1
0
mirror of https://github.com/golang/go synced 2024-11-06 05:36:13 -07:00
go/internal/lsp/cache/mod.go

358 lines
9.4 KiB
Go
Raw Normal View History

// Copyright 2019 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
package cache
import (
"context"
"fmt"
"os"
"path/filepath"
"regexp"
"strconv"
"strings"
"golang.org/x/mod/modfile"
"golang.org/x/tools/internal/event"
"golang.org/x/tools/internal/lsp/debug/tag"
"golang.org/x/tools/internal/lsp/protocol"
"golang.org/x/tools/internal/lsp/source"
"golang.org/x/tools/internal/memoize"
"golang.org/x/tools/internal/span"
errors "golang.org/x/xerrors"
)
const (
SyntaxError = "syntax"
)
type parseModHandle struct {
handle *memoize.Handle
}
type parseModData struct {
parsed *source.ParsedModule
// err is any error encountered while parsing the file.
err error
}
internal/memoize: switch from GC-driven to explicit deletion The GC-based cache has given us a number of problems. First, memory leaks driven by reference cycles: the Go runtime cannot collect cycles involving finalizers, which prevents us from writing natural code in Bind callbacks. If we screw it up, we get a mysterious leak that takes a long time to track down. Second, the behavior is generally mysterious; it's hard to predict how long a value lasts, and harder to tell if a value being live is a bug. Third, we think that it may be interacting poorly with the GC, resulting in unnecessary memory usage. The structure of the values we put in the cache is not actually that complicated -- there are only 5 significant types: parse, typecheck, analyze, parse mod, and analyze mod. Managing them manually should not be conceptually difficult, and in fact we already do most of the work in (*snapshot).clone. In this CL the cache adds the concept of "generations", which function as reference counts on cache entries. Entries are still global and shared across generations, but will be explicitly deleted once no generations refer to them. The idea is that each snapshot is a new generation, and can inherit entries from the previous snapshot or leave them behind to be deleted. One obvious risk of this scheme is that we'll leave dangling references to values without actually inheriting them across generations. To prevent that, getting a value requires passing in the generation at which it's being read, and an error will be returned if that generation is dead. Change-Id: I4b30891efd7be4e10f2b84f4c067b0dee43dcf9c Reviewed-on: https://go-review.googlesource.com/c/tools/+/242838 Run-TryBot: Heschi Kreinick <heschi@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Rebecca Stambler <rstambler@golang.org> Reviewed-by: Robert Findley <rfindley@google.com>
2020-07-24 15:17:13 -06:00
func (mh *parseModHandle) parse(ctx context.Context, snapshot *snapshot) (*source.ParsedModule, error) {
v, err := mh.handle.Get(ctx, snapshot.generation, snapshot)
if err != nil {
return nil, err
}
data := v.(*parseModData)
return data.parsed, data.err
}
func (s *snapshot) ParseMod(ctx context.Context, modFH source.FileHandle) (*source.ParsedModule, error) {
if handle := s.getModHandle(modFH.URI()); handle != nil {
return handle.parse(ctx, s)
}
internal/memoize: switch from GC-driven to explicit deletion The GC-based cache has given us a number of problems. First, memory leaks driven by reference cycles: the Go runtime cannot collect cycles involving finalizers, which prevents us from writing natural code in Bind callbacks. If we screw it up, we get a mysterious leak that takes a long time to track down. Second, the behavior is generally mysterious; it's hard to predict how long a value lasts, and harder to tell if a value being live is a bug. Third, we think that it may be interacting poorly with the GC, resulting in unnecessary memory usage. The structure of the values we put in the cache is not actually that complicated -- there are only 5 significant types: parse, typecheck, analyze, parse mod, and analyze mod. Managing them manually should not be conceptually difficult, and in fact we already do most of the work in (*snapshot).clone. In this CL the cache adds the concept of "generations", which function as reference counts on cache entries. Entries are still global and shared across generations, but will be explicitly deleted once no generations refer to them. The idea is that each snapshot is a new generation, and can inherit entries from the previous snapshot or leave them behind to be deleted. One obvious risk of this scheme is that we'll leave dangling references to values without actually inheriting them across generations. To prevent that, getting a value requires passing in the generation at which it's being read, and an error will be returned if that generation is dead. Change-Id: I4b30891efd7be4e10f2b84f4c067b0dee43dcf9c Reviewed-on: https://go-review.googlesource.com/c/tools/+/242838 Run-TryBot: Heschi Kreinick <heschi@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Rebecca Stambler <rstambler@golang.org> Reviewed-by: Robert Findley <rfindley@google.com>
2020-07-24 15:17:13 -06:00
h := s.generation.Bind(modFH.FileIdentity(), func(ctx context.Context, _ memoize.Arg) interface{} {
_, done := event.Start(ctx, "cache.ParseModHandle", tag.URI.Of(modFH.URI()))
defer done()
contents, err := modFH.Read()
if err != nil {
return &parseModData{err: err}
}
m := &protocol.ColumnMapper{
URI: modFH.URI(),
Converter: span.NewContentConverter(modFH.URI().Filename(), contents),
Content: contents,
}
data := &parseModData{
parsed: &source.ParsedModule{
Mapper: m,
},
}
data.parsed.File, data.err = modfile.Parse(modFH.URI().Filename(), contents, nil)
if data.err != nil {
// Attempt to convert the error to a non-fatal parse error.
if parseErr, extractErr := extractModParseErrors(modFH.URI(), m, data.err, contents); extractErr == nil {
data.err = nil
data.parsed.ParseErrors = []source.Error{*parseErr}
}
}
return data
})
pmh := &parseModHandle{handle: h}
s.mu.Lock()
s.parseModHandles[modFH.URI()] = pmh
s.mu.Unlock()
return pmh.parse(ctx, s)
}
func (s *snapshot) sumFH(ctx context.Context, modFH source.FileHandle) (source.FileHandle, error) {
// Get the go.sum file, either from the snapshot or directly from the
// cache. Avoid (*snapshot).GetFile here, as we don't want to add
// nonexistent file handles to the snapshot if the file does not exist.
sumURI := span.URIFromPath(sumFilename(modFH.URI()))
var sumFH source.FileHandle = s.FindFile(sumURI)
if sumFH == nil {
var err error
sumFH, err = s.view.session.cache.getFile(ctx, sumURI)
if err != nil {
return nil, err
}
}
_, err := sumFH.Read()
if err != nil {
return nil, err
}
return sumFH, nil
}
func sumFilename(modURI span.URI) string {
return strings.TrimSuffix(modURI.Filename(), ".mod") + ".sum"
}
// extractModParseErrors processes the raw errors returned by modfile.Parse,
// extracting the filenames and line numbers that correspond to the errors.
func extractModParseErrors(uri span.URI, m *protocol.ColumnMapper, parseErr error, content []byte) (*source.Error, error) {
re := regexp.MustCompile(`.*:([\d]+): (.+)`)
matches := re.FindStringSubmatch(strings.TrimSpace(parseErr.Error()))
if len(matches) < 3 {
return nil, errors.Errorf("could not parse go.mod error message: %s", parseErr)
}
line, err := strconv.Atoi(matches[1])
if err != nil {
return nil, err
}
lines := strings.Split(string(content), "\n")
if line > len(lines) {
return nil, errors.Errorf("could not parse go.mod error message %q, line number %v out of range", content, line)
}
// The error returned from the modfile package only returns a line number,
// so we assume that the diagnostic should be for the entire line.
endOfLine := len(lines[line-1])
sOffset, err := m.Converter.ToOffset(line, 0)
if err != nil {
return nil, err
}
eOffset, err := m.Converter.ToOffset(line, endOfLine)
if err != nil {
return nil, err
}
spn := span.New(uri, span.NewPoint(line, 0, sOffset), span.NewPoint(line, endOfLine, eOffset))
rng, err := m.Range(spn)
if err != nil {
return nil, err
}
return &source.Error{
Category: SyntaxError,
Message: matches[2],
Range: rng,
URI: uri,
}, nil
}
// modKey is uniquely identifies cached data for `go mod why` or dependencies
// to upgrade.
type modKey struct {
sessionID, cfg, view string
mod source.FileIdentity
verb modAction
}
type modAction int
const (
why modAction = iota
upgrade
)
type modWhyHandle struct {
handle *memoize.Handle
}
type modWhyData struct {
// why keeps track of the `go mod why` results for each require statement
// in the go.mod file.
why map[string]string
err error
}
internal/memoize: switch from GC-driven to explicit deletion The GC-based cache has given us a number of problems. First, memory leaks driven by reference cycles: the Go runtime cannot collect cycles involving finalizers, which prevents us from writing natural code in Bind callbacks. If we screw it up, we get a mysterious leak that takes a long time to track down. Second, the behavior is generally mysterious; it's hard to predict how long a value lasts, and harder to tell if a value being live is a bug. Third, we think that it may be interacting poorly with the GC, resulting in unnecessary memory usage. The structure of the values we put in the cache is not actually that complicated -- there are only 5 significant types: parse, typecheck, analyze, parse mod, and analyze mod. Managing them manually should not be conceptually difficult, and in fact we already do most of the work in (*snapshot).clone. In this CL the cache adds the concept of "generations", which function as reference counts on cache entries. Entries are still global and shared across generations, but will be explicitly deleted once no generations refer to them. The idea is that each snapshot is a new generation, and can inherit entries from the previous snapshot or leave them behind to be deleted. One obvious risk of this scheme is that we'll leave dangling references to values without actually inheriting them across generations. To prevent that, getting a value requires passing in the generation at which it's being read, and an error will be returned if that generation is dead. Change-Id: I4b30891efd7be4e10f2b84f4c067b0dee43dcf9c Reviewed-on: https://go-review.googlesource.com/c/tools/+/242838 Run-TryBot: Heschi Kreinick <heschi@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Rebecca Stambler <rstambler@golang.org> Reviewed-by: Robert Findley <rfindley@google.com>
2020-07-24 15:17:13 -06:00
func (mwh *modWhyHandle) why(ctx context.Context, snapshot *snapshot) (map[string]string, error) {
v, err := mwh.handle.Get(ctx, snapshot.generation, snapshot)
if err != nil {
return nil, err
}
data := v.(*modWhyData)
return data.why, data.err
}
func (s *snapshot) ModWhy(ctx context.Context) (map[string]string, error) {
if err := s.awaitLoaded(ctx); err != nil {
return nil, err
}
fh, err := s.GetFile(ctx, s.view.modURI)
if err != nil {
return nil, err
}
key := modKey{
sessionID: s.view.session.id,
cfg: hashConfig(s.config(ctx)),
mod: fh.FileIdentity(),
view: s.view.root.Filename(),
verb: why,
}
internal/memoize: switch from GC-driven to explicit deletion The GC-based cache has given us a number of problems. First, memory leaks driven by reference cycles: the Go runtime cannot collect cycles involving finalizers, which prevents us from writing natural code in Bind callbacks. If we screw it up, we get a mysterious leak that takes a long time to track down. Second, the behavior is generally mysterious; it's hard to predict how long a value lasts, and harder to tell if a value being live is a bug. Third, we think that it may be interacting poorly with the GC, resulting in unnecessary memory usage. The structure of the values we put in the cache is not actually that complicated -- there are only 5 significant types: parse, typecheck, analyze, parse mod, and analyze mod. Managing them manually should not be conceptually difficult, and in fact we already do most of the work in (*snapshot).clone. In this CL the cache adds the concept of "generations", which function as reference counts on cache entries. Entries are still global and shared across generations, but will be explicitly deleted once no generations refer to them. The idea is that each snapshot is a new generation, and can inherit entries from the previous snapshot or leave them behind to be deleted. One obvious risk of this scheme is that we'll leave dangling references to values without actually inheriting them across generations. To prevent that, getting a value requires passing in the generation at which it's being read, and an error will be returned if that generation is dead. Change-Id: I4b30891efd7be4e10f2b84f4c067b0dee43dcf9c Reviewed-on: https://go-review.googlesource.com/c/tools/+/242838 Run-TryBot: Heschi Kreinick <heschi@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Rebecca Stambler <rstambler@golang.org> Reviewed-by: Robert Findley <rfindley@google.com>
2020-07-24 15:17:13 -06:00
h := s.generation.Bind(key, func(ctx context.Context, arg memoize.Arg) interface{} {
ctx, done := event.Start(ctx, "cache.ModWhyHandle", tag.URI.Of(fh.URI()))
defer done()
snapshot := arg.(*snapshot)
pm, err := snapshot.ParseMod(ctx, fh)
if err != nil {
return &modWhyData{err: err}
}
// No requires to explain.
if len(pm.File.Require) == 0 {
return &modWhyData{}
}
// Run `go mod why` on all the dependencies.
args := []string{"why", "-m"}
for _, req := range pm.File.Require {
args = append(args, req.Mod.Path)
}
stdout, err := snapshot.RunGoCommand(ctx, "mod", args)
if err != nil {
return &modWhyData{err: err}
}
whyList := strings.Split(stdout.String(), "\n\n")
if len(whyList) != len(pm.File.Require) {
return &modWhyData{
err: fmt.Errorf("mismatched number of results: got %v, want %v", len(whyList), len(pm.File.Require)),
}
}
why := make(map[string]string, len(pm.File.Require))
for i, req := range pm.File.Require {
why[req.Mod.Path] = whyList[i]
}
return &modWhyData{why: why}
})
mwh := &modWhyHandle{handle: h}
s.mu.Lock()
s.modWhyHandle = mwh
s.mu.Unlock()
return s.modWhyHandle.why(ctx, s)
}
type modUpgradeHandle struct {
handle *memoize.Handle
}
type modUpgradeData struct {
// upgrades maps modules to their latest versions.
upgrades map[string]string
err error
}
internal/memoize: switch from GC-driven to explicit deletion The GC-based cache has given us a number of problems. First, memory leaks driven by reference cycles: the Go runtime cannot collect cycles involving finalizers, which prevents us from writing natural code in Bind callbacks. If we screw it up, we get a mysterious leak that takes a long time to track down. Second, the behavior is generally mysterious; it's hard to predict how long a value lasts, and harder to tell if a value being live is a bug. Third, we think that it may be interacting poorly with the GC, resulting in unnecessary memory usage. The structure of the values we put in the cache is not actually that complicated -- there are only 5 significant types: parse, typecheck, analyze, parse mod, and analyze mod. Managing them manually should not be conceptually difficult, and in fact we already do most of the work in (*snapshot).clone. In this CL the cache adds the concept of "generations", which function as reference counts on cache entries. Entries are still global and shared across generations, but will be explicitly deleted once no generations refer to them. The idea is that each snapshot is a new generation, and can inherit entries from the previous snapshot or leave them behind to be deleted. One obvious risk of this scheme is that we'll leave dangling references to values without actually inheriting them across generations. To prevent that, getting a value requires passing in the generation at which it's being read, and an error will be returned if that generation is dead. Change-Id: I4b30891efd7be4e10f2b84f4c067b0dee43dcf9c Reviewed-on: https://go-review.googlesource.com/c/tools/+/242838 Run-TryBot: Heschi Kreinick <heschi@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Rebecca Stambler <rstambler@golang.org> Reviewed-by: Robert Findley <rfindley@google.com>
2020-07-24 15:17:13 -06:00
func (muh *modUpgradeHandle) Upgrades(ctx context.Context, snapshot *snapshot) (map[string]string, error) {
v, err := muh.handle.Get(ctx, snapshot.generation, snapshot)
if v == nil {
return nil, err
}
data := v.(*modUpgradeData)
return data.upgrades, data.err
}
func (s *snapshot) ModUpgrade(ctx context.Context) (map[string]string, error) {
if err := s.awaitLoaded(ctx); err != nil {
return nil, err
}
fh, err := s.GetFile(ctx, s.view.modURI)
if err != nil {
return nil, err
}
cfg := s.config(ctx)
key := modKey{
sessionID: s.view.session.id,
cfg: hashConfig(cfg),
mod: fh.FileIdentity(),
view: s.view.root.Filename(),
verb: upgrade,
}
internal/memoize: switch from GC-driven to explicit deletion The GC-based cache has given us a number of problems. First, memory leaks driven by reference cycles: the Go runtime cannot collect cycles involving finalizers, which prevents us from writing natural code in Bind callbacks. If we screw it up, we get a mysterious leak that takes a long time to track down. Second, the behavior is generally mysterious; it's hard to predict how long a value lasts, and harder to tell if a value being live is a bug. Third, we think that it may be interacting poorly with the GC, resulting in unnecessary memory usage. The structure of the values we put in the cache is not actually that complicated -- there are only 5 significant types: parse, typecheck, analyze, parse mod, and analyze mod. Managing them manually should not be conceptually difficult, and in fact we already do most of the work in (*snapshot).clone. In this CL the cache adds the concept of "generations", which function as reference counts on cache entries. Entries are still global and shared across generations, but will be explicitly deleted once no generations refer to them. The idea is that each snapshot is a new generation, and can inherit entries from the previous snapshot or leave them behind to be deleted. One obvious risk of this scheme is that we'll leave dangling references to values without actually inheriting them across generations. To prevent that, getting a value requires passing in the generation at which it's being read, and an error will be returned if that generation is dead. Change-Id: I4b30891efd7be4e10f2b84f4c067b0dee43dcf9c Reviewed-on: https://go-review.googlesource.com/c/tools/+/242838 Run-TryBot: Heschi Kreinick <heschi@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Rebecca Stambler <rstambler@golang.org> Reviewed-by: Robert Findley <rfindley@google.com>
2020-07-24 15:17:13 -06:00
h := s.generation.Bind(key, func(ctx context.Context, arg memoize.Arg) interface{} {
ctx, done := event.Start(ctx, "cache.ModUpgradeHandle", tag.URI.Of(fh.URI()))
defer done()
snapshot := arg.(*snapshot)
pm, err := s.ParseMod(ctx, fh)
if err != nil {
return &modUpgradeData{err: err}
}
// No requires to upgrade.
if len(pm.File.Require) == 0 {
return &modUpgradeData{}
}
// Run "go list -mod readonly -u -m all" to be able to see which deps can be
// upgraded without modifying mod file.
args := []string{"-u", "-m", "all"}
if !snapshot.view.tmpMod || containsVendor(fh.URI()) {
// Use -mod=readonly if the module contains a vendor directory
// (see golang/go#38711).
args = append([]string{"-mod", "readonly"}, args...)
}
stdout, err := snapshot.RunGoCommand(ctx, "list", args)
if err != nil {
return &modUpgradeData{err: err}
}
upgradesList := strings.Split(stdout.String(), "\n")
if len(upgradesList) <= 1 {
return nil
}
upgrades := make(map[string]string)
for _, upgrade := range upgradesList[1:] {
// Example: "github.com/x/tools v1.1.0 [v1.2.0]"
info := strings.Split(upgrade, " ")
if len(info) != 3 {
continue
}
dep, version := info[0], info[2]
// Make sure that the format matches our expectation.
if len(version) < 2 {
continue
}
if version[0] != '[' || version[len(version)-1] != ']' {
continue
}
latest := version[1 : len(version)-1] // remove the "[" and "]"
upgrades[dep] = latest
}
return &modUpgradeData{
upgrades: upgrades,
}
})
muh := &modUpgradeHandle{handle: h}
s.mu.Lock()
s.modUpgradeHandle = muh
s.mu.Unlock()
return s.modUpgradeHandle.Upgrades(ctx, s)
}
// containsVendor reports whether the module has a vendor folder.
func containsVendor(modURI span.URI) bool {
dir := filepath.Dir(modURI.Filename())
f, err := os.Stat(filepath.Join(dir, "vendor"))
if err != nil {
return false
}
return f.IsDir()
}