githaven/modules/git/blame.go
zeripath 858c35b731
Ensure BlameReaders close at end of request (#12102)
#11716 reports multiple git blame processes hanging around
this was thought to be due to timeouts, however on closer look this
appears to be due to the Close() function of the BlameReader hanging
with a blocked stdout pipe.

This PR fixes this Close function to:

* Cancel the context of the cmd
* Close the StdoutReader - ensuring that the output pipe is closed

Further it makes the context of the `git blame` command a child of the
request context - ensuring that even if Close() is not called, on
cancellation of the Request the blame is command will also be cancelled.

Fixes #11716
Closes #11727

Signed-off-by: Andrew Thornton <art27@cantab.net>
2020-07-01 14:01:17 +01:00

135 lines
2.9 KiB
Go

// Copyright 2019 The Gitea Authors. All rights reserved.
// Use of this source code is governed by a MIT-style
// license that can be found in the LICENSE file.
package git
import (
"bufio"
"context"
"fmt"
"io"
"os"
"os/exec"
"regexp"
"code.gitea.io/gitea/modules/process"
)
// BlamePart represents block of blame - continuous lines with one sha
type BlamePart struct {
Sha string
Lines []string
}
// BlameReader returns part of file blame one by one
type BlameReader struct {
cmd *exec.Cmd
pid int64
output io.ReadCloser
scanner *bufio.Scanner
lastSha *string
cancel context.CancelFunc
}
var shaLineRegex = regexp.MustCompile("^([a-z0-9]{40})")
// NextPart returns next part of blame (sequencial code lines with the same commit)
func (r *BlameReader) NextPart() (*BlamePart, error) {
var blamePart *BlamePart
scanner := r.scanner
if r.lastSha != nil {
blamePart = &BlamePart{*r.lastSha, make([]string, 0)}
}
for scanner.Scan() {
line := scanner.Text()
// Skip empty lines
if len(line) == 0 {
continue
}
lines := shaLineRegex.FindStringSubmatch(line)
if lines != nil {
sha1 := lines[1]
if blamePart == nil {
blamePart = &BlamePart{sha1, make([]string, 0)}
}
if blamePart.Sha != sha1 {
r.lastSha = &sha1
return blamePart, nil
}
} else if line[0] == '\t' {
code := line[1:]
blamePart.Lines = append(blamePart.Lines, code)
}
}
r.lastSha = nil
return blamePart, nil
}
// Close BlameReader - don't run NextPart after invoking that
func (r *BlameReader) Close() error {
defer process.GetManager().Remove(r.pid)
r.cancel()
_ = r.output.Close()
if err := r.cmd.Wait(); err != nil {
return fmt.Errorf("Wait: %v", err)
}
return nil
}
// CreateBlameReader creates reader for given repository, commit and file
func CreateBlameReader(ctx context.Context, repoPath, commitID, file string) (*BlameReader, error) {
gitRepo, err := OpenRepository(repoPath)
if err != nil {
return nil, err
}
gitRepo.Close()
return createBlameReader(ctx, repoPath, GitExecutable, "blame", commitID, "--porcelain", "--", file)
}
func createBlameReader(ctx context.Context, dir string, command ...string) (*BlameReader, error) {
// Here we use the provided context - this should be tied to the request performing the blame so that it does not hang around.
ctx, cancel := context.WithCancel(ctx)
cmd := exec.CommandContext(ctx, command[0], command[1:]...)
cmd.Dir = dir
cmd.Stderr = os.Stderr
stdout, err := cmd.StdoutPipe()
if err != nil {
defer cancel()
return nil, fmt.Errorf("StdoutPipe: %v", err)
}
if err = cmd.Start(); err != nil {
defer cancel()
return nil, fmt.Errorf("Start: %v", err)
}
pid := process.GetManager().Add(fmt.Sprintf("GetBlame [repo_path: %s]", dir), cancel)
scanner := bufio.NewScanner(stdout)
return &BlameReader{
cmd,
pid,
stdout,
scanner,
nil,
cancel,
}, nil
}