on: https://go-review.googlesource.com/c/go/+/391014 Trust: Russ Cox <rsc@golang.org> Run-TryBot: Russ Cox <rsc@golang.org> TryBot-Result: Gopher Robot <gobot@golang.org> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2022-03-08 16:16:35 -07:00
cmd := []string{goTool(), "tool", "compile", "-e", "-p=p"}
cmd = append(cmd, flags...)
if *linkshared {
cmd = append(cmd, "-dynlink", "-installsuffix=dynlink")
}
cmd = append(cmd, longname)
return runcmd(cmd...)
}
func compileInDir(runcmd runCmd, dir string, flags []string, pkgname string, names ...string) (out []byte, err error) {
cmd := []string{goTool(), "tool", "compile", "-e", "-D", "test", "-I", "."}
if pkgname == "main" {
cmd = append(cmd, "-p=main")
} else {
pkgname = path.Join("test", strings.TrimSuffix(names[0], ".go"))
cmd = append(cmd, "-o", pkgname+".a", "-p", pkgname)
}
cmd = append(cmd, flags...)
if *linkshared {
cmd = append(cmd, "-dynlink", "-installsuffix=dynlink")
}
for _, name := range names {
cmd = append(cmd, filepath.Join(dir, name))
}
return runcmd(cmd...)
}
func linkFile(runcmd runCmd, goname string, ldflags []string) (err error) {
pfile := strings.Replace(goname, ".go", ".o", -1)
cmd := []string{goTool(), "tool", "link", "-w", "-o", "a.exe", "-L", "."}
if *linkshared {
cmd = append(cmd, "-linkshared", "-installsuffix=dynlink")
}
if ldflags != nil {
cmd = append(cmd, ldflags...)
}
cmd = append(cmd, pfile)
_, err = runcmd(cmd...)
return
}
// skipError describes why a test was skipped.
type skipError string
func (s skipError) Error() string { return string(s) }
// test holds the state of a test.
type test struct {
dir, gofile string
donec chan bool // closed when done
dt time.Duration
src string
tempDir string
err error
// expectFail indicates whether the (overall) test recipe is
// expected to fail under the current test configuration (e.g.,
// GOEXPERIMENT=unified).
expectFail bool
}
// initExpectFail initializes t.expectFail based on the build+test
// configuration.
func (t *test) initExpectFail() {
if *force {
return
}
failureSets := []map[string]bool{types2Failures}
// Note: gccgo supports more 32-bit architectures than this, but
// hopefully the 32-bit failures are fixed before this matters.
switch goarch {
case "386", "arm", "mips", "mipsle":
failureSets = append(failureSets, types2Failures32Bit)
}
if unifiedEnabled {
failureSets = append(failureSets, unifiedFailures)
} else {
failureSets = append(failureSets, go118Failures)
}
filename := strings.Replace(t.goFileName(), "\\", "/", -1) // goFileName() uses \ on Windows
for _, set := range failureSets {
if set[filename] {
t.expectFail = true
return
}
}
}
func startTest(dir, gofile string) *test {
t := &test{
dir: dir,
gofile: gofile,
donec: make(chan bool, 1),
}
if toRun == nil {
toRun = make(chan *test, maxTests)
go runTests()
}
select {
case toRun <- t:
default:
panic("toRun buffer size (maxTests) is too small")
}
return t
}
// runTests runs tests in parallel, but respecting the order they
// were enqueued on the toRun channel.
func runTests() {
for {
ratec <- true
t := <-toRun
go func() {
t.run()
<-ratec
}()
}
}
var cwd, _ = os.Getwd()
func (t *test) goFileName() string {
return filepath.Join(t.dir, t.gofile)
}
func (t *test) goDirName() string {
return filepath.Join(t.dir, strings.Replace(t.gofile, ".go", ".dir", -1))
}
func goDirFiles(longdir string) (filter []os.FileInfo, err error) {
files, dirErr := ioutil.ReadDir(longdir)
if dirErr != nil {
return nil, dirErr
}
for _, gofile := range files {
if filepath.Ext(gofile.Name()) == ".go" {
filter = append(filter, gofile)
}
}
return
}
var packageRE = regexp.MustCompile(`(?m)^package ([\p{Lu}\p{Ll}\w]+)`)
func getPackageNameFromSource(fn string) (string, error) {
data, err := ioutil.ReadFile(fn)
if err != nil {
return "", err
}
pkgname := packageRE.FindStringSubmatch(string(data))
if pkgname == nil {
return "", fmt.Errorf("cannot find package name in %s", fn)
}
return pkgname[1], nil
}
cmd/compile: require -p flag The -p flag specifies the import path of the package being compiled. This CL makes it required when invoking the compiler and adjusts tests that invoke the compiler directly to conform to this new requirement. The go command already passes the flag, so it is unmodified in this CL. It is expected that any other Go build systems also already pass -p, or else they will need to arrange to do so before updating to Go 1.19. Of particular note, Bazel already does for rules with an importpath= attribute, which includes all Gazelle-generated rules. There is more cleanup possible now in cmd/compile, cmd/link, and other consumers of Go object files, but that is left to future CLs. Additional historical background follows but can be ignored. Long ago, before the go command, or modules, or any kind of versioning, symbols in Go archive files were named using just the package name, so that for example func F in math/rand and func F in crypto/rand would both be the object file symbol 'rand.F'. This led to collisions even in small source trees, which made certain packages unusable in the presence of other packages and generally was a problem for Go's goal of scaling to very large source trees. Fixing this problem required changing from package names to import paths in symbol names, which was mostly straightforward. One wrinkle, though, is that the compiler did not know the import path of the package being compiled; it only knew the package name. At the time, there was no go command, just Makefiles that people had invoking 6g (now “go tool compile”) and then copying the resulting object file to an importable location. That is, everyone had a custom build setup for Go, because there was no standard one. So it was not particularly attractive to change how the compiler was invoked, since that would break approximately every Go user at the time. Instead, we arranged for the compiler to emit, and other tools reading object files to recognize, a special import path (the empty string, it turned out) denoting “the import path of this object file”. This worked well enough at the time and maintained complete command-line compatibility with existing Go usage. The changes implementing this transition can be found by searching the Git history for “package global name space”, which is what they eliminated. In particular, CL 190076 (a6736fa4), CL 186263 (758f2bc5), CL 193080 (1cecac81), CL 194053 (19126320), and CL 194071 (531e6b77) did the bulk of this transformation in January 2010. Later, in September 2011, we added the -p flag to the compiler for diagnostic purposes. The problem was that it was easy to create import cycles, especially in tests, and these could not be diagnosed until link time. You'd really want the compiler to diagnose these, for example if the compilation of package sort noticed it was importing a package that itself imported "sort". But the compilation of package sort didn't know its own import path, and so it could not tell whether it had found itself as a transitive dependency. Adding the -p flag solved this problem, and its use was optional, since the linker would still diagnose the import cycle in builds that had not updated to start passing -p. This was CL 4972057 (1e480cd1). There was still no go command at this point, but when we introduced the go command we made it pass -p, which it has for many years at this point. Over time, parts of the compiler began to depend on the presence of the -p flag for various reasonable purposes. For example: In CL 6497074 (041fc8bf; Oct 2012), the race detector used -p to detect packages that should not have race annotations, such as runtime/race and sync/atomic. In CL 13367052 (7276c02b; Sep 2013), a bug fix used -p to detect the compilation of package reflect. In CL 30539 (8aadcc55; Oct 2016), the compiler started using -p to identify package math, to be able to intrinsify calls to Sqrt inside that package. In CL 61019 (9daee931; Sep 2017), CL 71430 (2c1d2e06; Oct 2017), and later related CLs, the compiler started using the -p value when creating various DWARF debugging information. In CL 174657 (cc5eaf93; May 2019), the compiler started writing symbols without the magic empty string whenever -p was used, to reduce the amount of work required in the linker. In CL 179861 (dde7c770; Jun 2019), the compiler made the second argument to //go:linkname optional when -p is used, because in that case the compiler can derive an appropriate default. There are more examples. Today it is impossible to compile the Go standard library without using -p, and DWARF debug information is incomplete without using -p. All known Go build systems pass -p. In particular, the go command does, which is what nearly all Go developers invoke to build Go code. And Bazel does, for go_library rules that set the importpath attribute, which is all rules generated by Gazelle. Gccgo has an equivalent of -p and has required its use in order to disambiguate packages with the same name but different import paths since 2010. On top of all this, various parts of code generation for generics are made more complicated by needing to cope with the case where -p is not specified, even though it's essentially always specified. In summary, the current state is: - Use of the -p flag with cmd/compile is required for building the standard library, and for complete DWARF information, and to enable certain linker speedups. - The go command and Bazel, which we expect account for just about 100% of Go builds, both invoke cmd/compile with -p. - The code in cmd/compile to support builds without -p is complex and has become more complex with generics, but it is almost always dead code and therefore not worth maintaining. - Gccgo already requires its equivalent of -p in any build where two packages have the same name. All this supports the change in this CL, which makes -p required and adjusts tests that invoke cmd/compile to add -p appropriately. Future CLs will be able to remove all the code dealing with the possibility of -p not having been specified. Change-Id: I6b95b9d4cffe59c7bac82eb273ef6c4a67bb0e43 Reviewed-on: https://go-review.googlesource.com/c/go/+/391014 Trust: Russ Cox <rsc@golang.org> Run-TryBot: Russ Cox <rsc@golang.org> TryBot-Result: Gopher Robot <gobot@golang.org> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2022-03-08 16:16:35 -07:00
type goDirPkg struct {
name string
files []string
}
// If singlefilepkgs is set, each file is considered a separate package
// even if the package names are the same.
cmd/compile: require -p flag The -p flag specifies the import path of the package being compiled. This CL makes it required when invoking the compiler and adjusts tests that invoke the compiler directly to conform to this new requirement. The go command already passes the flag, so it is unmodified in this CL. It is expected that any other Go build systems also already pass -p, or else they will need to arrange to do so before updating to Go 1.19. Of particular note, Bazel already does for rules with an importpath= attribute, which includes all Gazelle-generated rules. There is more cleanup possible now in cmd/compile, cmd/link, and other consumers of Go object files, but that is left to future CLs. Additional historical background follows but can be ignored. Long ago, before the go command, or modules, or any kind of versioning, symbols in Go archive files were named using just the package name, so that for example func F in math/rand and func F in crypto/rand would both be the object file symbol 'rand.F'. This led to collisions even in small source trees, which made certain packages unusable in the presence of other packages and generally was a problem for Go's goal of scaling to very large source trees. Fixing this problem required changing from package names to import paths in symbol names, which was mostly straightforward. One wrinkle, though, is that the compiler did not know the import path of the package being compiled; it only knew the package name. At the time, there was no go command, just Makefiles that people had invoking 6g (now “go tool compile”) and then copying the resulting object file to an importable location. That is, everyone had a custom build setup for Go, because there was no standard one. So it was not particularly attractive to change how the compiler was invoked, since that would break approximately every Go user at the time. Instead, we arranged for the compiler to emit, and other tools reading object files to recognize, a special import path (the empty string, it turned out) denoting “the import path of this object file”. This worked well enough at the time and maintained complete command-line compatibility with existing Go usage. The changes implementing this transition can be found by searching the Git history for “package global name space”, which is what they eliminated. In particular, CL 190076 (a6736fa4), CL 186263 (758f2bc5), CL 193080 (1cecac81), CL 194053 (19126320), and CL 194071 (531e6b77) did the bulk of this transformation in January 2010. Later, in September 2011, we added the -p flag to the compiler for diagnostic purposes. The problem was that it was easy to create import cycles, especially in tests, and these could not be diagnosed until link time. You'd really want the compiler to diagnose these, for example if the compilation of package sort noticed it was importing a package that itself imported "sort". But the compilation of package sort didn't know its own import path, and so it could not tell whether it had found itself as a transitive dependency. Adding the -p flag solved this problem, and its use was optional, since the linker would still diagnose the import cycle in builds that had not updated to start passing -p. This was CL 4972057 (1e480cd1). There was still no go command at this point, but when we introduced the go command we made it pass -p, which it has for many years at this point. Over time, parts of the compiler began to depend on the presence of the -p flag for various reasonable purposes. For example: In CL 6497074 (041fc8bf; Oct 2012), the race detector used -p to detect packages that should not have race annotations, such as runtime/race and sync/atomic. In CL 13367052 (7276c02b; Sep 2013), a bug fix used -p to detect the compilation of package reflect. In CL 30539 (8aadcc55; Oct 2016), the compiler started using -p to identify package math, to be able to intrinsify calls to Sqrt inside that package. In CL 61019 (9daee931; Sep 2017), CL 71430 (2c1d2e06; Oct 2017), and later related CLs, the compiler started using the -p value when creating various DWARF debugging information. In CL 174657 (cc5eaf93; May 2019), the compiler started writing symbols without the magic empty string whenever -p was used, to reduce the amount of work required in the linker. In CL 179861 (dde7c770; Jun 2019), the compiler made the second argument to //go:linkname optional when -p is used, because in that case the compiler can derive an appropriate default. There are more examples. Today it is impossible to compile the Go standard library without using -p, and DWARF debug information is incomplete without using -p. All known Go build systems pass -p. In particular, the go command does, which is what nearly all Go developers invoke to build Go code. And Bazel does, for go_library rules that set the importpath attribute, which is all rules generated by Gazelle. Gccgo has an equivalent of -p and has required its use in order to disambiguate packages with the same name but different import paths since 2010. On top of all this, various parts of code generation for generics are made more complicated by needing to cope with the case where -p is not specified, even though it's essentially always specified. In summary, the current state is: - Use of the -p flag with cmd/compile is required for building the standard library, and for complete DWARF information, and to enable certain linker speedups. - The go command and Bazel, which we expect account for just about 100% of Go builds, both invoke cmd/compile with -p. - The code in cmd/compile to support builds without -p is complex and has become more complex with generics, but it is almost always dead code and therefore not worth maintaining. - Gccgo already requires its equivalent of -p in any build where two packages have the same name. All this supports the change in this CL, which makes -p required and adjusts tests that invoke cmd/compile to add -p appropriately. Future CLs will be able to remove all the code dealing with the possibility of -p not having been specified. Change-Id: I6b95b9d4cffe59c7bac82eb273ef6c4a67bb0e43 Reviewed-on: https://go-review.googlesource.com/c/go/+/391014 Trust: Russ Cox <rsc@golang.org> Run-TryBot: Russ Cox <rsc@golang.org> TryBot-Result: Gopher Robot <gobot@golang.org> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2022-03-08 16:16:35 -07:00
func goDirPackages(longdir string, singlefilepkgs bool) ([]*goDirPkg, error) {
files, err := goDirFiles(longdir)
if err != nil {
return nil, err
}
cmd/compile: require -p flag The -p flag specifies the import path of the package being compiled. This CL makes it required when invoking the compiler and adjusts tests that invoke the compiler directly to conform to this new requirement. The go command already passes the flag, so it is unmodified in this CL. It is expected that any other Go build systems also already pass -p, or else they will need to arrange to do so before updating to Go 1.19. Of particular note, Bazel already does for rules with an importpath= attribute, which includes all Gazelle-generated rules. There is more cleanup possible now in cmd/compile, cmd/link, and other consumers of Go object files, but that is left to future CLs. Additional historical background follows but can be ignored. Long ago, before the go command, or modules, or any kind of versioning, symbols in Go archive files were named using just the package name, so that for example func F in math/rand and func F in crypto/rand would both be the object file symbol 'rand.F'. This led to collisions even in small source trees, which made certain packages unusable in the presence of other packages and generally was a problem for Go's goal of scaling to very large source trees. Fixing this problem required changing from package names to import paths in symbol names, which was mostly straightforward. One wrinkle, though, is that the compiler did not know the import path of the package being compiled; it only knew the package name. At the time, there was no go command, just Makefiles that people had invoking 6g (now “go tool compile”) and then copying the resulting object file to an importable location. That is, everyone had a custom build setup for Go, because there was no standard one. So it was not particularly attractive to change how the compiler was invoked, since that would break approximately every Go user at the time. Instead, we arranged for the compiler to emit, and other tools reading object files to recognize, a special import path (the empty string, it turned out) denoting “the import path of this object file”. This worked well enough at the time and maintained complete command-line compatibility with existing Go usage. The changes implementing this transition can be found by searching the Git history for “package global name space”, which is what they eliminated. In particular, CL 190076 (a6736fa4), CL 186263 (758f2bc5), CL 193080 (1cecac81), CL 194053 (19126320), and CL 194071 (531e6b77) did the bulk of this transformation in January 2010. Later, in September 2011, we added the -p flag to the compiler for diagnostic purposes. The problem was that it was easy to create import cycles, especially in tests, and these could not be diagnosed until link time. You'd really want the compiler to diagnose these, for example if the compilation of package sort noticed it was importing a package that itself imported "sort". But the compilation of package sort didn't know its own import path, and so it could not tell whether it had found itself as a transitive dependency. Adding the -p flag solved this problem, and its use was optional, since the linker would still diagnose the import cycle in builds that had not updated to start passing -p. This was CL 4972057 (1e480cd1). There was still no go command at this point, but when we introduced the go command we made it pass -p, which it has for many years at this point. Over time, parts of the compiler began to depend on the presence of the -p flag for various reasonable purposes. For example: In CL 6497074 (041fc8bf; Oct 2012), the race detector used -p to detect packages that should not have race annotations, such as runtime/race and sync/atomic. In CL 13367052 (7276c02b; Sep 2013), a bug fix used -p to detect the compilation of package reflect. In CL 30539 (8aadcc55; Oct 2016), the compiler started using -p to identify package math, to be able to intrinsify calls to Sqrt inside that package. In CL 61019 (9daee931; Sep 2017), CL 71430 (2c1d2e06; Oct 2017), and later related CLs, the compiler started using the -p value when creating various DWARF debugging information. In CL 174657 (cc5eaf93; May 2019), the compiler started writing symbols without the magic empty string whenever -p was used, to reduce the amount of work required in the linker. In CL 179861 (dde7c770; Jun 2019), the compiler made the second argument to //go:linkname optional when -p is used, because in that case the compiler can derive an appropriate default. There are more examples. Today it is impossible to compile the Go standard library without using -p, and DWARF debug information is incomplete without using -p. All known Go build systems pass -p. In particular, the go command does, which is what nearly all Go developers invoke to build Go code. And Bazel does, for go_library rules that set the importpath attribute, which is all rules generated by Gazelle. Gccgo has an equivalent of -p and has required its use in order to disambiguate packages with the same name but different import paths since 2010. On top of all this, various parts of code generation for generics are made more complicated by needing to cope with the case where -p is not specified, even though it's essentially always specified. In summary, the current state is: - Use of the -p flag with cmd/compile is required for building the standard library, and for complete DWARF information, and to enable certain linker speedups. - The go command and Bazel, which we expect account for just about 100% of Go builds, both invoke cmd/compile with -p. - The code in cmd/compile to support builds without -p is complex and has become more complex with generics, but it is almost always dead code and therefore not worth maintaining. - Gccgo already requires its equivalent of -p in any build where two packages have the same name. All this supports the change in this CL, which makes -p required and adjusts tests that invoke cmd/compile to add -p appropriately. Future CLs will be able to remove all the code dealing with the possibility of -p not having been specified. Change-Id: I6b95b9d4cffe59c7bac82eb273ef6c4a67bb0e43 Reviewed-on: https://go-review.googlesource.com/c/go/+/391014 Trust: Russ Cox <rsc@golang.org> Run-TryBot: Russ Cox <rsc@golang.org> TryBot-Result: Gopher Robot <gobot@golang.org> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2022-03-08 16:16:35 -07:00
var pkgs []*goDirPkg
m := make(map[string]*goDirPkg)
for _, file := range files {
name := file.Name()
pkgname, err := getPackageNameFromSource(filepath.Join(longdir, name))
if err != nil {
log.Fatal(err)
}
cmd/compile: require -p flag The -p flag specifies the import path of the package being compiled. This CL makes it required when invoking the compiler and adjusts tests that invoke the compiler directly to conform to this new requirement. The go command already passes the flag, so it is unmodified in this CL. It is expected that any other Go build systems also already pass -p, or else they will need to arrange to do so before updating to Go 1.19. Of particular note, Bazel already does for rules with an importpath= attribute, which includes all Gazelle-generated rules. There is more cleanup possible now in cmd/compile, cmd/link, and other consumers of Go object files, but that is left to future CLs. Additional historical background follows but can be ignored. Long ago, before the go command, or modules, or any kind of versioning, symbols in Go archive files were named using just the package name, so that for example func F in math/rand and func F in crypto/rand would both be the object file symbol 'rand.F'. This led to collisions even in small source trees, which made certain packages unusable in the presence of other packages and generally was a problem for Go's goal of scaling to very large source trees. Fixing this problem required changing from package names to import paths in symbol names, which was mostly straightforward. One wrinkle, though, is that the compiler did not know the import path of the package being compiled; it only knew the package name. At the time, there was no go command, just Makefiles that people had invoking 6g (now “go tool compile”) and then copying the resulting object file to an importable location. That is, everyone had a custom build setup for Go, because there was no standard one. So it was not particularly attractive to change how the compiler was invoked, since that would break approximately every Go user at the time. Instead, we arranged for the compiler to emit, and other tools reading object files to recognize, a special import path (the empty string, it turned out) denoting “the import path of this object file”. This worked well enough at the time and maintained complete command-line compatibility with existing Go usage. The changes implementing this transition can be found by searching the Git history for “package global name space”, which is what they eliminated. In particular, CL 190076 (a6736fa4), CL 186263 (758f2bc5), CL 193080 (1cecac81), CL 194053 (19126320), and CL 194071 (531e6b77) did the bulk of this transformation in January 2010. Later, in September 2011, we added the -p flag to the compiler for diagnostic purposes. The problem was that it was easy to create import cycles, especially in tests, and these could not be diagnosed until link time. You'd really want the compiler to diagnose these, for example if the compilation of package sort noticed it was importing a package that itself imported "sort". But the compilation of package sort didn't know its own import path, and so it could not tell whether it had found itself as a transitive dependency. Adding the -p flag solved this problem, and its use was optional, since the linker would still diagnose the import cycle in builds that had not updated to start passing -p. This was CL 4972057 (1e480cd1). There was still no go command at this point, but when we introduced the go command we made it pass -p, which it has for many years at this point. Over time, parts of the compiler began to depend on the presence of the -p flag for various reasonable purposes. For example: In CL 6497074 (041fc8bf; Oct 2012), the race detector used -p to detect packages that should not have race annotations, such as runtime/race and sync/atomic. In CL 13367052 (7276c02b; Sep 2013), a bug fix used -p to detect the compilation of package reflect. In CL 30539 (8aadcc55; Oct 2016), the compiler started using -p to identify package math, to be able to intrinsify calls to Sqrt inside that package. In CL 61019 (9daee931; Sep 2017), CL 71430 (2c1d2e06; Oct 2017), and later related CLs, the compiler started using the -p value when creating various DWARF debugging information. In CL 174657 (cc5eaf93; May 2019), the compiler started writing symbols without the magic empty string whenever -p was used, to reduce the amount of work required in the linker. In CL 179861 (dde7c770; Jun 2019), the compiler made the second argument to //go:linkname optional when -p is used, because in that case the compiler can derive an appropriate default. There are more examples. Today it is impossible to compile the Go standard library without using -p, and DWARF debug information is incomplete without using -p. All known Go build systems pass -p. In particular, the go command does, which is what nearly all Go developers invoke to build Go code. And Bazel does, for go_library rules that set the importpath attribute, which is all rules generated by Gazelle. Gccgo has an equivalent of -p and has required its use in order to disambiguate packages with the same name but different import paths since 2010. On top of all this, various parts of code generation for generics are made more complicated by needing to cope with the case where -p is not specified, even though it's essentially always specified. In summary, the current state is: - Use of the -p flag with cmd/compile is required for building the standard library, and for complete DWARF information, and to enable certain linker speedups. - The go command and Bazel, which we expect account for just about 100% of Go builds, both invoke cmd/compile with -p. - The code in cmd/compile to support builds without -p is complex and has become more complex with generics, but it is almost always dead code and therefore not worth maintaining. - Gccgo already requires its equivalent of -p in any build where two packages have the same name. All this supports the change in this CL, which makes -p required and adjusts tests that invoke cmd/compile to add -p appropriately. Future CLs will be able to remove all the code dealing with the possibility of -p not having been specified. Change-Id: I6b95b9d4cffe59c7bac82eb273ef6c4a67bb0e43 Reviewed-on: https://go-review.googlesource.com/c/go/+/391014 Trust: Russ Cox <rsc@golang.org> Run-TryBot: Russ Cox <rsc@golang.org> TryBot-Result: Gopher Robot <gobot@golang.org> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2022-03-08 16:16:35 -07:00
p, ok := m[pkgname]
if singlefilepkgs || !ok {
cmd/compile: require -p flag The -p flag specifies the import path of the package being compiled. This CL makes it required when invoking the compiler and adjusts tests that invoke the compiler directly to conform to this new requirement. The go command already passes the flag, so it is unmodified in this CL. It is expected that any other Go build systems also already pass -p, or else they will need to arrange to do so before updating to Go 1.19. Of particular note, Bazel already does for rules with an importpath= attribute, which includes all Gazelle-generated rules. There is more cleanup possible now in cmd/compile, cmd/link, and other consumers of Go object files, but that is left to future CLs. Additional historical background follows but can be ignored. Long ago, before the go command, or modules, or any kind of versioning, symbols in Go archive files were named using just the package name, so that for example func F in math/rand and func F in crypto/rand would both be the object file symbol 'rand.F'. This led to collisions even in small source trees, which made certain packages unusable in the presence of other packages and generally was a problem for Go's goal of scaling to very large source trees. Fixing this problem required changing from package names to import paths in symbol names, which was mostly straightforward. One wrinkle, though, is that the compiler did not know the import path of the package being compiled; it only knew the package name. At the time, there was no go command, just Makefiles that people had invoking 6g (now “go tool compile”) and then copying the resulting object file to an importable location. That is, everyone had a custom build setup for Go, because there was no standard one. So it was not particularly attractive to change how the compiler was invoked, since that would break approximately every Go user at the time. Instead, we arranged for the compiler to emit, and other tools reading object files to recognize, a special import path (the empty string, it turned out) denoting “the import path of this object file”. This worked well enough at the time and maintained complete command-line compatibility with existing Go usage. The changes implementing this transition can be found by searching the Git history for “package global name space”, which is what they eliminated. In particular, CL 190076 (a6736fa4), CL 186263 (758f2bc5), CL 193080 (1cecac81), CL 194053 (19126320), and CL 194071 (531e6b77) did the bulk of this transformation in January 2010. Later, in September 2011, we added the -p flag to the compiler for diagnostic purposes. The problem was that it was easy to create import cycles, especially in tests, and these could not be diagnosed until link time. You'd really want the compiler to diagnose these, for example if the compilation of package sort noticed it was importing a package that itself imported "sort". But the compilation of package sort didn't know its own import path, and so it could not tell whether it had found itself as a transitive dependency. Adding the -p flag solved this problem, and its use was optional, since the linker would still diagnose the import cycle in builds that had not updated to start passing -p. This was CL 4972057 (1e480cd1). There was still no go command at this point, but when we introduced the go command we made it pass -p, which it has for many years at this point. Over time, parts of the compiler began to depend on the presence of the -p flag for various reasonable purposes. For example: In CL 6497074 (041fc8bf; Oct 2012), the race detector used -p to detect packages that should not have race annotations, such as runtime/race and sync/atomic. In CL 13367052 (7276c02b; Sep 2013), a bug fix used -p to detect the compilation of package reflect. In CL 30539 (8aadcc55; Oct 2016), the compiler started using -p to identify package math, to be able to intrinsify calls to Sqrt inside that package. In CL 61019 (9daee931; Sep 2017), CL 71430 (2c1d2e06; Oct 2017), and later related CLs, the compiler started using the -p value when creating various DWARF debugging information. In CL 174657 (cc5eaf93; May 2019), the compiler started writing symbols without the magic empty string whenever -p was used, to reduce the amount of work required in the linker. In CL 179861 (dde7c770; Jun 2019), the compiler made the second argument to //go:linkname optional when -p is used, because in that case the compiler can derive an appropriate default. There are more examples. Today it is impossible to compile the Go standard library without using -p, and DWARF debug information is incomplete without using -p. All known Go build systems pass -p. In particular, the go command does, which is what nearly all Go developers invoke to build Go code. And Bazel does, for go_library rules that set the importpath attribute, which is all rules generated by Gazelle. Gccgo has an equivalent of -p and has required its use in order to disambiguate packages with the same name but different import paths since 2010. On top of all this, various parts of code generation for generics are made more complicated by needing to cope with the case where -p is not specified, even though it's essentially always specified. In summary, the current state is: - Use of the -p flag with cmd/compile is required for building the standard library, and for complete DWARF information, and to enable certain linker speedups. - The go command and Bazel, which we expect account for just about 100% of Go builds, both invoke cmd/compile with -p. - The code in cmd/compile to support builds without -p is complex and has become more complex with generics, but it is almost always dead code and therefore not worth maintaining. - Gccgo already requires its equivalent of -p in any build where two packages have the same name. All this supports the change in this CL, which makes -p required and adjusts tests that invoke cmd/compile to add -p appropriately. Future CLs will be able to remove all the code dealing with the possibility of -p not having been specified. Change-Id: I6b95b9d4cffe59c7bac82eb273ef6c4a67bb0e43 Reviewed-on: https://go-review.googlesource.com/c/go/+/391014 Trust: Russ Cox <rsc@golang.org> Run-TryBot: Russ Cox <rsc@golang.org> TryBot-Result: Gopher Robot <gobot@golang.org> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2022-03-08 16:16:35 -07:00
p = &goDirPkg{name: pkgname}
pkgs = append(pkgs, p)
m[pkgname] = p
}
cmd/compile: require -p flag The -p flag specifies the import path of the package being compiled. This CL makes it required when invoking the compiler and adjusts tests that invoke the compiler directly to conform to this new requirement. The go command already passes the flag, so it is unmodified in this CL. It is expected that any other Go build systems also already pass -p, or else they will need to arrange to do so before updating to Go 1.19. Of particular note, Bazel already does for rules with an importpath= attribute, which includes all Gazelle-generated rules. There is more cleanup possible now in cmd/compile, cmd/link, and other consumers of Go object files, but that is left to future CLs. Additional historical background follows but can be ignored. Long ago, before the go command, or modules, or any kind of versioning, symbols in Go archive files were named using just the package name, so that for example func F in math/rand and func F in crypto/rand would both be the object file symbol 'rand.F'. This led to collisions even in small source trees, which made certain packages unusable in the presence of other packages and generally was a problem for Go's goal of scaling to very large source trees. Fixing this problem required changing from package names to import paths in symbol names, which was mostly straightforward. One wrinkle, though, is that the compiler did not know the import path of the package being compiled; it only knew the package name. At the time, there was no go command, just Makefiles that people had invoking 6g (now “go tool compile”) and then copying the resulting object file to an importable location. That is, everyone had a custom build setup for Go, because there was no standard one. So it was not particularly attractive to change how the compiler was invoked, since that would break approximately every Go user at the time. Instead, we arranged for the compiler to emit, and other tools reading object files to recognize, a special import path (the empty string, it turned out) denoting “the import path of this object file”. This worked well enough at the time and maintained complete command-line compatibility with existing Go usage. The changes implementing this transition can be found by searching the Git history for “package global name space”, which is what they eliminated. In particular, CL 190076 (a6736fa4), CL 186263 (758f2bc5), CL 193080 (1cecac81), CL 194053 (19126320), and CL 194071 (531e6b77) did the bulk of this transformation in January 2010. Later, in September 2011, we added the -p flag to the compiler for diagnostic purposes. The problem was that it was easy to create import cycles, especially in tests, and these could not be diagnosed until link time. You'd really want the compiler to diagnose these, for example if the compilation of package sort noticed it was importing a package that itself imported "sort". But the compilation of package sort didn't know its own import path, and so it could not tell whether it had found itself as a transitive dependency. Adding the -p flag solved this problem, and its use was optional, since the linker would still diagnose the import cycle in builds that had not updated to start passing -p. This was CL 4972057 (1e480cd1). There was still no go command at this point, but when we introduced the go command we made it pass -p, which it has for many years at this point. Over time, parts of the compiler began to depend on the presence of the -p flag for various reasonable purposes. For example: In CL 6497074 (041fc8bf; Oct 2012), the race detector used -p to detect packages that should not have race annotations, such as runtime/race and sync/atomic. In CL 13367052 (7276c02b; Sep 2013), a bug fix used -p to detect the compilation of package reflect. In CL 30539 (8aadcc55; Oct 2016), the compiler started using -p to identify package math, to be able to intrinsify calls to Sqrt inside that package. In CL 61019 (9daee931; Sep 2017), CL 71430 (2c1d2e06; Oct 2017), and later related CLs, the compiler started using the -p value when creating various DWARF debugging information. In CL 174657 (cc5eaf93; May 2019), the compiler started writing symbols without the magic empty string whenever -p was used, to reduce the amount of work required in the linker. In CL 179861 (dde7c770; Jun 2019), the compiler made the second argument to //go:linkname optional when -p is used, because in that case the compiler can derive an appropriate default. There are more examples. Today it is impossible to compile the Go standard library without using -p, and DWARF debug information is incomplete without using -p. All known Go build systems pass -p. In particular, the go command does, which is what nearly all Go developers invoke to build Go code. And Bazel does, for go_library rules that set the importpath attribute, which is all rules generated by Gazelle. Gccgo has an equivalent of -p and has required its use in order to disambiguate packages with the same name but different import paths since 2010. On top of all this, various parts of code generation for generics are made more complicated by needing to cope with the case where -p is not specified, even though it's essentially always specified. In summary, the current state is: - Use of the -p flag with cmd/compile is required for building the standard library, and for complete DWARF information, and to enable certain linker speedups. - The go command and Bazel, which we expect account for just about 100% of Go builds, both invoke cmd/compile with -p. - The code in cmd/compile to support builds without -p is complex and has become more complex with generics, but it is almost always dead code and therefore not worth maintaining. - Gccgo already requires its equivalent of -p in any build where two packages have the same name. All this supports the change in this CL, which makes -p required and adjusts tests that invoke cmd/compile to add -p appropriately. Future CLs will be able to remove all the code dealing with the possibility of -p not having been specified. Change-Id: I6b95b9d4cffe59c7bac82eb273ef6c4a67bb0e43 Reviewed-on: https://go-review.googlesource.com/c/go/+/391014 Trust: Russ Cox <rsc@golang.org> Run-TryBot: Russ Cox <rsc@golang.org> TryBot-Result: Gopher Robot <gobot@golang.org> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2022-03-08 16:16:35 -07:00
p.files = append(p.files, name)
}
return pkgs, nil
}
type context struct {
GOOS string
GOARCH string
cgoEnabled bool
noOptEnv bool
}
// shouldTest looks for build tags in a source file and returns
// whether the file should be used according to the tags.
func shouldTest(src string, goos, goarch string) (ok bool, whyNot string) {
if *runSkips {
return true, ""
}
for _, line := range strings.Split(src, "\n") {
if strings.HasPrefix(line, "package ") {
break
}
if expr, err := constraint.Parse(line); err == nil {
gcFlags := os.Getenv("GO_GCFLAGS")
ctxt := &context{
GOOS: goos,
GOARCH: goarch,
cgoEnabled: cgoEnabled,
noOptEnv: strings.Contains(gcFlags, "-N") || strings.Contains(gcFlags, "-l"),
}
if !expr.Eval(ctxt.match) {
return false, line
}
}
}
return true, ""
}
func (ctxt *context) match(name string) bool {
if name == "" {
return false
}
// Tags must be letters, digits, underscores or dots.
// Unlike in Go identifiers, all digits are fine (e.g., "386").
for _, c := range name {
if !unicode.IsLetter(c) && !unicode.IsDigit(c) && c != '_' && c != '.' {
return false
}
}
if strings.HasPrefix(name, "goexperiment.") {
for _, tag := range build.Default.ToolTags {
if tag == name {
return true
}
}
return false
}
if name == "cgo" && ctxt.cgoEnabled {
return true
}
if name == ctxt.GOOS || name == ctxt.GOARCH || name == "gc" {
return true
}
if ctxt.noOptEnv && name == "gcflags_noopt" {
return true
}
if name == "test_run" {
return true
}
return false
}
func init() { checkShouldTest() }
// goGcflags returns the -gcflags argument to use with go build / go run.
// This must match the flags used for building the standard library,
// or else the commands will rebuild any needed packages (like runtime)
// over and over.
func (t *test) goGcflags() string {
return "-gcflags=all=" + os.Getenv("GO_GCFLAGS")
}
func (t *test) goGcflagsIsEmpty() bool {
return "" == os.Getenv("GO_GCFLAGS")
}
var errTimeout = errors.New("command exceeded time limit")
// run runs a test.
func (t *test) run() {
start := time.Now()
defer func() {
t.dt = time.Since(start)
close(t.donec)
}()
srcBytes, err := ioutil.ReadFile(t.goFileName())
if err != nil {
t.err = err
return
}
t.src = string(srcBytes)
if t.src[0] == '\n' {
t.err = skipError("starts with newline")
return
}
// Execution recipe stops at first blank line.
action, _, ok := strings.Cut(t.src, "\n\n")
if !ok {
t.err = fmt.Errorf("double newline ending execution recipe not found in %s", t.goFileName())
return
}
if firstLine, rest, ok := strings.Cut(action, "\n"); ok && strings.Contains(firstLine, "+build") {
// skip first line
action = rest
}
action = strings.TrimPrefix(action, "//")
// Check for build constraints only up to the actual code.
header, _, ok := strings.Cut(t.src, "\npackage")
if !ok {
header = action // some files are intentionally malformed
}
if ok, why := shouldTest(header, goos, goarch); !ok {
if *showSkips {
fmt.Printf("%-20s %-20s: %s\n", "skip", t.goFileName(), why)
}
return
}
var args, flags, runenv []string
var tim int
wantError := false
wantAuto := false
singlefilepkgs := false
f, err := splitQuoted(action)
if err != nil {
t.err = fmt.Errorf("invalid test recipe: %v", err)
return
}
if len(f) > 0 {
action = f[0]
args = f[1:]
}
// TODO: Clean up/simplify this switch statement.
switch action {
case "compile", "compiledir", "build", "builddir", "buildrundir", "run", "buildrun", "runoutput", "rundir", "runindir", "asmcheck":
// nothing to do
case "errorcheckandrundir":
wantError = false // should be no error if also will run
case "errorcheckwithauto":
action = "errorcheck"
wantAuto = true
wantError = true
case "errorcheck", "errorcheckdir", "errorcheckoutput":
wantError = true
case "skip":
if *runSkips {
break
}
return
default:
t.err = skipError("skipped; unknown pattern: " + action)
return
}
goexp := env.GOEXPERIMENT
// collect flags
for len(args) > 0 && strings.HasPrefix(args[0], "-") {
switch args[0] {
case "-1":
wantError = true
case "-0":
wantError = false
case "-s":
singlefilepkgs = true
case "-t": // timeout in seconds
args = args[1:]
var err error
tim, err = strconv.Atoi(args[0])
if err != nil {
t.err = fmt.Errorf("need number of seconds for -t timeout, got %s instead", args[0])
}
if s := os.Getenv("GO_TEST_TIMEOUT_SCALE"); s != "" {
timeoutScale, err := strconv.Atoi(s)
if err != nil {
log.Fatalf("failed to parse $GO_TEST_TIMEOUT_SCALE = %q as integer: %v", s, err)
}
tim *= timeoutScale
}
case "-goexperiment": // set GOEXPERIMENT environment
args = args[1:]
if goexp != "" {
goexp += ","
}
goexp += args[0]
runenv = append(runenv, "GOEXPERIMENT="+goexp)
default:
flags = append(flags, args[0])
}
args = args[1:]
}
if action == "errorcheck" {
found := false
for i, f := range flags {
if strings.HasPrefix(f, "-d=") {
flags[i] = f + ",ssa/check/on"
found = true
break
}
}
if !found {
flags = append(flags, "-d=ssa/check/on")
}
}
t.initExpectFail()
t.makeTempDir()
if !*keep {
defer os.RemoveAll(t.tempDir)
}
err = ioutil.WriteFile(filepath.Join(t.tempDir, t.gofile), srcBytes, 0644)
if err != nil {
log.Fatal(err)
}
// A few tests (of things like the environment) require these to be set.
if os.Getenv("GOOS") == "" {
os.Setenv("GOOS", runtime.GOOS)
}
if os.Getenv("GOARCH") == "" {
os.Setenv("GOARCH", runtime.GOARCH)
}
var (
runInDir = t.tempDir
tempDirIsGOPATH = false
)
runcmd := func(args ...string) ([]byte, error) {
cmd := exec.Command(args[0], args[1:]...)
var buf bytes.Buffer
cmd.Stdout = &buf
cmd.Stderr = &buf
cmd.Env = append(os.Environ(), "GOENV=off", "GOFLAGS=")
if runInDir != "" {
cmd.Dir = runInDir
// Set PWD to match Dir to speed up os.Getwd in the child process.
cmd.Env = append(cmd.Env, "PWD="+cmd.Dir)
}
if tempDirIsGOPATH {
cmd.Env = append(cmd.Env, "GOPATH="+t.tempDir)
}
cmd.Env = append(cmd.Env, runenv...)
var err error
if tim != 0 {
err = cmd.Start()
// This command-timeout code adapted from cmd/go/test.go
// Note: the Go command uses a more sophisticated timeout
// strategy, first sending SIGQUIT (if appropriate for the
// OS in question) to try to trigger a stack trace, then
// finally much later SIGKILL. If timeouts prove to be a
// common problem here, it would be worth porting over
// that code as well. See https://do.dev/issue/50973
// for more discussion.
if err == nil {
tick := time.NewTimer(time.Duration(tim) * time.Second)
done := make(chan error)
go func() {
done <- cmd.Wait()
}()
select {
case err = <-done:
// ok
case <-tick.C:
cmd.Process.Signal(os.Interrupt)
time.Sleep(1 * time.Second)
cmd.Process.Kill()
<-done
err = errTimeout
}
tick.Stop()
}
} else {
err = cmd.Run()
}
if err != nil && err != errTimeout {
err = fmt.Errorf("%s\n%s", err, buf.Bytes())
}
return buf.Bytes(), err
}
long := filepath.Join(cwd, t.goFileName())
switch action {
default:
t.err = fmt.Errorf("unimplemented action %q", action)
case "asmcheck":
// Compile Go file and match the generated assembly
// against a set of regexps in comments.
ops := t.wantedAsmOpcodes(long)
self := runtime.GOOS + "/" + runtime.GOARCH
for _, env := range ops.Envs() {
// Only run checks relevant to the current GOOS/GOARCH,
// to avoid triggering a cross-compile of the runtime.
if string(env) != self && !strings.HasPrefix(string(env), self+"/") && !*allCodegen {
continue
}
// -S=2 forces outermost line numbers when disassembling inlined code.
cmdline := []string{"build", "-gcflags", "-S=2"}
// Append flags, but don't override -gcflags=-S=2; add to it instead.
for i := 0; i < len(flags); i++ {
flag := flags[i]
switch {
case strings.HasPrefix(flag, "-gcflags="):
cmdline[2] += " " + strings.TrimPrefix(flag, "-gcflags=")
case strings.HasPrefix(flag, "--gcflags="):
cmdline[2] += " " + strings.TrimPrefix(flag, "--gcflags=")
case flag == "-gcflags", flag == "--gcflags":
i++
if i < len(flags) {
cmdline[2] += " " + flags[i]
}
default:
cmdline = append(cmdline, flag)
}
}
cmdline = append(cmdline, long)
cmd := exec.Command(goTool(), cmdline...)
cmd.Env = append(os.Environ(), env.Environ()...)
if len(flags) > 0 && flags[0] == "-race" {
cmd.Env = append(cmd.Env, "CGO_ENABLED=1")
}
var buf bytes.Buffer
cmd.Stdout, cmd.Stderr = &buf, &buf
if err := cmd.Run(); err != nil {
fmt.Println(env, "\n", cmd.Stderr)
t.err = err
return
}
t.err = t.asmCheck(buf.String(), long, env, ops[env])
if t.err != nil {
return
}
}
return
case "errorcheck":
// Compile Go file.
// Fail if wantError is true and compilation was successful and vice versa.
// Match errors produced by gc against errors in comments.
// TODO(gri) remove need for -C (disable printing of columns in error messages)
cmd/compile: require -p flag The -p flag specifies the import path of the package being compiled. This CL makes it required when invoking the compiler and adjusts tests that invoke the compiler directly to conform to this new requirement. The go command already passes the flag, so it is unmodified in this CL. It is expected that any other Go build systems also already pass -p, or else they will need to arrange to do so before updating to Go 1.19. Of particular note, Bazel already does for rules with an importpath= attribute, which includes all Gazelle-generated rules. There is more cleanup possible now in cmd/compile, cmd/link, and other consumers of Go object files, but that is left to future CLs. Additional historical background follows but can be ignored. Long ago, before the go command, or modules, or any kind of versioning, symbols in Go archive files were named using just the package name, so that for example func F in math/rand and func F in crypto/rand would both be the object file symbol 'rand.F'. This led to collisions even in small source trees, which made certain packages unusable in the presence of other packages and generally was a problem for Go's goal of scaling to very large source trees. Fixing this problem required changing from package names to import paths in symbol names, which was mostly straightforward. One wrinkle, though, is that the compiler did not know the import path of the package being compiled; it only knew the package name. At the time, there was no go command, just Makefiles that people had invoking 6g (now “go tool compile”) and then copying the resulting object file to an importable location. That is, everyone had a custom build setup for Go, because there was no standard one. So it was not particularly attractive to change how the compiler was invoked, since that would break approximately every Go user at the time. Instead, we arranged for the compiler to emit, and other tools reading object files to recognize, a special import path (the empty string, it turned out) denoting “the import path of this object file”. This worked well enough at the time and maintained complete command-line compatibility with existing Go usage. The changes implementing this transition can be found by searching the Git history for “package global name space”, which is what they eliminated. In particular, CL 190076 (a6736fa4), CL 186263 (758f2bc5), CL 193080 (1cecac81), CL 194053 (19126320), and CL 194071 (531e6b77) did the bulk of this transformation in January 2010. Later, in September 2011, we added the -p flag to the compiler for diagnostic purposes. The problem was that it was easy to create import cycles, especially in tests, and these could not be diagnosed until link time. You'd really want the compiler to diagnose these, for example if the compilation of package sort noticed it was importing a package that itself imported "sort". But the compilation of package sort didn't know its own import path, and so it could not tell whether it had found itself as a transitive dependency. Adding the -p flag solved this problem, and its use was optional, since the linker would still diagnose the import cycle in builds that had not updated to start passing -p. This was CL 4972057 (1e480cd1). There was still no go command at this point, but when we introduced the go command we made it pass -p, which it has for many years at this point. Over time, parts of the compiler began to depend on the presence of the -p flag for various reasonable purposes. For example: In CL 6497074 (041fc8bf; Oct 2012), the race detector used -p to detect packages that should not have race annotations, such as runtime/race and sync/atomic. In CL 13367052 (7276c02b; Sep 2013), a bug fix used -p to detect the compilation of package reflect. In CL 30539 (8aadcc55; Oct 2016), the compiler started using -p to identify package math, to be able to intrinsify calls to Sqrt inside that package. In CL 61019 (9daee931; Sep 2017), CL 71430 (2c1d2e06; Oct 2017), and later related CLs, the compiler started using the -p value when creating various DWARF debugging information. In CL 174657 (cc5eaf93; May 2019), the compiler started writing symbols without the magic empty string whenever -p was used, to reduce the amount of work required in the linker. In CL 179861 (dde7c770; Jun 2019), the compiler made the second argument to //go:linkname optional when -p is used, because in that case the compiler can derive an appropriate default. There are more examples. Today it is impossible to compile the Go standard library without using -p, and DWARF debug information is incomplete without using -p. All known Go build systems pass -p. In particular, the go command does, which is what nearly all Go developers invoke to build Go code. And Bazel does, for go_library rules that set the importpath attribute, which is all rules generated by Gazelle. Gccgo has an equivalent of -p and has required its use in order to disambiguate packages with the same name but different import paths since 2010. On top of all this, various parts of code generation for generics are made more complicated by needing to cope with the case where -p is not specified, even though it's essentially always specified. In summary, the current state is: - Use of the -p flag with cmd/compile is required for building the standard library, and for complete DWARF information, and to enable certain linker speedups. - The go command and Bazel, which we expect account for just about 100% of Go builds, both invoke cmd/compile with -p. - The code in cmd/compile to support builds without -p is complex and has become more complex with generics, but it is almost always dead code and therefore not worth maintaining. - Gccgo already requires its equivalent of -p in any build where two packages have the same name. All this supports the change in this CL, which makes -p required and adjusts tests that invoke cmd/compile to add -p appropriately. Future CLs will be able to remove all the code dealing with the possibility of -p not having been specified. Change-Id: I6b95b9d4cffe59c7bac82eb273ef6c4a67bb0e43 Reviewed-on: https://go-review.googlesource.com/c/go/+/391014 Trust: Russ Cox <rsc@golang.org> Run-TryBot: Russ Cox <rsc@golang.org> TryBot-Result: Gopher Robot <gobot@golang.org> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2022-03-08 16:16:35 -07:00
cmdline := []string{goTool(), "tool", "compile", "-p=p", "-d=panic", "-C", "-e", "-o", "a.o"}
// No need to add -dynlink even if linkshared if we're just checking for errors...
cmdline = append(cmdline, flags...)
cmdline = append(cmdline, long)
out, err := runcmd(cmdline...)
if wantError {
if err == nil {
t.err = fmt.Errorf("compilation succeeded unexpectedly\n%s", out)
return
}
if err == errTimeout {
t.err = fmt.Errorf("compilation timed out")
return
}
} else {
if err != nil {
t.err = err
return
}
}
if *updateErrors {
t.updateErrors(string(out), long)
}
t.err = t.errorCheck(string(out), wantAuto, long, t.gofile)
case "compile":
// Compile Go file.
_, t.err = compileFile(runcmd, long, flags)
case "compiledir":
// Compile all files in the directory as packages in lexicographic order.
longdir := filepath.Join(cwd, t.goDirName())
pkgs, err := goDirPackages(longdir, singlefilepkgs)
if err != nil {
t.err = err
return
}
cmd/compile: require -p flag The -p flag specifies the import path of the package being compiled. This CL makes it required when invoking the compiler and adjusts tests that invoke the compiler directly to conform to this new requirement. The go command already passes the flag, so it is unmodified in this CL. It is expected that any other Go build systems also already pass -p, or else they will need to arrange to do so before updating to Go 1.19. Of particular note, Bazel already does for rules with an importpath= attribute, which includes all Gazelle-generated rules. There is more cleanup possible now in cmd/compile, cmd/link, and other consumers of Go object files, but that is left to future CLs. Additional historical background follows but can be ignored. Long ago, before the go command, or modules, or any kind of versioning, symbols in Go archive files were named using just the package name, so that for example func F in math/rand and func F in crypto/rand would both be the object file symbol 'rand.F'. This led to collisions even in small source trees, which made certain packages unusable in the presence of other packages and generally was a problem for Go's goal of scaling to very large source trees. Fixing this problem required changing from package names to import paths in symbol names, which was mostly straightforward. One wrinkle, though, is that the compiler did not know the import path of the package being compiled; it only knew the package name. At the time, there was no go command, just Makefiles that people had invoking 6g (now “go tool compile”) and then copying the resulting object file to an importable location. That is, everyone had a custom build setup for Go, because there was no standard one. So it was not particularly attractive to change how the compiler was invoked, since that would break approximately every Go user at the time. Instead, we arranged for the compiler to emit, and other tools reading object files to recognize, a special import path (the empty string, it turned out) denoting “the import path of this object file”. This worked well enough at the time and maintained complete command-line compatibility with existing Go usage. The changes implementing this transition can be found by searching the Git history for “package global name space”, which is what they eliminated. In particular, CL 190076 (a6736fa4), CL 186263 (758f2bc5), CL 193080 (1cecac81), CL 194053 (19126320), and CL 194071 (531e6b77) did the bulk of this transformation in January 2010. Later, in September 2011, we added the -p flag to the compiler for diagnostic purposes. The problem was that it was easy to create import cycles, especially in tests, and these could not be diagnosed until link time. You'd really want the compiler to diagnose these, for example if the compilation of package sort noticed it was importing a package that itself imported "sort". But the compilation of package sort didn't know its own import path, and so it could not tell whether it had found itself as a transitive dependency. Adding the -p flag solved this problem, and its use was optional, since the linker would still diagnose the import cycle in builds that had not updated to start passing -p. This was CL 4972057 (1e480cd1). There was still no go command at this point, but when we introduced the go command we made it pass -p, which it has for many years at this point. Over time, parts of the compiler began to depend on the presence of the -p flag for various reasonable purposes. For example: In CL 6497074 (041fc8bf; Oct 2012), the race detector used -p to detect packages that should not have race annotations, such as runtime/race and sync/atomic. In CL 13367052 (7276c02b; Sep 2013), a bug fix used -p to detect the compilation of package reflect. In CL 30539 (8aadcc55; Oct 2016), the compiler started using -p to identify package math, to be able to intrinsify calls to Sqrt inside that package. In CL 61019 (9daee931; Sep 2017), CL 71430 (2c1d2e06; Oct 2017), and later related CLs, the compiler started using the -p value when creating various DWARF debugging information. In CL 174657 (cc5eaf93; May 2019), the compiler started writing symbols without the magic empty string whenever -p was used, to reduce the amount of work required in the linker. In CL 179861 (dde7c770; Jun 2019), the compiler made the second argument to //go:linkname optional when -p is used, because in that case the compiler can derive an appropriate default. There are more examples. Today it is impossible to compile the Go standard library without using -p, and DWARF debug information is incomplete without using -p. All known Go build systems pass -p. In particular, the go command does, which is what nearly all Go developers invoke to build Go code. And Bazel does, for go_library rules that set the importpath attribute, which is all rules generated by Gazelle. Gccgo has an equivalent of -p and has required its use in order to disambiguate packages with the same name but different import paths since 2010. On top of all this, various parts of code generation for generics are made more complicated by needing to cope with the case where -p is not specified, even though it's essentially always specified. In summary, the current state is: - Use of the -p flag with cmd/compile is required for building the standard library, and for complete DWARF information, and to enable certain linker speedups. - The go command and Bazel, which we expect account for just about 100% of Go builds, both invoke cmd/compile with -p. - The code in cmd/compile to support builds without -p is complex and has become more complex with generics, but it is almost always dead code and therefore not worth maintaining. - Gccgo already requires its equivalent of -p in any build where two packages have the same name. All this supports the change in this CL, which makes -p required and adjusts tests that invoke cmd/compile to add -p appropriately. Future CLs will be able to remove all the code dealing with the possibility of -p not having been specified. Change-Id: I6b95b9d4cffe59c7bac82eb273ef6c4a67bb0e43 Reviewed-on: https://go-review.googlesource.com/c/go/+/391014 Trust: Russ Cox <rsc@golang.org> Run-TryBot: Russ Cox <rsc@golang.org> TryBot-Result: Gopher Robot <gobot@golang.org> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2022-03-08 16:16:35 -07:00
for _, pkg := range pkgs {
_, t.err = compileInDir(runcmd, longdir, flags, pkg.name, pkg.files...)
if t.err != nil {
return
}
}
case "errorcheckdir", "errorcheckandrundir":
flags = append(flags, "-d=panic")
// Compile and errorCheck all files in the directory as packages in lexicographic order.
// If errorcheckdir and wantError, compilation of the last package must fail.
// If errorcheckandrundir and wantError, compilation of the package prior the last must fail.
longdir := filepath.Join(cwd, t.goDirName())
pkgs, err := goDirPackages(longdir, singlefilepkgs)
if err != nil {
t.err = err
return
}
errPkg := len(pkgs) - 1
if wantError && action == "errorcheckandrundir" {
// The last pkg should compiled successfully and will be run in next case.
// Preceding pkg must return an error from compileInDir.
errPkg--
}
cmd/compile: require -p flag The -p flag specifies the import path of the package being compiled. This CL makes it required when invoking the compiler and adjusts tests that invoke the compiler directly to conform to this new requirement. The go command already passes the flag, so it is unmodified in this CL. It is expected that any other Go build systems also already pass -p, or else they will need to arrange to do so before updating to Go 1.19. Of particular note, Bazel already does for rules with an importpath= attribute, which includes all Gazelle-generated rules. There is more cleanup possible now in cmd/compile, cmd/link, and other consumers of Go object files, but that is left to future CLs. Additional historical background follows but can be ignored. Long ago, before the go command, or modules, or any kind of versioning, symbols in Go archive files were named using just the package name, so that for example func F in math/rand and func F in crypto/rand would both be the object file symbol 'rand.F'. This led to collisions even in small source trees, which made certain packages unusable in the presence of other packages and generally was a problem for Go's goal of scaling to very large source trees. Fixing this problem required changing from package names to import paths in symbol names, which was mostly straightforward. One wrinkle, though, is that the compiler did not know the import path of the package being compiled; it only knew the package name. At the time, there was no go command, just Makefiles that people had invoking 6g (now “go tool compile”) and then copying the resulting object file to an importable location. That is, everyone had a custom build setup for Go, because there was no standard one. So it was not particularly attractive to change how the compiler was invoked, since that would break approximately every Go user at the time. Instead, we arranged for the compiler to emit, and other tools reading object files to recognize, a special import path (the empty string, it turned out) denoting “the import path of this object file”. This worked well enough at the time and maintained complete command-line compatibility with existing Go usage. The changes implementing this transition can be found by searching the Git history for “package global name space”, which is what they eliminated. In particular, CL 190076 (a6736fa4), CL 186263 (758f2bc5), CL 193080 (1cecac81), CL 194053 (19126320), and CL 194071 (531e6b77) did the bulk of this transformation in January 2010. Later, in September 2011, we added the -p flag to the compiler for diagnostic purposes. The problem was that it was easy to create import cycles, especially in tests, and these could not be diagnosed until link time. You'd really want the compiler to diagnose these, for example if the compilation of package sort noticed it was importing a package that itself imported "sort". But the compilation of package sort didn't know its own import path, and so it could not tell whether it had found itself as a transitive dependency. Adding the -p flag solved this problem, and its use was optional, since the linker would still diagnose the import cycle in builds that had not updated to start passing -p. This was CL 4972057 (1e480cd1). There was still no go command at this point, but when we introduced the go command we made it pass -p, which it has for many years at this point. Over time, parts of the compiler began to depend on the presence of the -p flag for various reasonable purposes. For example: In CL 6497074 (041fc8bf; Oct 2012), the race detector used -p to detect packages that should not have race annotations, such as runtime/race and sync/atomic. In CL 13367052 (7276c02b; Sep 2013), a bug fix used -p to detect the compilation of package reflect. In CL 30539 (8aadcc55; Oct 2016), the compiler started using -p to identify package math, to be able to intrinsify calls to Sqrt inside that package. In CL 61019 (9daee931; Sep 2017), CL 71430 (2c1d2e06; Oct 2017), and later related CLs, the compiler started using the -p value when creating various DWARF debugging information. In CL 174657 (cc5eaf93; May 2019), the compiler started writing symbols without the magic empty string whenever -p was used, to reduce the amount of work required in the linker. In CL 179861 (dde7c770; Jun 2019), the compiler made the second argument to //go:linkname optional when -p is used, because in that case the compiler can derive an appropriate default. There are more examples. Today it is impossible to compile the Go standard library without using -p, and DWARF debug information is incomplete without using -p. All known Go build systems pass -p. In particular, the go command does, which is what nearly all Go developers invoke to build Go code. And Bazel does, for go_library rules that set the importpath attribute, which is all rules generated by Gazelle. Gccgo has an equivalent of -p and has required its use in order to disambiguate packages with the same name but different import paths since 2010. On top of all this, various parts of code generation for generics are made more complicated by needing to cope with the case where -p is not specified, even though it's essentially always specified. In summary, the current state is: - Use of the -p flag with cmd/compile is required for building the standard library, and for complete DWARF information, and to enable certain linker speedups. - The go command and Bazel, which we expect account for just about 100% of Go builds, both invoke cmd/compile with -p. - The code in cmd/compile to support builds without -p is complex and has become more complex with generics, but it is almost always dead code and therefore not worth maintaining. - Gccgo already requires its equivalent of -p in any build where two packages have the same name. All this supports the change in this CL, which makes -p required and adjusts tests that invoke cmd/compile to add -p appropriately. Future CLs will be able to remove all the code dealing with the possibility of -p not having been specified. Change-Id: I6b95b9d4cffe59c7bac82eb273ef6c4a67bb0e43 Reviewed-on: https://go-review.googlesource.com/c/go/+/391014 Trust: Russ Cox <rsc@golang.org> Run-TryBot: Russ Cox <rsc@golang.org> TryBot-Result: Gopher Robot <gobot@golang.org> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2022-03-08 16:16:35 -07:00
for i, pkg := range pkgs {
out, err := compileInDir(runcmd, longdir, flags, pkg.name, pkg.files...)
if i == errPkg {
if wantError && err == nil {
t.err = fmt.Errorf("compilation succeeded unexpectedly\n%s", out)
return
} else if !wantError && err != nil {
t.err = err
return
}
} else if err != nil {
t.err = err
return
}
var fullshort []string
cmd/compile: require -p flag The -p flag specifies the import path of the package being compiled. This CL makes it required when invoking the compiler and adjusts tests that invoke the compiler directly to conform to this new requirement. The go command already passes the flag, so it is unmodified in this CL. It is expected that any other Go build systems also already pass -p, or else they will need to arrange to do so before updating to Go 1.19. Of particular note, Bazel already does for rules with an importpath= attribute, which includes all Gazelle-generated rules. There is more cleanup possible now in cmd/compile, cmd/link, and other consumers of Go object files, but that is left to future CLs. Additional historical background follows but can be ignored. Long ago, before the go command, or modules, or any kind of versioning, symbols in Go archive files were named using just the package name, so that for example func F in math/rand and func F in crypto/rand would both be the object file symbol 'rand.F'. This led to collisions even in small source trees, which made certain packages unusable in the presence of other packages and generally was a problem for Go's goal of scaling to very large source trees. Fixing this problem required changing from package names to import paths in symbol names, which was mostly straightforward. One wrinkle, though, is that the compiler did not know the import path of the package being compiled; it only knew the package name. At the time, there was no go command, just Makefiles that people had invoking 6g (now “go tool compile”) and then copying the resulting object file to an importable location. That is, everyone had a custom build setup for Go, because there was no standard one. So it was not particularly attractive to change how the compiler was invoked, since that would break approximately every Go user at the time. Instead, we arranged for the compiler to emit, and other tools reading object files to recognize, a special import path (the empty string, it turned out) denoting “the import path of this object file”. This worked well enough at the time and maintained complete command-line compatibility with existing Go usage. The changes implementing this transition can be found by searching the Git history for “package global name space”, which is what they eliminated. In particular, CL 190076 (a6736fa4), CL 186263 (758f2bc5), CL 193080 (1cecac81), CL 194053 (19126320), and CL 194071 (531e6b77) did the bulk of this transformation in January 2010. Later, in September 2011, we added the -p flag to the compiler for diagnostic purposes. The problem was that it was easy to create import cycles, especially in tests, and these could not be diagnosed until link time. You'd really want the compiler to diagnose these, for example if the compilation of package sort noticed it was importing a package that itself imported "sort". But the compilation of package sort didn't know its own import path, and so it could not tell whether it had found itself as a transitive dependency. Adding the -p flag solved this problem, and its use was optional, since the linker would still diagnose the import cycle in builds that had not updated to start passing -p. This was CL 4972057 (1e480cd1). There was still no go command at this point, but when we introduced the go command we made it pass -p, which it has for many years at this point. Over time, parts of the compiler began to depend on the presence of the -p flag for various reasonable purposes. For example: In CL 6497074 (041fc8bf; Oct 2012), the race detector used -p to detect packages that should not have race annotations, such as runtime/race and sync/atomic. In CL 13367052 (7276c02b; Sep 2013), a bug fix used -p to detect the compilation of package reflect. In CL 30539 (8aadcc55; Oct 2016), the compiler started using -p to identify package math, to be able to intrinsify calls to Sqrt inside that package. In CL 61019 (9daee931; Sep 2017), CL 71430 (2c1d2e06; Oct 2017), and later related CLs, the compiler started using the -p value when creating various DWARF debugging information. In CL 174657 (cc5eaf93; May 2019), the compiler started writing symbols without the magic empty string whenever -p was used, to reduce the amount of work required in the linker. In CL 179861 (dde7c770; Jun 2019), the compiler made the second argument to //go:linkname optional when -p is used, because in that case the compiler can derive an appropriate default. There are more examples. Today it is impossible to compile the Go standard library without using -p, and DWARF debug information is incomplete without using -p. All known Go build systems pass -p. In particular, the go command does, which is what nearly all Go developers invoke to build Go code. And Bazel does, for go_library rules that set the importpath attribute, which is all rules generated by Gazelle. Gccgo has an equivalent of -p and has required its use in order to disambiguate packages with the same name but different import paths since 2010. On top of all this, various parts of code generation for generics are made more complicated by needing to cope with the case where -p is not specified, even though it's essentially always specified. In summary, the current state is: - Use of the -p flag with cmd/compile is required for building the standard library, and for complete DWARF information, and to enable certain linker speedups. - The go command and Bazel, which we expect account for just about 100% of Go builds, both invoke cmd/compile with -p. - The code in cmd/compile to support builds without -p is complex and has become more complex with generics, but it is almost always dead code and therefore not worth maintaining. - Gccgo already requires its equivalent of -p in any build where two packages have the same name. All this supports the change in this CL, which makes -p required and adjusts tests that invoke cmd/compile to add -p appropriately. Future CLs will be able to remove all the code dealing with the possibility of -p not having been specified. Change-Id: I6b95b9d4cffe59c7bac82eb273ef6c4a67bb0e43 Reviewed-on: https://go-review.googlesource.com/c/go/+/391014 Trust: Russ Cox <rsc@golang.org> Run-TryBot: Russ Cox <rsc@golang.org> TryBot-Result: Gopher Robot <gobot@golang.org> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2022-03-08 16:16:35 -07:00
for _, name := range pkg.files {
fullshort = append(fullshort, filepath.Join(longdir, name), name)
}
t.err = t.errorCheck(string(out), wantAuto, fullshort...)
if t.err != nil {
break
}
}
if action == "errorcheckdir" {
return
}
fallthrough
case "rundir":
// Compile all files in the directory as packages in lexicographic order.
// In case of errorcheckandrundir, ignore failed compilation of the package before the last.
// Link as if the last file is the main package, run it.
// Verify the expected output.
longdir := filepath.Join(cwd, t.goDirName())
pkgs, err := goDirPackages(longdir, singlefilepkgs)
if err != nil {
t.err = err
return
}
// Split flags into gcflags and ldflags
ldflags := []string{}
for i, fl := range flags {
if fl == "-ldflags" {
ldflags = flags[i+1:]
flags = flags[0:i]
break
}
}
cmd/compile: require -p flag The -p flag specifies the import path of the package being compiled. This CL makes it required when invoking the compiler and adjusts tests that invoke the compiler directly to conform to this new requirement. The go command already passes the flag, so it is unmodified in this CL. It is expected that any other Go build systems also already pass -p, or else they will need to arrange to do so before updating to Go 1.19. Of particular note, Bazel already does for rules with an importpath= attribute, which includes all Gazelle-generated rules. There is more cleanup possible now in cmd/compile, cmd/link, and other consumers of Go object files, but that is left to future CLs. Additional historical background follows but can be ignored. Long ago, before the go command, or modules, or any kind of versioning, symbols in Go archive files were named using just the package name, so that for example func F in math/rand and func F in crypto/rand would both be the object file symbol 'rand.F'. This led to collisions even in small source trees, which made certain packages unusable in the presence of other packages and generally was a problem for Go's goal of scaling to very large source trees. Fixing this problem required changing from package names to import paths in symbol names, which was mostly straightforward. One wrinkle, though, is that the compiler did not know the import path of the package being compiled; it only knew the package name. At the time, there was no go command, just Makefiles that people had invoking 6g (now “go tool compile”) and then copying the resulting object file to an importable location. That is, everyone had a custom build setup for Go, because there was no standard one. So it was not particularly attractive to change how the compiler was invoked, since that would break approximately every Go user at the time. Instead, we arranged for the compiler to emit, and other tools reading object files to recognize, a special import path (the empty string, it turned out) denoting “the import path of this object file”. This worked well enough at the time and maintained complete command-line compatibility with existing Go usage. The changes implementing this transition can be found by searching the Git history for “package global name space”, which is what they eliminated. In particular, CL 190076 (a6736fa4), CL 186263 (758f2bc5), CL 193080 (1cecac81), CL 194053 (19126320), and CL 194071 (531e6b77) did the bulk of this transformation in January 2010. Later, in September 2011, we added the -p flag to the compiler for diagnostic purposes. The problem was that it was easy to create import cycles, especially in tests, and these could not be diagnosed until link time. You'd really want the compiler to diagnose these, for example if the compilation of package sort noticed it was importing a package that itself imported "sort". But the compilation of package sort didn't know its own import path, and so it could not tell whether it had found itself as a transitive dependency. Adding the -p flag solved this problem, and its use was optional, since the linker would still diagnose the import cycle in builds that had not updated to start passing -p. This was CL 4972057 (1e480cd1). There was still no go command at this point, but when we introduced the go command we made it pass -p, which it has for many years at this point. Over time, parts of the compiler began to depend on the presence of the -p flag for various reasonable purposes. For example: In CL 6497074 (041fc8bf; Oct 2012), the race detector used -p to detect packages that should not have race annotations, such as runtime/race and sync/atomic. In CL 13367052 (7276c02b; Sep 2013), a bug fix used -p to detect the compilation of package reflect. In CL 30539 (8aadcc55; Oct 2016), the compiler started using -p to identify package math, to be able to intrinsify calls to Sqrt inside that package. In CL 61019 (9daee931; Sep 2017), CL 71430 (2c1d2e06; Oct 2017), and later related CLs, the compiler started using the -p value when creating various DWARF debugging information. In CL 174657 (cc5eaf93; May 2019), the compiler started writing symbols without the magic empty string whenever -p was used, to reduce the amount of work required in the linker. In CL 179861 (dde7c770; Jun 2019), the compiler made the second argument to //go:linkname optional when -p is used, because in that case the compiler can derive an appropriate default. There are more examples. Today it is impossible to compile the Go standard library without using -p, and DWARF debug information is incomplete without using -p. All known Go build systems pass -p. In particular, the go command does, which is what nearly all Go developers invoke to build Go code. And Bazel does, for go_library rules that set the importpath attribute, which is all rules generated by Gazelle. Gccgo has an equivalent of -p and has required its use in order to disambiguate packages with the same name but different import paths since 2010. On top of all this, various parts of code generation for generics are made more complicated by needing to cope with the case where -p is not specified, even though it's essentially always specified. In summary, the current state is: - Use of the -p flag with cmd/compile is required for building the standard library, and for complete DWARF information, and to enable certain linker speedups. - The go command and Bazel, which we expect account for just about 100% of Go builds, both invoke cmd/compile with -p. - The code in cmd/compile to support builds without -p is complex and has become more complex with generics, but it is almost always dead code and therefore not worth maintaining. - Gccgo already requires its equivalent of -p in any build where two packages have the same name. All this supports the change in this CL, which makes -p required and adjusts tests that invoke cmd/compile to add -p appropriately. Future CLs will be able to remove all the code dealing with the possibility of -p not having been specified. Change-Id: I6b95b9d4cffe59c7bac82eb273ef6c4a67bb0e43 Reviewed-on: https://go-review.googlesource.com/c/go/+/391014 Trust: Russ Cox <rsc@golang.org> Run-TryBot: Russ Cox <rsc@golang.org> TryBot-Result: Gopher Robot <gobot@golang.org> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2022-03-08 16:16:35 -07:00
for i, pkg := range pkgs {
_, err := compileInDir(runcmd, longdir, flags, pkg.name, pkg.files...)
// Allow this package compilation fail based on conditions below;
// its errors were checked in previous case.
if err != nil && !(wantError && action == "errorcheckandrundir" && i == len(pkgs)-2) {
t.err = err
return
}
if i == len(pkgs)-1 {
cmd/compile: require -p flag The -p flag specifies the import path of the package being compiled. This CL makes it required when invoking the compiler and adjusts tests that invoke the compiler directly to conform to this new requirement. The go command already passes the flag, so it is unmodified in this CL. It is expected that any other Go build systems also already pass -p, or else they will need to arrange to do so before updating to Go 1.19. Of particular note, Bazel already does for rules with an importpath= attribute, which includes all Gazelle-generated rules. There is more cleanup possible now in cmd/compile, cmd/link, and other consumers of Go object files, but that is left to future CLs. Additional historical background follows but can be ignored. Long ago, before the go command, or modules, or any kind of versioning, symbols in Go archive files were named using just the package name, so that for example func F in math/rand and func F in crypto/rand would both be the object file symbol 'rand.F'. This led to collisions even in small source trees, which made certain packages unusable in the presence of other packages and generally was a problem for Go's goal of scaling to very large source trees. Fixing this problem required changing from package names to import paths in symbol names, which was mostly straightforward. One wrinkle, though, is that the compiler did not know the import path of the package being compiled; it only knew the package name. At the time, there was no go command, just Makefiles that people had invoking 6g (now “go tool compile”) and then copying the resulting object file to an importable location. That is, everyone had a custom build setup for Go, because there was no standard one. So it was not particularly attractive to change how the compiler was invoked, since that would break approximately every Go user at the time. Instead, we arranged for the compiler to emit, and other tools reading object files to recognize, a special import path (the empty string, it turned out) denoting “the import path of this object file”. This worked well enough at the time and maintained complete command-line compatibility with existing Go usage. The changes implementing this transition can be found by searching the Git history for “package global name space”, which is what they eliminated. In particular, CL 190076 (a6736fa4), CL 186263 (758f2bc5), CL 193080 (1cecac81), CL 194053 (19126320), and CL 194071 (531e6b77) did the bulk of this transformation in January 2010. Later, in September 2011, we added the -p flag to the compiler for diagnostic purposes. The problem was that it was easy to create import cycles, especially in tests, and these could not be diagnosed until link time. You'd really want the compiler to diagnose these, for example if the compilation of package sort noticed it was importing a package that itself imported "sort". But the compilation of package sort didn't know its own import path, and so it could not tell whether it had found itself as a transitive dependency. Adding the -p flag solved this problem, and its use was optional, since the linker would still diagnose the import cycle in builds that had not updated to start passing -p. This was CL 4972057 (1e480cd1). There was still no go command at this point, but when we introduced the go command we made it pass -p, which it has for many years at this point. Over time, parts of the compiler began to depend on the presence of the -p flag for various reasonable purposes. For example: In CL 6497074 (041fc8bf; Oct 2012), the race detector used -p to detect packages that should not have race annotations, such as runtime/race and sync/atomic. In CL 13367052 (7276c02b; Sep 2013), a bug fix used -p to detect the compilation of package reflect. In CL 30539 (8aadcc55; Oct 2016), the compiler started using -p to identify package math, to be able to intrinsify calls to Sqrt inside that package. In CL 61019 (9daee931; Sep 2017), CL 71430 (2c1d2e06; Oct 2017), and later related CLs, the compiler started using the -p value when creating various DWARF debugging information. In CL 174657 (cc5eaf93; May 2019), the compiler started writing symbols without the magic empty string whenever -p was used, to reduce the amount of work required in the linker. In CL 179861 (dde7c770; Jun 2019), the compiler made the second argument to //go:linkname optional when -p is used, because in that case the compiler can derive an appropriate default. There are more examples. Today it is impossible to compile the Go standard library without using -p, and DWARF debug information is incomplete without using -p. All known Go build systems pass -p. In particular, the go command does, which is what nearly all Go developers invoke to build Go code. And Bazel does, for go_library rules that set the importpath attribute, which is all rules generated by Gazelle. Gccgo has an equivalent of -p and has required its use in order to disambiguate packages with the same name but different import paths since 2010. On top of all this, various parts of code generation for generics are made more complicated by needing to cope with the case where -p is not specified, even though it's essentially always specified. In summary, the current state is: - Use of the -p flag with cmd/compile is required for building the standard library, and for complete DWARF information, and to enable certain linker speedups. - The go command and Bazel, which we expect account for just about 100% of Go builds, both invoke cmd/compile with -p. - The code in cmd/compile to support builds without -p is complex and has become more complex with generics, but it is almost always dead code and therefore not worth maintaining. - Gccgo already requires its equivalent of -p in any build where two packages have the same name. All this supports the change in this CL, which makes -p required and adjusts tests that invoke cmd/compile to add -p appropriately. Future CLs will be able to remove all the code dealing with the possibility of -p not having been specified. Change-Id: I6b95b9d4cffe59c7bac82eb273ef6c4a67bb0e43 Reviewed-on: https://go-review.googlesource.com/c/go/+/391014 Trust: Russ Cox <rsc@golang.org> Run-TryBot: Russ Cox <rsc@golang.org> TryBot-Result: Gopher Robot <gobot@golang.org> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2022-03-08 16:16:35 -07:00
err = linkFile(runcmd, pkg.files[0], ldflags)
if err != nil {
t.err = err
return
}
var cmd []string
cmd = append(cmd, findExecCmd()...)
cmd = append(cmd, filepath.Join(t.tempDir, "a.exe"))
cmd = append(cmd, args...)
out, err := runcmd(cmd...)
if err != nil {
t.err = err
return
}
t.checkExpectedOutput(out)
}
}
case "runindir":
// Make a shallow copy of t.goDirName() in its own module and GOPATH, and
// run "go run ." in it. The module path (and hence import path prefix) of
// the copy is equal to the basename of the source directory.
//
// It's used when test a requires a full 'go build' in order to compile
// the sources, such as when importing multiple packages (issue29612.dir)
// or compiling a package containing assembly files (see issue15609.dir),
// but still needs to be run to verify the expected output.
tempDirIsGOPATH = true
srcDir := t.goDirName()
modName := filepath.Base(srcDir)
gopathSrcDir := filepath.Join(t.tempDir, "src", modName)
runInDir = gopathSrcDir
if err := overlayDir(gopathSrcDir, srcDir); err != nil {
t.err = err
return
}
modFile := fmt.Sprintf("module %s\ngo 1.14\n", modName)
if err := ioutil.WriteFile(filepath.Join(gopathSrcDir, "go.mod"), []byte(modFile), 0666); err != nil {
t.err = err
return
}
cmd := []string{goTool(), "run", t.goGcflags()}
if *linkshared {
cmd = append(cmd, "-linkshared")
}
cmd = append(cmd, flags...)
cmd = append(cmd, ".")
out, err := runcmd(cmd...)
if err != nil {
t.err = err
return
}
t.checkExpectedOutput(out)
case "build":
// Build Go file.
cmd := []string{goTool(), "build", t.goGcflags()}
cmd = append(cmd, flags...)
cmd = append(cmd, "-o", "a.exe", long)
_, err := runcmd(cmd...)
if err != nil {
t.err = err
}
case "builddir", "buildrundir":
// Build an executable from all the .go and .s files in a subdirectory.
// Run it and verify its output in the buildrundir case.
longdir := filepath.Join(cwd, t.goDirName())
files, dirErr := ioutil.ReadDir(longdir)
if dirErr != nil {
t.err = dirErr
break
}
var gos []string
var asms []string
for _, file := range files {
switch filepath.Ext(file.Name()) {
case ".go":
gos = append(gos, filepath.Join(longdir, file.Name()))
case ".s":
asms = append(asms, filepath.Join(longdir, file.Name()))
}
}
if len(asms) > 0 {
emptyHdrFile := filepath.Join(t.tempDir, "go_asm.h")
if err := ioutil.WriteFile(emptyHdrFile, nil, 0666); err != nil {
t.err = fmt.Errorf("write empty go_asm.h: %s", err)
return
}
cmd := []string{goTool(), "tool", "asm", "-p=main", "-gensymabis", "-o", "symabis"}
cmd = append(cmd, asms...)
_, err = runcmd(cmd...)
if err != nil {
t.err = err
break
}
}
var objs []string
cmd/compile: require -p flag The -p flag specifies the import path of the package being compiled. This CL makes it required when invoking the compiler and adjusts tests that invoke the compiler directly to conform to this new requirement. The go command already passes the flag, so it is unmodified in this CL. It is expected that any other Go build systems also already pass -p, or else they will need to arrange to do so before updating to Go 1.19. Of particular note, Bazel already does for rules with an importpath= attribute, which includes all Gazelle-generated rules. There is more cleanup possible now in cmd/compile, cmd/link, and other consumers of Go object files, but that is left to future CLs. Additional historical background follows but can be ignored. Long ago, before the go command, or modules, or any kind of versioning, symbols in Go archive files were named using just the package name, so that for example func F in math/rand and func F in crypto/rand would both be the object file symbol 'rand.F'. This led to collisions even in small source trees, which made certain packages unusable in the presence of other packages and generally was a problem for Go's goal of scaling to very large source trees. Fixing this problem required changing from package names to import paths in symbol names, which was mostly straightforward. One wrinkle, though, is that the compiler did not know the import path of the package being compiled; it only knew the package name. At the time, there was no go command, just Makefiles that people had invoking 6g (now “go tool compile”) and then copying the resulting object file to an importable location. That is, everyone had a custom build setup for Go, because there was no standard one. So it was not particularly attractive to change how the compiler was invoked, since that would break approximately every Go user at the time. Instead, we arranged for the compiler to emit, and other tools reading object files to recognize, a special import path (the empty string, it turned out) denoting “the import path of this object file”. This worked well enough at the time and maintained complete command-line compatibility with existing Go usage. The changes implementing this transition can be found by searching the Git history for “package global name space”, which is what they eliminated. In particular, CL 190076 (a6736fa4), CL 186263 (758f2bc5), CL 193080 (1cecac81), CL 194053 (19126320), and CL 194071 (531e6b77) did the bulk of this transformation in January 2010. Later, in September 2011, we added the -p flag to the compiler for diagnostic purposes. The problem was that it was easy to create import cycles, especially in tests, and these could not be diagnosed until link time. You'd really want the compiler to diagnose these, for example if the compilation of package sort noticed it was importing a package that itself imported "sort". But the compilation of package sort didn't know its own import path, and so it could not tell whether it had found itself as a transitive dependency. Adding the -p flag solved this problem, and its use was optional, since the linker would still diagnose the import cycle in builds that had not updated to start passing -p. This was CL 4972057 (1e480cd1). There was still no go command at this point, but when we introduced the go command we made it pass -p, which it has for many years at this point. Over time, parts of the compiler began to depend on the presence of the -p flag for various reasonable purposes. For example: In CL 6497074 (041fc8bf; Oct 2012), the race detector used -p to detect packages that should not have race annotations, such as runtime/race and sync/atomic. In CL 13367052 (7276c02b; Sep 2013), a bug fix used -p to detect the compilation of package reflect. In CL 30539 (8aadcc55; Oct 2016), the compiler started using -p to identify package math, to be able to intrinsify calls to Sqrt inside that package. In CL 61019 (9daee931; Sep 2017), CL 71430 (2c1d2e06; Oct 2017), and later related CLs, the compiler started using the -p value when creating various DWARF debugging information. In CL 174657 (cc5eaf93; May 2019), the compiler started writing symbols without the magic empty string whenever -p was used, to reduce the amount of work required in the linker. In CL 179861 (dde7c770; Jun 2019), the compiler made the second argument to //go:linkname optional when -p is used, because in that case the compiler can derive an appropriate default. There are more examples. Today it is impossible to compile the Go standard library without using -p, and DWARF debug information is incomplete without using -p. All known Go build systems pass -p. In particular, the go command does, which is what nearly all Go developers invoke to build Go code. And Bazel does, for go_library rules that set the importpath attribute, which is all rules generated by Gazelle. Gccgo has an equivalent of -p and has required its use in order to disambiguate packages with the same name but different import paths since 2010. On top of all this, various parts of code generation for generics are made more complicated by needing to cope with the case where -p is not specified, even though it's essentially always specified. In summary, the current state is: - Use of the -p flag with cmd/compile is required for building the standard library, and for complete DWARF information, and to enable certain linker speedups. - The go command and Bazel, which we expect account for just about 100% of Go builds, both invoke cmd/compile with -p. - The code in cmd/compile to support builds without -p is complex and has become more complex with generics, but it is almost always dead code and therefore not worth maintaining. - Gccgo already requires its equivalent of -p in any build where two packages have the same name. All this supports the change in this CL, which makes -p required and adjusts tests that invoke cmd/compile to add -p appropriately. Future CLs will be able to remove all the code dealing with the possibility of -p not having been specified. Change-Id: I6b95b9d4cffe59c7bac82eb273ef6c4a67bb0e43 Reviewed-on: https://go-review.googlesource.com/c/go/+/391014 Trust: Russ Cox <rsc@golang.org> Run-TryBot: Russ Cox <rsc@golang.org> TryBot-Result: Gopher Robot <gobot@golang.org> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2022-03-08 16:16:35 -07:00
cmd := []string{goTool(), "tool", "compile", "-p=main", "-e", "-D", ".", "-I", ".", "-o", "go.o"}
if len(asms) > 0 {
cmd = append(cmd, "-asmhdr", "go_asm.h", "-symabis", "symabis")
}
cmd = append(cmd, gos...)
_, err := runcmd(cmd...)
if err != nil {
t.err = err
break
}
objs = append(objs, "go.o")
if len(asms) > 0 {
cmd = []string{goTool(), "tool", "asm", "-p=main", "-e", "-I", ".", "-o", "asm.o"}
cmd = append(cmd, asms...)
_, err = runcmd(cmd...)
if err != nil {
t.err = err
break
}
objs = append(objs, "asm.o")
}
cmd = []string{goTool(), "tool", "pack", "c", "all.a"}
cmd = append(cmd, objs...)
_, err = runcmd(cmd...)
if err != nil {
t.err = err
break
}
cmd = []string{goTool(), "tool", "link", "-o", "a.exe", "all.a"}
_, err = runcmd(cmd...)
if err != nil {
t.err = err
break
}
if action == "buildrundir" {
cmd = append(findExecCmd(), filepath.Join(t.tempDir, "a.exe"))
out, err := runcmd(cmd...)
if err != nil {
t.err = err
break
}
t.checkExpectedOutput(out)
}
case "buildrun":
// Build an executable from Go file, then run it, verify its output.
// Useful for timeout tests where failure mode is infinite loop.
// TODO: not supported on NaCl
cmd := []string{goTool(), "build", t.goGcflags(), "-o", "a.exe"}
if *linkshared {
cmd = append(cmd, "-linkshared")
}
longdirgofile := filepath.Join(filepath.Join(cwd, t.dir), t.gofile)
cmd = append(cmd, flags...)
cmd = append(cmd, longdirgofile)
_, err := runcmd(cmd...)
if err != nil {
t.err = err
return
}
cmd = []string{"./a.exe"}
out, err := runcmd(append(cmd, args...)...)
if err != nil {
t.err = err
return
}
t.checkExpectedOutput(out)
case "run":
// Run Go file if no special go command flags are provided;
// otherwise build an executable and run it.
// Verify the output.
runInDir = ""
var out []byte
var err error
if len(flags)+len(args) == 0 && t.goGcflagsIsEmpty() && !*linkshared && goarch == runtime.GOARCH && goos == runtime.GOOS && goexp == env.GOEXPERIMENT {
// If we're not using special go command flags,
// skip all the go command machinery.
// This avoids any time the go command would
// spend checking whether, for example, the installed
// package runtime is up to date.
// Because we run lots of trivial test programs,
// the time adds up.
pkg := filepath.Join(t.tempDir, "pkg.a")
cmd/compile: require -p flag The -p flag specifies the import path of the package being compiled. This CL makes it required when invoking the compiler and adjusts tests that invoke the compiler directly to conform to this new requirement. The go command already passes the flag, so it is unmodified in this CL. It is expected that any other Go build systems also already pass -p, or else they will need to arrange to do so before updating to Go 1.19. Of particular note, Bazel already does for rules with an importpath= attribute, which includes all Gazelle-generated rules. There is more cleanup possible now in cmd/compile, cmd/link, and other consumers of Go object files, but that is left to future CLs. Additional historical background follows but can be ignored. Long ago, before the go command, or modules, or any kind of versioning, symbols in Go archive files were named using just the package name, so that for example func F in math/rand and func F in crypto/rand would both be the object file symbol 'rand.F'. This led to collisions even in small source trees, which made certain packages unusable in the presence of other packages and generally was a problem for Go's goal of scaling to very large source trees. Fixing this problem required changing from package names to import paths in symbol names, which was mostly straightforward. One wrinkle, though, is that the compiler did not know the import path of the package being compiled; it only knew the package name. At the time, there was no go command, just Makefiles that people had invoking 6g (now “go tool compile”) and then copying the resulting object file to an importable location. That is, everyone had a custom build setup for Go, because there was no standard one. So it was not particularly attractive to change how the compiler was invoked, since that would break approximately every Go user at the time. Instead, we arranged for the compiler to emit, and other tools reading object files to recognize, a special import path (the empty string, it turned out) denoting “the import path of this object file”. This worked well enough at the time and maintained complete command-line compatibility with existing Go usage. The changes implementing this transition can be found by searching the Git history for “package global name space”, which is what they eliminated. In particular, CL 190076 (a6736fa4), CL 186263 (758f2bc5), CL 193080 (1cecac81), CL 194053 (19126320), and CL 194071 (531e6b77) did the bulk of this transformation in January 2010. Later, in September 2011, we added the -p flag to the compiler for diagnostic purposes. The problem was that it was easy to create import cycles, especially in tests, and these could not be diagnosed until link time. You'd really want the compiler to diagnose these, for example if the compilation of package sort noticed it was importing a package that itself imported "sort". But the compilation of package sort didn't know its own import path, and so it could not tell whether it had found itself as a transitive dependency. Adding the -p flag solved this problem, and its use was optional, since the linker would still diagnose the import cycle in builds that had not updated to start passing -p. This was CL 4972057 (1e480cd1). There was still no go command at this point, but when we introduced the go command we made it pass -p, which it has for many years at this point. Over time, parts of the compiler began to depend on the presence of the -p flag for various reasonable purposes. For example: In CL 6497074 (041fc8bf; Oct 2012), the race detector used -p to detect packages that should not have race annotations, such as runtime/race and sync/atomic. In CL 13367052 (7276c02b; Sep 2013), a bug fix used -p to detect the compilation of package reflect. In CL 30539 (8aadcc55; Oct 2016), the compiler started using -p to identify package math, to be able to intrinsify calls to Sqrt inside that package. In CL 61019 (9daee931; Sep 2017), CL 71430 (2c1d2e06; Oct 2017), and later related CLs, the compiler started using the -p value when creating various DWARF debugging information. In CL 174657 (cc5eaf93; May 2019), the compiler started writing symbols without the magic empty string whenever -p was used, to reduce the amount of work required in the linker. In CL 179861 (dde7c770; Jun 2019), the compiler made the second argument to //go:linkname optional when -p is used, because in that case the compiler can derive an appropriate default. There are more examples. Today it is impossible to compile the Go standard library without using -p, and DWARF debug information is incomplete without using -p. All known Go build systems pass -p. In particular, the go command does, which is what nearly all Go developers invoke to build Go code. And Bazel does, for go_library rules that set the importpath attribute, which is all rules generated by Gazelle. Gccgo has an equivalent of -p and has required its use in order to disambiguate packages with the same name but different import paths since 2010. On top of all this, various parts of code generation for generics are made more complicated by needing to cope with the case where -p is not specified, even though it's essentially always specified. In summary, the current state is: - Use of the -p flag with cmd/compile is required for building the standard library, and for complete DWARF information, and to enable certain linker speedups. - The go command and Bazel, which we expect account for just about 100% of Go builds, both invoke cmd/compile with -p. - The code in cmd/compile to support builds without -p is complex and has become more complex with generics, but it is almost always dead code and therefore not worth maintaining. - Gccgo already requires its equivalent of -p in any build where two packages have the same name. All this supports the change in this CL, which makes -p required and adjusts tests that invoke cmd/compile to add -p appropriately. Future CLs will be able to remove all the code dealing with the possibility of -p not having been specified. Change-Id: I6b95b9d4cffe59c7bac82eb273ef6c4a67bb0e43 Reviewed-on: https://go-review.googlesource.com/c/go/+/391014 Trust: Russ Cox <rsc@golang.org> Run-TryBot: Russ Cox <rsc@golang.org> TryBot-Result: Gopher Robot <gobot@golang.org> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2022-03-08 16:16:35 -07:00
if _, err := runcmd(goTool(), "tool", "compile", "-p=main", "-o", pkg, t.goFileName()); err != nil {
t.err = err
return
}
exe := filepath.Join(t.tempDir, "test.exe")
cmd := []string{goTool(), "tool", "link", "-s", "-w"}
cmd = append(cmd, "-o", exe, pkg)
if _, err := runcmd(cmd...); err != nil {
t.err = err
return
}
out, err = runcmd(append([]string{exe}, args...)...)
} else {
cmd := []string{goTool(), "run", t.goGcflags()}
if *linkshared {
cmd = append(cmd, "-linkshared")
}
cmd = append(cmd, flags...)
cmd = append(cmd, t.goFileName())
out, err = runcmd(append(cmd, args...)...)
}
if err != nil {
t.err = err
return
}
t.checkExpectedOutput(out)
case "runoutput":
// Run Go file and write its output into temporary Go file.
// Run generated Go file and verify its output.
rungatec <- true
defer func() {
<-rungatec
}()
runInDir = ""
cmd := []string{goTool(), "run", t.goGcflags()}
if *linkshared {
cmd = append(cmd, "-linkshared")
}
cmd = append(cmd, t.goFileName())
out, err := runcmd(append(cmd, args...)...)
if err != nil {
t.err = err
return
}
tfile := filepath.Join(t.tempDir, "tmp__.go")
if err := ioutil.WriteFile(tfile, out, 0666); err != nil {
t.err = fmt.Errorf("write tempfile:%s", err)
return
}
cmd = []string{goTool(), "run", t.goGcflags()}
if *linkshared {
cmd = append(cmd, "-linkshared")
}
cmd = append(cmd, tfile)
out, err = runcmd(cmd...)
if err != nil {
t.err = err
return
}
t.checkExpectedOutput(out)
case "errorcheckoutput":
// Run Go file and write its output into temporary Go file.
// Compile and errorCheck generated Go file.
runInDir = ""
cmd := []string{goTool(), "run", t.goGcflags()}
if *linkshared {
cmd = append(cmd, "-linkshared")
}
cmd = append(cmd, t.goFileName())
out, err := runcmd(append(cmd, args...)...)
if err != nil {
t.err = err
return
}
tfile := filepath.Join(t.tempDir, "tmp__.go")
err = ioutil.WriteFile(tfile, out, 0666)
if err != nil {
t.err = fmt.Errorf("write tempfile:%s", err)
return
}
cmd/compile: require -p flag The -p flag specifies the import path of the package being compiled. This CL makes it required when invoking the compiler and adjusts tests that invoke the compiler directly to conform to this new requirement. The go command already passes the flag, so it is unmodified in this CL. It is expected that any other Go build systems also already pass -p, or else they will need to arrange to do so before updating to Go 1.19. Of particular note, Bazel already does for rules with an importpath= attribute, which includes all Gazelle-generated rules. There is more cleanup possible now in cmd/compile, cmd/link, and other consumers of Go object files, but that is left to future CLs. Additional historical background follows but can be ignored. Long ago, before the go command, or modules, or any kind of versioning, symbols in Go archive files were named using just the package name, so that for example func F in math/rand and func F in crypto/rand would both be the object file symbol 'rand.F'. This led to collisions even in small source trees, which made certain packages unusable in the presence of other packages and generally was a problem for Go's goal of scaling to very large source trees. Fixing this problem required changing from package names to import paths in symbol names, which was mostly straightforward. One wrinkle, though, is that the compiler did not know the import path of the package being compiled; it only knew the package name. At the time, there was no go command, just Makefiles that people had invoking 6g (now “go tool compile”) and then copying the resulting object file to an importable location. That is, everyone had a custom build setup for Go, because there was no standard one. So it was not particularly attractive to change how the compiler was invoked, since that would break approximately every Go user at the time. Instead, we arranged for the compiler to emit, and other tools reading object files to recognize, a special import path (the empty string, it turned out) denoting “the import path of this object file”. This worked well enough at the time and maintained complete command-line compatibility with existing Go usage. The changes implementing this transition can be found by searching the Git history for “package global name space”, which is what they eliminated. In particular, CL 190076 (a6736fa4), CL 186263 (758f2bc5), CL 193080 (1cecac81), CL 194053 (19126320), and CL 194071 (531e6b77) did the bulk of this transformation in January 2010. Later, in September 2011, we added the -p flag to the compiler for diagnostic purposes. The problem was that it was easy to create import cycles, especially in tests, and these could not be diagnosed until link time. You'd really want the compiler to diagnose these, for example if the compilation of package sort noticed it was importing a package that itself imported "sort". But the compilation of package sort didn't know its own import path, and so it could not tell whether it had found itself as a transitive dependency. Adding the -p flag solved this problem, and its use was optional, since the linker would still diagnose the import cycle in builds that had not updated to start passing -p. This was CL 4972057 (1e480cd1). There was still no go command at this point, but when we introduced the go command we made it pass -p, which it has for many years at this point. Over time, parts of the compiler began to depend on the presence of the -p flag for various reasonable purposes. For example: In CL 6497074 (041fc8bf; Oct 2012), the race detector used -p to detect packages that should not have race annotations, such as runtime/race and sync/atomic. In CL 13367052 (7276c02b; Sep 2013), a bug fix used -p to detect the compilation of package reflect. In CL 30539 (8aadcc55; Oct 2016), the compiler started using -p to identify package math, to be able to intrinsify calls to Sqrt inside that package. In CL 61019 (9daee931; Sep 2017), CL 71430 (2c1d2e06; Oct 2017), and later related CLs, the compiler started using the -p value when creating various DWARF debugging information. In CL 174657 (cc5eaf93; May 2019), the compiler started writing symbols without the magic empty string whenever -p was used, to reduce the amount of work required in the linker. In CL 179861 (dde7c770; Jun 2019), the compiler made the second argument to //go:linkname optional when -p is used, because in that case the compiler can derive an appropriate default. There are more examples. Today it is impossible to compile the Go standard library without using -p, and DWARF debug information is incomplete without using -p. All known Go build systems pass -p. In particular, the go command does, which is what nearly all Go developers invoke to build Go code. And Bazel does, for go_library rules that set the importpath attribute, which is all rules generated by Gazelle. Gccgo has an equivalent of -p and has required its use in order to disambiguate packages with the same name but different import paths since 2010. On top of all this, various parts of code generation for generics are made more complicated by needing to cope with the case where -p is not specified, even though it's essentially always specified. In summary, the current state is: - Use of the -p flag with cmd/compile is required for building the standard library, and for complete DWARF information, and to enable certain linker speedups. - The go command and Bazel, which we expect account for just about 100% of Go builds, both invoke cmd/compile with -p. - The code in cmd/compile to support builds without -p is complex and has become more complex with generics, but it is almost always dead code and therefore not worth maintaining. - Gccgo already requires its equivalent of -p in any build where two packages have the same name. All this supports the change in this CL, which makes -p required and adjusts tests that invoke cmd/compile to add -p appropriately. Future CLs will be able to remove all the code dealing with the possibility of -p not having been specified. Change-Id: I6b95b9d4cffe59c7bac82eb273ef6c4a67bb0e43 Reviewed-on: https://go-review.googlesource.com/c/go/+/391014 Trust: Russ Cox <rsc@golang.org> Run-TryBot: Russ Cox <rsc@golang.org> TryBot-Result: Gopher Robot <gobot@golang.org> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2022-03-08 16:16:35 -07:00
cmdline := []string{goTool(), "tool", "compile", "-p=p", "-d=panic", "-e", "-o", "a.o"}
cmdline = append(cmdline, flags...)
cmdline = append(cmdline, tfile)
out, err = runcmd(cmdline...)
if wantError {
if err == nil {
t.err = fmt.Errorf("compilation succeeded unexpectedly\n%s", out)
return
}
} else {
if err != nil {
t.err = err
return
}
}
t.err = t.errorCheck(string(out), false, tfile, "tmp__.go")
return
}
}
var execCmd []string
func findExecCmd() []string {
if execCmd != nil {
return execCmd
}
execCmd = []string{} // avoid work the second time
if goos == runtime.GOOS && goarch == runtime.GOARCH {
return execCmd
}
path, err := exec.LookPath(fmt.Sprintf("go_%s_%s_exec", goos, goarch))
if err == nil {
execCmd = []string{path}
}
return execCmd
}
func (t *test) String() string {
return filepath.Join(t.dir, t.gofile)
}
func (t *test) makeTempDir() {
var err error
t.tempDir, err = ioutil.TempDir("", "")
if err != nil {
log.Fatal(err)
}
if *keep {
log.Printf("Temporary directory is %s", t.tempDir)
}
err = os.Mkdir(filepath.Join(t.tempDir, "test"), 0o755)
if err != nil {
log.Fatal(err)
}
}
// checkExpectedOutput compares the output from compiling and/or running with the contents
// of the corresponding reference output file, if any (replace ".go" with ".out").
// If they don't match, fail with an informative message.
func (t *test) checkExpectedOutput(gotBytes []byte) {
got := string(gotBytes)
filename := filepath.Join(t.dir, t.gofile)
filename = filename[:len(filename)-len(".go")]
filename += ".out"
b, err := ioutil.ReadFile(filename)
// File is allowed to be missing (err != nil) in which case output should be empty.
got = strings.Replace(got, "\r\n", "\n", -1)
if got != string(b) {
if err == nil {
t.err = fmt.Errorf("output does not match expected in %s. Instead saw\n%s", filename, got)
} else {
t.err = fmt.Errorf("output should be empty when (optional) expected-output file %s is not present. Instead saw\n%s", filename, got)
}
}
}
func splitOutput(out string, wantAuto bool) []string {
// gc error messages continue onto additional lines with leading tabs.
// Split the output at the beginning of each line that doesn't begin with a tab.
// <autogenerated> lines are impossible to match so those are filtered out.
var res []string
for _, line := range strings.Split(out, "\n") {
if strings.HasSuffix(line, "\r") { // remove '\r', output by compiler on windows
line = line[:len(line)-1]
}
if strings.HasPrefix(line, "\t") {
res[len(res)-1] += "\n" + line
} else if strings.HasPrefix(line, "go tool") || strings.HasPrefix(line, "#") || !wantAuto && strings.HasPrefix(line, "<autogenerated>") {
continue
} else if strings.TrimSpace(line) != "" {
res = append(res, line)
}
}
return res
}
// errorCheck matches errors in outStr against comments in source files.
// For each line of the source files which should generate an error,
// there should be a comment of the form // ERROR "regexp".
// If outStr has an error for a line which has no such comment,
// this function will report an error.
// Likewise if outStr does not have an error for a line which has a comment,
// or if the error message does not match the <regexp>.
// The <regexp> syntax is Perl but it's best to stick to egrep.
//
// Sources files are supplied as fullshort slice.
// It consists of pairs: full path to source file and its base name.
func (t *test) errorCheck(outStr string, wantAuto bool, fullshort ...string) (err error) {
defer func() {
if *verbose && err != nil {
log.Printf("%s gc output:\n%s", t, outStr)
}
}()
var errs []error
out := splitOutput(outStr, wantAuto)
// Cut directory name.
for i := range out {
for j := 0; j < len(fullshort); j += 2 {
full, short := fullshort[j], fullshort[j+1]
out[i] = strings.Replace(out[i], full, short, -1)
}
}
var want []wantedError
for j := 0; j < len(fullshort); j += 2 {
full, short := fullshort[j], fullshort[j+1]
want = append(want, t.wantedErrors(full, short)...)
}
for _, we := range want {
var errmsgs []string
if we.auto {
errmsgs, out = partitionStrings("<autogenerated>", out)
} else {
errmsgs, out = partitionStrings(we.prefix, out)
}
if len(errmsgs) == 0 {
errs = append(errs, fmt.Errorf("%s:%d: missing error %q", we.file, we.lineNum, we.reStr))
continue
}
matched := false
n := len(out)
for _, errmsg := range errmsgs {
// Assume errmsg says "file:line: foo".
// Cut leading "file:line: " to avoid accidental matching of file name instead of message.
text := errmsg
if _, suffix, ok := strings.Cut(text, " "); ok {
text = suffix
}
if we.re.MatchString(text) {
matched = true
} else {
out = append(out, errmsg)
}
}
if !matched {
errs = append(errs, fmt.Errorf("%s:%d: no match for %#q in:\n\t%s", we.file, we.lineNum, we.reStr, strings.Join(out[n:], "\n\t")))
continue
}
}
if len(out) > 0 {
errs = append(errs, fmt.Errorf("Unmatched Errors:"))
for _, errLine := range out {
errs = append(errs, fmt.Errorf("%s", errLine))
}
}
if len(errs) == 0 {
return nil
}
if len(errs) == 1 {
return errs[0]
}
var buf bytes.Buffer
fmt.Fprintf(&buf, "\n")
for _, err := range errs {
fmt.Fprintf(&buf, "%s\n", err.Error())
}
return errors.New(buf.String())
}
func (t *test) updateErrors(out, file string) {
base := path.Base(file)
// Read in source file.
src, err := ioutil.ReadFile(file)
if err != nil {
fmt.Fprintln(os.Stderr, err)
return
}
lines := strings.Split(string(src), "\n")
// Remove old errors.
for i := range lines {
lines[i], _, _ = strings.Cut(lines[i], " // ERROR ")
}
// Parse new errors.
errors := make(map[int]map[string]bool)
tmpRe := regexp.MustCompile(`autotmp_[0-9]+`)
for _, errStr := range splitOutput(out, false) {
errFile, rest, ok := strings.Cut(errStr, ":")
if !ok || errFile != file {
continue
}
lineStr, msg, ok := strings.Cut(rest, ":")
if !ok {
continue
}
line, err := strconv.Atoi(lineStr)
line--
if err != nil || line < 0 || line >= len(lines) {
continue
}
msg = strings.Replace(msg, file, base, -1) // normalize file mentions in error itself
msg = strings.TrimLeft(msg, " \t")
for _, r := range []string{`\`, `*`, `+`, `?`, `[`, `]`, `(`, `)`} {
msg = strings.Replace(msg, r, `\`+r, -1)
}
msg = strings.Replace(msg, `"`, `.`, -1)
msg = tmpRe.ReplaceAllLiteralString(msg, `autotmp_[0-9]+`)
if errors[line] == nil {
errors[line] = make(map[string]bool)
}
errors[line][msg] = true
}
// Add new errors.
for line, errs := range errors {
var sorted []string
for e := range errs {
sorted = append(sorted, e)
}
sort.Strings(sorted)
lines[line] += " // ERROR"
for _, e := range sorted {
lines[line] += fmt.Sprintf(` "%s$"`, e)
}
}
// Write new file.
err = ioutil.WriteFile(file, []byte(strings.Join(lines, "\n")), 0640)
if err != nil {
fmt.Fprintln(os.Stderr, err)
return
}
// Polish.
exec.Command(goTool(), "fmt", file).CombinedOutput()
}
// matchPrefix reports whether s is of the form ^(.*/)?prefix(:|[),
// That is, it needs the file name prefix followed by a : or a [,
// and possibly preceded by a directory name.
func matchPrefix(s, prefix string) bool {
i := strings.Index(s, ":")
if i < 0 {
return false
}
j := strings.LastIndex(s[:i], "/")
s = s[j+1:]
if len(s) <= len(prefix) || s[:len(prefix)] != prefix {
return false
}
switch s[len(prefix)] {
case '[', ':':
return true
}
return false
}
func partitionStrings(prefix string, strs []string) (matched, unmatched []string) {
for _, s := range strs {
if matchPrefix(s, prefix) {
matched = append(matched, s)
} else {
unmatched = append(unmatched, s)
}
}
return
}
type wantedError struct {
reStr string
re *regexp.Regexp
lineNum int
auto bool // match <autogenerated> line
file string
prefix string
}
var (
errRx = regexp.MustCompile(`// (?:GC_)?ERROR (.*)`)
errAutoRx = regexp.MustCompile(`// (?:GC_)?ERRORAUTO (.*)`)
errQuotesRx = regexp.MustCompile(`"([^"]*)"`)
lineRx = regexp.MustCompile(`LINE(([+-])([0-9]+))?`)
)
func (t *test) wantedErrors(file, short string) (errs []wantedError) {
cache := make(map[string]*regexp.Regexp)
src, _ := ioutil.ReadFile(file)
for i, line := range strings.Split(string(src), "\n") {
lineNum := i + 1
if strings.Contains(line, "////") {
// double comment disables ERROR
continue
}
var auto bool
m := errAutoRx.FindStringSubmatch(line)
if m != nil {
auto = true
} else {
m = errRx.FindStringSubmatch(line)
}
if m == nil {
continue
}
all := m[1]
mm := errQuotesRx.FindAllStringSubmatch(all, -1)
if mm == nil {
log.Fatalf("%s:%d: invalid errchk line: %s", t.goFileName(), lineNum, line)
}
for _, m := range mm {
rx := lineRx.ReplaceAllStringFunc(m[1], func(m string) string {
n := lineNum
if strings.HasPrefix(m, "LINE+") {
delta, _ := strconv.Atoi(m[5:])
n += delta
} else if strings.HasPrefix(m, "LINE-") {
delta, _ := strconv.Atoi(m[5:])
n -= delta
}
return fmt.Sprintf("%s:%d", short, n)
})
re := cache[rx]
if re == nil {
var err error
re, err = regexp.Compile(rx)
if err != nil {
log.Fatalf("%s:%d: invalid regexp \"%s\" in ERROR line: %v", t.goFileName(), lineNum, rx, err)
}
cache[rx] = re
}
prefix := fmt.Sprintf("%s:%d", short, lineNum)
errs = append(errs, wantedError{
reStr: rx,
re: re,
prefix: prefix,
auto: auto,
lineNum: lineNum,
file: short,
})
}
}
return
}
const (
// Regexp to match a single opcode check: optionally begin with "-" (to indicate
// a negative check), followed by a string literal enclosed in "" or ``. For "",
// backslashes must be handled.
reMatchCheck = `-?(?:\x60[^\x60]*\x60|"(?:[^"\\]|\\.)*")`
)
var (
// Regexp to split a line in code and comment, trimming spaces
rxAsmComment = regexp.MustCompile(`^\s*(.*?)\s*(?://\s*(.+)\s*)?$`)
// Regexp to extract an architecture check: architecture name (or triplet),
// followed by semi-colon, followed by a comma-separated list of opcode checks.
// Extraneous spaces are ignored.
rxAsmPlatform = regexp.MustCompile(`(\w+)(/\w+)?(/\w*)?\s*:\s*(` + reMatchCheck + `(?:\s*,\s*` + reMatchCheck + `)*)`)
// Regexp to extract a single opcoded check
rxAsmCheck = regexp.MustCompile(reMatchCheck)
// List of all architecture variants. Key is the GOARCH architecture,
// value[0] is the variant-changing environment variable, and values[1:]
// are the supported variants.
archVariants = map[string][]string{
"386": {"GO386", "sse2", "softfloat"},
"amd64": {"GOAMD64", "v1", "v2", "v3", "v4"},
"arm": {"GOARM", "5", "6", "7"},
"arm64": {},
"loong64": {},
"mips": {"GOMIPS", "hardfloat", "softfloat"},
"mips64": {"GOMIPS64", "hardfloat", "softfloat"},
"ppc64": {"GOPPC64", "power8", "power9"},
"ppc64le": {"GOPPC64", "power8", "power9"},
"s390x": {},
"wasm": {},
"riscv64": {},
}
)
// wantedAsmOpcode is a single asmcheck check
type wantedAsmOpcode struct {
fileline string // original source file/line (eg: "/path/foo.go:45")
line int // original source line
opcode *regexp.Regexp // opcode check to be performed on assembly output
negative bool // true if the check is supposed to fail rather than pass
found bool // true if the opcode check matched at least one in the output
}
// A build environment triplet separated by slashes (eg: linux/386/sse2).
// The third field can be empty if the arch does not support variants (eg: "plan9/amd64/")
type buildEnv string
// Environ returns the environment it represents in cmd.Environ() "key=val" format
// For instance, "linux/386/sse2".Environ() returns {"GOOS=linux", "GOARCH=386", "GO386=sse2"}
func (b buildEnv) Environ() []string {
fields := strings.Split(string(b), "/")
if len(fields) != 3 {
panic("invalid buildEnv string: " + string(b))
}
env := []string{"GOOS=" + fields[0], "GOARCH=" + fields[1]}
if fields[2] != "" {
env = append(env, archVariants[fields[1]][0]+"="+fields[2])
}
return env
}
// asmChecks represents all the asmcheck checks present in a test file
// The outer map key is the build triplet in which the checks must be performed.
// The inner map key represent the source file line ("filename.go:1234") at which the
// checks must be performed.
type asmChecks map[buildEnv]map[string][]wantedAsmOpcode
// Envs returns all the buildEnv in which at least one check is present
func (a asmChecks) Envs() []buildEnv {
var envs []buildEnv
for e := range a {
envs = append(envs, e)
}
sort.Slice(envs, func(i, j int) bool {
return string(envs[i]) < string(envs[j])
})
return envs
}
func (t *test) wantedAsmOpcodes(fn string) asmChecks {
ops := make(asmChecks)
comment := ""
src, _ := ioutil.ReadFile(fn)
for i, line := range strings.Split(string(src), "\n") {
matches := rxAsmComment.FindStringSubmatch(line)
code, cmt := matches[1], matches[2]
// Keep comments pending in the comment variable until
// we find a line that contains some code.
comment += " " + cmt
if code == "" {
continue
}
// Parse and extract any architecture check from comments,
// made by one architecture name and multiple checks.
lnum := fn + ":" + strconv.Itoa(i+1)
for _, ac := range rxAsmPlatform.FindAllStringSubmatch(comment, -1) {
archspec, allchecks := ac[1:4], ac[4]
var arch, subarch, os string
switch {
case archspec[2] != "": // 3 components: "linux/386/sse2"
os, arch, subarch = archspec[0], archspec[1][1:], archspec[2][1:]
case archspec[1] != "": // 2 components: "386/sse2"
os, arch, subarch = "linux", archspec[0], archspec[1][1:]
default: // 1 component: "386"
os, arch, subarch = "linux", archspec[0], ""
if arch == "wasm" {
os = "js"
}
}
if _, ok := archVariants[arch]; !ok {
log.Fatalf("%s:%d: unsupported architecture: %v", t.goFileName(), i+1, arch)
}
// Create the build environments corresponding the above specifiers
envs := make([]buildEnv, 0, 4)
if subarch != "" {
envs = append(envs, buildEnv(os+"/"+arch+"/"+subarch))
} else {
subarchs := archVariants[arch]
if len(subarchs) == 0 {
envs = append(envs, buildEnv(os+"/"+arch+"/"))
} else {
for _, sa := range archVariants[arch][1:] {
envs = append(envs, buildEnv(os+"/"+arch+"/"+sa))
}
}
}
for _, m := range rxAsmCheck.FindAllString(allchecks, -1) {
negative := false
if m[0] == '-' {
negative = true
m = m[1:]
}
rxsrc, err := strconv.Unquote(m)
if err != nil {
log.Fatalf("%s:%d: error unquoting string: %v", t.goFileName(), i+1, err)
}
// Compile the checks as regular expressions. Notice that we
// consider checks as matching from the beginning of the actual
// assembler source (that is, what is left on each line of the
// compile -S output after we strip file/line info) to avoid
// trivial bugs such as "ADD" matching "FADD". This
// doesn't remove genericity: it's still possible to write
// something like "F?ADD", but we make common cases simpler
// to get right.
oprx, err := regexp.Compile("^" + rxsrc)
if err != nil {
log.Fatalf("%s:%d: %v", t.goFileName(), i+1, err)
}
for _, env := range envs {
if ops[env] == nil {
ops[env] = make(map[string][]wantedAsmOpcode)
}
ops[env][lnum] = append(ops[env][lnum], wantedAsmOpcode{
negative: negative,
fileline: lnum,
line: i + 1,
opcode: oprx,
})
}
}
}
comment = ""
}
return ops
}
func (t *test) asmCheck(outStr string, fn string, env buildEnv, fullops map[string][]wantedAsmOpcode) (err error) {
// The assembly output contains the concatenated dump of multiple functions.
// the first line of each function begins at column 0, while the rest is
// indented by a tabulation. These data structures help us index the
// output by function.
functionMarkers := make([]int, 1)
lineFuncMap := make(map[string]int)
lines := strings.Split(outStr, "\n")
rxLine := regexp.MustCompile(fmt.Sprintf(`\((%s:\d+)\)\s+(.*)`, regexp.QuoteMeta(fn)))
for nl, line := range lines {
// Check if this line begins a function
if len(line) > 0 && line[0] != '\t' {
functionMarkers = append(functionMarkers, nl)
}
// Search if this line contains a assembly opcode (which is prefixed by the
// original source file/line in parenthesis)
matches := rxLine.FindStringSubmatch(line)
if len(matches) == 0 {
continue
}
srcFileLine, asm := matches[1], matches[2]
// Associate the original file/line information to the current
// function in the output; it will be useful to dump it in case
// of error.
lineFuncMap[srcFileLine] = len(functionMarkers) - 1
// If there are opcode checks associated to this source file/line,
// run the checks.
if ops, found := fullops[srcFileLine]; found {
for i := range ops {
if !ops[i].found && ops[i].opcode.FindString(asm) != "" {
ops[i].found = true
}
}
}
}
functionMarkers = append(functionMarkers, len(lines))
var failed []wantedAsmOpcode
for _, ops := range fullops {
for _, o := range ops {
// There's a failure if a negative match was found,
// or a positive match was not found.
if o.negative == o.found {
failed = append(failed, o)
}
}
}
if len(failed) == 0 {
return
}
// At least one asmcheck failed; report them
sort.Slice(failed, func(i, j int) bool {
return failed[i].line < failed[j].line
})
lastFunction := -1
var errbuf bytes.Buffer
fmt.Fprintln(&errbuf)
for _, o := range failed {
// Dump the function in which this opcode check was supposed to
// pass but failed.
funcIdx := lineFuncMap[o.fileline]
if funcIdx != 0 && funcIdx != lastFunction {
funcLines := lines[functionMarkers[funcIdx]:functionMarkers[funcIdx+1]]
log.Println(strings.Join(funcLines, "\n"))
lastFunction = funcIdx // avoid printing same function twice
}
if o.negative {
fmt.Fprintf(&errbuf, "%s:%d: %s: wrong opcode found: %q\n", t.goFileName(), o.line, env, o.opcode.String())
} else {
fmt.Fprintf(&errbuf, "%s:%d: %s: opcode not found: %q\n", t.goFileName(), o.line, env, o.opcode.String())
}
}
err = errors.New(errbuf.String())
return
}
// defaultRunOutputLimit returns the number of runoutput tests that
// can be executed in parallel.
func defaultRunOutputLimit() int {
const maxArmCPU = 2
cpu := runtime.NumCPU()
if runtime.GOARCH == "arm" && cpu > maxArmCPU {
cpu = maxArmCPU
}
return cpu
}
// checkShouldTest runs sanity checks on the shouldTest function.
func checkShouldTest() {
assert := func(ok bool, _ string) {
if !ok {
panic("fail")
}
}
assertNot := func(ok bool, _ string) { assert(!ok, "") }
// Simple tests.
assert(shouldTest("// +build linux", "linux", "arm"))
assert(shouldTest("// +build !windows", "linux", "arm"))
assertNot(shouldTest("// +build !windows", "windows", "amd64"))
// A file with no build tags will always be tested.
assert(shouldTest("// This is a test.", "os", "arch"))
// Build tags separated by a space are OR-ed together.
assertNot(shouldTest("// +build arm 386", "linux", "amd64"))
// Build tags separated by a comma are AND-ed together.
assertNot(shouldTest("// +build !windows,!plan9", "windows", "amd64"))
assertNot(shouldTest("// +build !windows,!plan9", "plan9", "386"))
// Build tags on multiple lines are AND-ed together.
assert(shouldTest("// +build !windows\n// +build amd64", "linux", "amd64"))
assertNot(shouldTest("// +build !windows\n// +build amd64", "windows", "amd64"))
// Test that (!a OR !b) matches anything.
assert(shouldTest("// +build !windows !plan9", "windows", "amd64"))
}
func getenv(key, def string) string {
value := os.Getenv(key)
if value != "" {
return value
}
return def
}
// overlayDir makes a minimal-overhead copy of srcRoot in which new files may be added.
func overlayDir(dstRoot, srcRoot string) error {
dstRoot = filepath.Clean(dstRoot)
if err := os.MkdirAll(dstRoot, 0777); err != nil {
return err
}
srcRoot, err := filepath.Abs(srcRoot)
if err != nil {
return err
}
return filepath.WalkDir(srcRoot, func(srcPath string, d fs.DirEntry, err error) error {
if err != nil || srcPath == srcRoot {
return err
}
suffix := strings.TrimPrefix(srcPath, srcRoot)
for len(suffix) > 0 && suffix[0] == filepath.Separator {
suffix = suffix[1:]
}
dstPath := filepath.Join(dstRoot, suffix)
var info fs.FileInfo
if d.Type()&os.ModeSymlink != 0 {
info, err = os.Stat(srcPath)
} else {
info, err = d.Info()
}
if err != nil {
return err
}
perm := info.Mode() & os.ModePerm
// Always copy directories (don't symlink them).
// If we add a file in the overlay, we don't want to add it in the original.
if info.IsDir() {
return os.MkdirAll(dstPath, perm|0200)
}
// If the OS supports symlinks, use them instead of copying bytes.
if err := os.Symlink(srcPath, dstPath); err == nil {
return nil
}
// Otherwise, copy the bytes.
src, err := os.Open(srcPath)
if err != nil {
return err
}
defer src.Close()
dst, err := os.OpenFile(dstPath, os.O_WRONLY|os.O_CREATE|os.O_EXCL, perm)
if err != nil {
return err
}
_, err = io.Copy(dst, src)
if closeErr := dst.Close(); err == nil {
err = closeErr
}
return err
})
}
// The following sets of files are excluded from testing depending on configuration.
// The types2Failures(32Bit) files pass with the 1.17 compiler but don't pass with
// the 1.18 compiler using the new types2 type checker, or pass with sub-optimal
// error(s).
// List of files that the compiler cannot errorcheck with the new typechecker (types2).
var types2Failures = setOf(
test: re-enable most go/tests that were disabled because of types2 differences I made the default be that, where there are differences between types2 and -G=0 error messages, we want errorcheck tests to pass types2. Typically, we can get errorcheck to pass on types2 and -G=0 if they give the same number of error messages on the same lines, just different wording. If they give a different number of error messages, then I made types2 pass. I added an exception list for -G=0 to cover those cases where -G=0 and types give different numbers of error messages. Because types2 does not run if there are syntax errors, for several tests, I had to split the tests into two parts in order to get all the indicated errors to be reported in types2 (bug228.go, bug388.go, issue11610.go, issue14520.go) I tried to preserve the GCCGO labeling correctly (but may have gotten some wrong). When types2 now matches where a GCCGO error previously occurred, I transformed GCCGO_ERROR -> ERROR. When types2 no longer reports an error in a certain place, I transformed ERROR -> GCCGO_ERROR. When types2 reports an error in a new place, I used GC_ERROR. The remaining entries in types2Failures are things that I think we probably still need to fix - either actually missing errors in types2, or cases where types2 gives worse errors than -G=0. Change-Id: I7f01e82b322b16094096b67d7ed2bb39b410c34f Reviewed-on: https://go-review.googlesource.com/c/go/+/372854 Trust: Dan Scales <danscales@google.com> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2021-12-03 17:10:10 -07:00
"notinheap.go", // types2 doesn't report errors about conversions that are invalid due to //go:notinheap
"shift1.go", // types2 reports two new errors which are probably not right
"fixedbugs/issue10700.go", // types2 should give hint about ptr to interface
"fixedbugs/issue18331.go", // missing error about misuse of //go:noescape (irgen needs code from noder)
"fixedbugs/issue18419.go", // types2 reports no field or method member, but should say unexported
"fixedbugs/issue20233.go", // types2 reports two instead of one error (preference: 1.17 compiler)
"fixedbugs/issue20245.go", // types2 reports two instead of one error (preference: 1.17 compiler)
test: re-enable most go/tests that were disabled because of types2 differences I made the default be that, where there are differences between types2 and -G=0 error messages, we want errorcheck tests to pass types2. Typically, we can get errorcheck to pass on types2 and -G=0 if they give the same number of error messages on the same lines, just different wording. If they give a different number of error messages, then I made types2 pass. I added an exception list for -G=0 to cover those cases where -G=0 and types give different numbers of error messages. Because types2 does not run if there are syntax errors, for several tests, I had to split the tests into two parts in order to get all the indicated errors to be reported in types2 (bug228.go, bug388.go, issue11610.go, issue14520.go) I tried to preserve the GCCGO labeling correctly (but may have gotten some wrong). When types2 now matches where a GCCGO error previously occurred, I transformed GCCGO_ERROR -> ERROR. When types2 no longer reports an error in a certain place, I transformed ERROR -> GCCGO_ERROR. When types2 reports an error in a new place, I used GC_ERROR. The remaining entries in types2Failures are things that I think we probably still need to fix - either actually missing errors in types2, or cases where types2 gives worse errors than -G=0. Change-Id: I7f01e82b322b16094096b67d7ed2bb39b410c34f Reviewed-on: https://go-review.googlesource.com/c/go/+/372854 Trust: Dan Scales <danscales@google.com> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2021-12-03 17:10:10 -07:00
"fixedbugs/issue31053.go", // types2 reports "unknown field" instead of "cannot refer to unexported field"
)
var types2Failures32Bit = setOf(
"printbig.go", // large untyped int passed to print (32-bit)
"fixedbugs/bug114.go", // large untyped int passed to println (32-bit)
"fixedbugs/issue23305.go", // large untyped int passed to println (32-bit)
)
var go118Failures = setOf(
"typeparam/nested.go", // 1.18 compiler doesn't support function-local types with generics
"typeparam/issue51521.go", // 1.18 compiler produces bad panic message and link error
)
// In all of these cases, the 1.17 compiler reports reasonable errors, but either the
// 1.17 or 1.18 compiler report extra errors, so we can't match correctly on both. We
// now set the patterns to match correctly on all the 1.18 errors.
// This list remains here just as a reference and for comparison - these files all pass.
var _ = setOf(
test: re-enable most go/tests that were disabled because of types2 differences I made the default be that, where there are differences between types2 and -G=0 error messages, we want errorcheck tests to pass types2. Typically, we can get errorcheck to pass on types2 and -G=0 if they give the same number of error messages on the same lines, just different wording. If they give a different number of error messages, then I made types2 pass. I added an exception list for -G=0 to cover those cases where -G=0 and types give different numbers of error messages. Because types2 does not run if there are syntax errors, for several tests, I had to split the tests into two parts in order to get all the indicated errors to be reported in types2 (bug228.go, bug388.go, issue11610.go, issue14520.go) I tried to preserve the GCCGO labeling correctly (but may have gotten some wrong). When types2 now matches where a GCCGO error previously occurred, I transformed GCCGO_ERROR -> ERROR. When types2 no longer reports an error in a certain place, I transformed ERROR -> GCCGO_ERROR. When types2 reports an error in a new place, I used GC_ERROR. The remaining entries in types2Failures are things that I think we probably still need to fix - either actually missing errors in types2, or cases where types2 gives worse errors than -G=0. Change-Id: I7f01e82b322b16094096b67d7ed2bb39b410c34f Reviewed-on: https://go-review.googlesource.com/c/go/+/372854 Trust: Dan Scales <danscales@google.com> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2021-12-03 17:10:10 -07:00
"import1.go", // types2 reports extra errors
"initializerr.go", // types2 reports extra error
"typecheck.go", // types2 reports extra error at function call
"fixedbugs/bug176.go", // types2 reports all errors (pref: types2)
"fixedbugs/bug195.go", // types2 reports slight different errors, and an extra error
"fixedbugs/bug412.go", // types2 produces a follow-on error
"fixedbugs/issue11614.go", // types2 reports an extra error
"fixedbugs/issue17038.go", // types2 doesn't report a follow-on error (pref: types2)
"fixedbugs/issue23732.go", // types2 reports different (but ok) line numbers
"fixedbugs/issue4510.go", // types2 reports different (but ok) line numbers
"fixedbugs/issue7525b.go", // types2 reports init cycle error on different line - ok otherwise
"fixedbugs/issue7525c.go", // types2 reports init cycle error on different line - ok otherwise
"fixedbugs/issue7525d.go", // types2 reports init cycle error on different line - ok otherwise
"fixedbugs/issue7525e.go", // types2 reports init cycle error on different line - ok otherwise
"fixedbugs/issue7525.go", // types2 reports init cycle error on different line - ok otherwise
)
var unifiedFailures = setOf(
"closure3.go", // unified IR numbers closures differently than -d=inlfuncswithclosures
"escape4.go", // unified IR can inline f5 and f6; test doesn't expect this
"inline.go", // unified IR reports function literal diagnostics on different lines than -d=inlfuncswithclosures
"linkname3.go", // unified IR is missing some linkname errors
cmd/compile: report channel size errors correctly for -G=3 First, we need to set base.Pos in varDecl() and typeDecl(), so it will be correct if we need to report type size errors while converting types. Changed error calls in types/sizes.go to use Errorf, not ErrorfAt, since we want to use base.Pos (which will set from t.Pos(), if that is available). Second, we need to add an extra call CalcSize(t1.Elem()) in the TCHANARGS case of CalcSize(). We can use CalcSize() rather than CheckSize(), since we know the top-level recursive type will have been calculated by the time we process the fake TCHANARGS type. In -G=0 mode, the size of the channel element has often been calculated because of some other processing (but not in the case of #49767). But in -G=3 mode, we just calculate sizes during the single noder2 pass, so we are more likely to have not gotten to calculating the size of the element yet, depending on the order of processing of the deferredTypeStack. Fixes the tests fixedbugs/issue{42058a,42058b}.go that were disabled for -G=3 mode. Had to add exceptions in stdlib_test.go for go/types and types2, because the types2 typechecker does not know about type size limits. Fixes #49814 Fixes #49771 Updates #49767 Change-Id: I77d058e8ceff68a58c4c386a8cf46799c54b04c3 Reviewed-on: https://go-review.googlesource.com/c/go/+/367955 Trust: Dan Scales <danscales@google.com> Run-TryBot: Dan Scales <danscales@google.com> TryBot-Result: Gopher Robot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
2021-11-29 12:45:17 -07:00
"fixedbugs/issue42284.go", // prints "T(0) does not escape", but test expects "a.I(a.T(0)) does not escape"
"fixedbugs/issue7921.go", // prints "… escapes to heap", but test expects "string(…) escapes to heap"
"typeparam/issue47631.go", // unified IR can handle local type declarations
"fixedbugs/issue42058a.go", // unified IR doesn't report channel element too large
"fixedbugs/issue42058b.go", // unified IR doesn't report channel element too large
"fixedbugs/issue49767.go", // unified IR doesn't report channel element too large
"fixedbugs/issue49814.go", // unified IR doesn't report array type too large
)
func setOf(keys ...string) map[string]bool {
m := make(map[string]bool, len(keys))
for _, key := range keys {
m[key] = true
}
return m
}
// splitQuoted splits the string s around each instance of one or more consecutive
// white space characters while taking into account quotes and escaping, and
// returns an array of substrings of s or an empty list if s contains only white space.
// Single quotes and double quotes are recognized to prevent splitting within the
// quoted region, and are removed from the resulting substrings. If a quote in s
// isn't closed err will be set and r will have the unclosed argument as the
// last element. The backslash is used for escaping.
//
// For example, the following string:
//
// a b:"c d" 'e''f' "g\""
//
// Would be parsed as:
//
// []string{"a", "b:c d", "ef", `g"`}
//
// [copied from src/go/build/build.go]
func splitQuoted(s string) (r []string, err error) {
var args []string
arg := make([]rune, len(s))
escaped := false
quoted := false
quote := '\x00'
i := 0
for _, rune := range s {
switch {
case escaped:
escaped = false
case rune == '\\':
escaped = true
continue
case quote != '\x00':
if rune == quote {
quote = '\x00'
continue
}
case rune == '"' || rune == '\'':
quoted = true
quote = rune
continue
case unicode.IsSpace(rune):
if quoted || i > 0 {
quoted = false
args = append(args, string(arg[:i]))
i = 0
}
continue
}
arg[i] = rune
i++
}
if quoted || i > 0 {
args = append(args, string(arg[:i]))
}
if quote != 0 {
err = errors.New("unclosed quote")
} else if escaped {
err = errors.New("unfinished escaping")
}
return args, err
}