f16d61cdfa
The server skip fallback for stages with a skipped edge now applies only when every earlier visit recorded a complete file-change capture with no changed files and no image artifact. Any change (no repo-specific UI path or extension list), any partial/unavailable/missing capture, a human decision or a checkpoint-retry prefix runs the model. auto_skip=false opts a stage out. The fallback verdict names the server fallback. Capture compares over-budget text by hash and, in Git, lists paths it could not compare when their git status changes, marked no_patch, so a large workspace no longer records files: [] for a real edit. Run detail gains a warnings list for partial/unavailable captures, also printed by inspect on stderr.
507 lines
16 KiB
Go
507 lines
16 KiB
Go
// Package filediff captures bounded, read-only stage-local workspace changes.
|
|
package filediff
|
|
|
|
import (
|
|
"bytes"
|
|
"context"
|
|
"crypto/sha256"
|
|
"encoding/json"
|
|
"errors"
|
|
"fmt"
|
|
"io"
|
|
"io/fs"
|
|
"os"
|
|
"os/exec"
|
|
"sort"
|
|
"strings"
|
|
"sync/atomic"
|
|
"syscall"
|
|
"unicode/utf8"
|
|
|
|
"git.a-wan.net/alan/pipelines/internal/protocol"
|
|
"github.com/pmezard/go-difflib/difflib"
|
|
)
|
|
|
|
const (
|
|
maxFiles = 10000
|
|
maxFileBytes = 1 << 20
|
|
maxSnapshotBytes = 32 << 20
|
|
maxPatchBytes = 128 << 10
|
|
maxFilePatchBytes = 64 << 10
|
|
maxLines = 5000
|
|
binarySniffBytes = 8 << 10
|
|
)
|
|
|
|
// maxHashBytes bounds one hash-only file; larger files are skipped (and, in a
|
|
// Git working tree, listed only when their Git status changes). It is a
|
|
// variable so package tests can exercise that path without huge files.
|
|
var maxHashBytes int64 = 256 << 20
|
|
|
|
// snapshotBudget is the retained text bytes per snapshot. Files beyond it (or
|
|
// beyond the per-file limit) are still compared by content hash and listed
|
|
// without a patch, so an over-budget workspace never hides which files changed.
|
|
var snapshotBudget atomic.Int64
|
|
|
|
func init() { snapshotBudget.Store(maxSnapshotBytes) }
|
|
|
|
// SetSnapshotBudgetForTesting lowers the retained-text budget so tests can
|
|
// exercise an over-budget workspace without writing tens of megabytes. It
|
|
// returns a function restoring the previous budget. Production never calls it.
|
|
func SetSnapshotBudgetForTesting(n int64) (restore func()) {
|
|
previous := snapshotBudget.Swap(n)
|
|
return func() { snapshotBudget.Store(previous) }
|
|
}
|
|
|
|
type file struct {
|
|
data []byte
|
|
hash [32]byte
|
|
mode fs.FileMode
|
|
binary bool
|
|
// hashOnly marks a text file compared by content hash only: it exceeded the
|
|
// per-file or per-snapshot text budget, so no patch can be produced.
|
|
hashOnly bool
|
|
}
|
|
|
|
type Snapshot struct {
|
|
files map[string]file
|
|
skipped map[string]bool
|
|
partial bool
|
|
available bool
|
|
diagnostic string
|
|
scope string
|
|
// gitStatus maps run-relative paths to their porcelain XY status code when
|
|
// the run directory is a Git working tree and the status read succeeded
|
|
// (gitOK). Compare uses it to list changed paths the content snapshot could
|
|
// not compare (time/file limit, unreadable); it never reads HEAD/index content.
|
|
gitStatus map[string]string
|
|
gitOK bool
|
|
}
|
|
|
|
// readGitStatus reads `git status --porcelain=v1 -z` for the run directory and
|
|
// returns its paths relative to that directory. It only reads: optional locks
|
|
// are disabled so the index is never refreshed or rewritten.
|
|
func readGitStatus(ctx context.Context, directory string) (map[string]string, bool) {
|
|
prefixOut, err := exec.CommandContext(ctx, "git", "-c", "core.fsmonitor=false", "-C", directory, "rev-parse", "--show-prefix").Output()
|
|
if err != nil {
|
|
return nil, false
|
|
}
|
|
prefix := strings.TrimSpace(string(prefixOut))
|
|
cmd := exec.CommandContext(ctx, "git", "--no-optional-locks", "-c", "core.fsmonitor=false", "-C", directory, "status", "--porcelain=v1", "-z", "--no-renames", "--untracked-files=all", "--", ".")
|
|
out := &boundedOutput{limit: 4 << 20}
|
|
cmd.Stdout = out
|
|
if err := cmd.Run(); err != nil {
|
|
return nil, false
|
|
}
|
|
status := map[string]string{}
|
|
for _, entry := range strings.Split(out.String(), "\x00") {
|
|
if len(entry) < 4 || entry[2] != ' ' {
|
|
continue
|
|
}
|
|
name, ok := strings.CutPrefix(entry[3:], prefix)
|
|
if !ok || name == "" {
|
|
continue // outside the run directory
|
|
}
|
|
status[name] = entry[:2]
|
|
}
|
|
return status, true
|
|
}
|
|
|
|
// hashFile streams f from its start into a content hash without retaining it.
|
|
// It returns ok=false when the file exceeds maxHashBytes or cannot be read.
|
|
func hashFile(f *os.File) (hash [32]byte, binary bool, ok bool) {
|
|
if _, err := f.Seek(0, io.SeekStart); err != nil {
|
|
return hash, false, false
|
|
}
|
|
h := sha256.New()
|
|
head := make([]byte, binarySniffBytes)
|
|
n, err := io.ReadFull(f, head)
|
|
if err != nil && !errors.Is(err, io.ErrUnexpectedEOF) && !errors.Is(err, io.EOF) {
|
|
return hash, false, false
|
|
}
|
|
head = head[:n]
|
|
h.Write(head)
|
|
copied, err := io.Copy(h, io.LimitReader(f, maxHashBytes+1-int64(n)))
|
|
if err != nil || int64(n)+copied > maxHashBytes {
|
|
return hash, false, false
|
|
}
|
|
copy(hash[:], h.Sum(nil))
|
|
return hash, bytes.IndexByte(head, 0) >= 0, true
|
|
}
|
|
|
|
func protected(name string) bool {
|
|
for _, part := range strings.Split(strings.ToLower(name), "/") {
|
|
if part == ".git" || part == "secrets" || part == ".ssh" || part == ".aws" || part == ".gnupg" || part == ".kube" || part == "node_modules" {
|
|
return true
|
|
}
|
|
if part == ".env" || strings.HasPrefix(part, ".env.") || strings.HasSuffix(part, ".env") || strings.HasSuffix(part, ".pem") || strings.HasSuffix(part, ".key") || strings.HasSuffix(part, ".db") || strings.HasSuffix(part, ".db-wal") || strings.HasSuffix(part, ".db-shm") || strings.HasSuffix(part, ".sqlite") || strings.HasSuffix(part, ".sqlite3") || part == "auth.json" || part == "credentials.json" {
|
|
return true
|
|
}
|
|
}
|
|
return false
|
|
}
|
|
|
|
type boundedOutput struct {
|
|
bytes.Buffer
|
|
limit int
|
|
}
|
|
|
|
func (b *boundedOutput) Write(p []byte) (int, error) {
|
|
if b.Len()+len(p) > b.limit {
|
|
return 0, fmt.Errorf("file inventory limit exceeded")
|
|
}
|
|
return b.Buffer.Write(p)
|
|
}
|
|
|
|
// Git only supplies an inventory; never compare against HEAD/index or alter them.
|
|
// Tracked dirty files and untracked, nonignored files are compared as they exist.
|
|
func inventory(ctx context.Context, directory string, root *os.Root) ([]string, bool, error) {
|
|
probe := exec.CommandContext(ctx, "git", "-c", "core.fsmonitor=false", "-C", directory, "rev-parse", "--is-inside-work-tree")
|
|
if data, err := probe.Output(); err == nil && strings.TrimSpace(string(data)) == "true" {
|
|
cmd := exec.CommandContext(ctx, "git", "-c", "core.fsmonitor=false", "-C", directory, "ls-files", "--cached", "--others", "--exclude-standard", "--deduplicate", "-z", "--", ".")
|
|
out := &boundedOutput{limit: 2 << 20}
|
|
cmd.Stdout = out
|
|
if err := cmd.Run(); err != nil {
|
|
return nil, true, fmt.Errorf("Git file inventory unavailable")
|
|
}
|
|
return strings.Split(strings.TrimSuffix(out.String(), "\x00"), "\x00"), true, nil
|
|
}
|
|
if ctx.Err() != nil {
|
|
return nil, false, ctx.Err()
|
|
}
|
|
var names []string
|
|
err := fs.WalkDir(root.FS(), ".", func(name string, entry fs.DirEntry, err error) error {
|
|
if ctx.Err() != nil {
|
|
return ctx.Err()
|
|
}
|
|
if err != nil {
|
|
return err
|
|
}
|
|
if name != "." && protected(name) {
|
|
names = append(names, name)
|
|
if entry.IsDir() {
|
|
return fs.SkipDir
|
|
}
|
|
return nil
|
|
}
|
|
if !entry.IsDir() {
|
|
names = append(names, name)
|
|
if len(names) > maxFiles {
|
|
return fmt.Errorf("file inventory limit exceeded")
|
|
}
|
|
}
|
|
return nil
|
|
})
|
|
return names, false, err
|
|
}
|
|
|
|
func Capture(ctx context.Context, directory string, previous ...*Snapshot) *Snapshot {
|
|
s := &Snapshot{files: map[string]file{}, skipped: map[string]bool{}}
|
|
root, err := os.OpenRoot(directory)
|
|
if err != nil {
|
|
s.diagnostic = "Workspace snapshot unavailable."
|
|
return s
|
|
}
|
|
defer root.Close()
|
|
s.available = true
|
|
names, git, err := inventory(ctx, directory, root)
|
|
s.scope = "Workspace regular files"
|
|
if git {
|
|
s.scope = "Git tracked and nonignored files in the run directory"
|
|
}
|
|
if err != nil {
|
|
s.partial = true
|
|
s.diagnostic = "Workspace inventory incomplete."
|
|
}
|
|
if git {
|
|
s.gitStatus, s.gitOK = readGitStatus(ctx, directory)
|
|
}
|
|
// Keep pre-stage paths even if a stage changes .gitignore/the index.
|
|
known := map[string]bool{}
|
|
for _, name := range names {
|
|
known[name] = true
|
|
}
|
|
for _, prior := range previous {
|
|
if prior != nil {
|
|
for name := range prior.files {
|
|
if !known[name] {
|
|
names = append(names, name)
|
|
known[name] = true
|
|
}
|
|
}
|
|
}
|
|
}
|
|
sort.Strings(names)
|
|
used := 0
|
|
for i, name := range names {
|
|
if name == "" {
|
|
continue
|
|
}
|
|
if i >= maxFiles || ctx.Err() != nil {
|
|
s.partial = true
|
|
s.diagnostic = "Workspace capture reached its time/file limit."
|
|
break
|
|
}
|
|
if !fs.ValidPath(name) || protected(name) {
|
|
s.skipped[name] = true
|
|
continue
|
|
}
|
|
// Reject symlinks in every component, including aliases to sensitive paths.
|
|
valid := true
|
|
parts := strings.Split(name, "/")
|
|
for j := range parts {
|
|
info, err := root.Lstat(strings.Join(parts[:j+1], "/"))
|
|
if errors.Is(err, fs.ErrNotExist) {
|
|
valid = false
|
|
break
|
|
} // tracked deletion
|
|
if err != nil || info.Mode()&os.ModeSymlink != 0 {
|
|
s.skipped[name] = true
|
|
valid = false
|
|
break
|
|
}
|
|
}
|
|
if !valid {
|
|
continue
|
|
}
|
|
info, err := root.Lstat(name)
|
|
if err != nil || !info.Mode().IsRegular() || info.Size() > maxHashBytes {
|
|
s.skipped[name] = true
|
|
continue
|
|
}
|
|
f, err := root.OpenFile(name, os.O_RDONLY|syscall.O_NONBLOCK|syscall.O_NOFOLLOW, 0)
|
|
if err != nil {
|
|
s.skipped[name] = true
|
|
continue
|
|
}
|
|
opened, statErr := f.Stat()
|
|
if statErr != nil || !opened.Mode().IsRegular() {
|
|
f.Close()
|
|
s.skipped[name] = true
|
|
continue
|
|
}
|
|
if opened.Size() <= maxFileBytes {
|
|
data, readErr := io.ReadAll(io.LimitReader(f, maxFileBytes+1))
|
|
if readErr != nil {
|
|
f.Close()
|
|
s.skipped[name] = true
|
|
continue
|
|
}
|
|
if len(data) <= maxFileBytes {
|
|
f.Close()
|
|
binary := bytes.IndexByte(data, 0) >= 0 || !utf8.Valid(data)
|
|
hash := sha256.Sum256(data)
|
|
hashOnly := false
|
|
// Binary changes need only metadata and a hash, never retained patch bytes.
|
|
// Counting screenshots against the retained-text budget can otherwise
|
|
// exhaust it before any source files are reached in sorted inventory order.
|
|
// Text beyond the budget keeps only its hash, so a change to it is still
|
|
// listed (without a patch) instead of being skipped unseen.
|
|
if binary {
|
|
data = nil
|
|
} else if int64(used+len(data)) > snapshotBudget.Load() {
|
|
data = nil
|
|
hashOnly = true
|
|
} else {
|
|
used += len(data)
|
|
}
|
|
s.files[name] = file{data: data, hash: hash, mode: opened.Mode().Perm(), binary: binary, hashOnly: hashOnly}
|
|
continue
|
|
}
|
|
// The file grew past the per-file limit while being read: hash it.
|
|
}
|
|
hash, binary, ok := hashFile(f)
|
|
f.Close()
|
|
if !ok {
|
|
s.skipped[name] = true
|
|
continue
|
|
}
|
|
s.files[name] = file{hash: hash, mode: opened.Mode().Perm(), binary: binary, hashOnly: !binary}
|
|
}
|
|
return s
|
|
}
|
|
|
|
func Compare(before, after *Snapshot) *protocol.StageFileChanges {
|
|
result := &protocol.StageFileChanges{Status: "complete", Files: []protocol.FileChange{}}
|
|
if before == nil || after == nil || !before.available || !after.available {
|
|
result.Status = "unavailable"
|
|
result.Diagnostic = "Workspace snapshot unavailable; changes could not be compared."
|
|
return result
|
|
}
|
|
result.Scope = before.scope
|
|
if before.partial || after.partial {
|
|
result.Status = "partial"
|
|
result.Diagnostic = "Workspace capture incomplete; some changes may be missing."
|
|
}
|
|
skipped := map[string]bool{}
|
|
for name := range before.skipped {
|
|
skipped[name] = true
|
|
}
|
|
for name := range after.skipped {
|
|
skipped[name] = true
|
|
}
|
|
result.Skipped = len(skipped)
|
|
if result.Skipped > 0 {
|
|
result.Status = "partial"
|
|
result.Diagnostic = "Excluded protected, symlinked, oversized or unreadable files; this is not a complete workspace diff."
|
|
}
|
|
keys := map[string]bool{}
|
|
for name := range before.files {
|
|
keys[name] = true
|
|
}
|
|
for name := range after.files {
|
|
keys[name] = true
|
|
}
|
|
names := make([]string, 0, len(keys))
|
|
for name := range keys {
|
|
names = append(names, name)
|
|
}
|
|
sort.Strings(names)
|
|
remaining := maxPatchBytes
|
|
retained := 0
|
|
add := func(change protocol.FileChange) {
|
|
if change.Truncated {
|
|
result.Status = "partial"
|
|
if !change.NoPatch {
|
|
result.Diagnostic = "Some patches were omitted/truncated by capture limits."
|
|
}
|
|
}
|
|
encoded, _ := json.Marshal(change)
|
|
if retained+len(encoded)+1 > maxPatchBytes {
|
|
change.Patch = ""
|
|
change.Truncated = true
|
|
encoded, _ = json.Marshal(change)
|
|
}
|
|
if retained+len(encoded)+1 > maxPatchBytes {
|
|
result.Omitted++
|
|
result.Status = "partial"
|
|
result.Diagnostic = "Additional file changes omitted by the retained evidence size limit."
|
|
return
|
|
}
|
|
if change.Truncated {
|
|
result.Status = "partial"
|
|
}
|
|
retained += len(encoded) + 1
|
|
result.Files = append(result.Files, change)
|
|
}
|
|
listed := 0
|
|
// decided records every path the content comparison settled (changed or
|
|
// not), so the Git-status pass below only speaks for the rest.
|
|
decided := map[string]bool{}
|
|
for _, name := range names {
|
|
if skipped[name] {
|
|
continue
|
|
}
|
|
a, aOK := before.files[name]
|
|
b, bOK := after.files[name]
|
|
// Missing files in an incomplete inventory cannot prove deletion/addition.
|
|
if !aOK && before.partial || !bOK && after.partial {
|
|
continue
|
|
}
|
|
decided[name] = true
|
|
if aOK && bOK && a.hash == b.hash && a.mode == b.mode {
|
|
continue
|
|
}
|
|
kind := "modified"
|
|
if !aOK {
|
|
kind = "added"
|
|
}
|
|
if !bOK {
|
|
kind = "deleted"
|
|
}
|
|
change := protocol.FileChange{Path: name, Kind: kind, Binary: a.binary || b.binary}
|
|
if aOK {
|
|
change.BeforeMode = fmt.Sprintf("%04o", a.mode)
|
|
}
|
|
if bOK {
|
|
change.AfterMode = fmt.Sprintf("%04o", b.mode)
|
|
}
|
|
if !change.Binary && a.hash != b.hash && (a.hashOnly || b.hashOnly) {
|
|
// Over-budget content was compared by hash: list it, without a patch.
|
|
change.Truncated = true
|
|
change.NoPatch = true
|
|
listed++
|
|
} else if !change.Binary && a.hash != b.hash {
|
|
if remaining == 0 || bytes.Count(a.data, []byte("\n"))+bytes.Count(b.data, []byte("\n")) > maxLines {
|
|
change.Truncated = true
|
|
} else {
|
|
from, to := "a/"+name, "b/"+name
|
|
if !aOK {
|
|
from = "/dev/null"
|
|
}
|
|
if !bOK {
|
|
to = "/dev/null"
|
|
}
|
|
// Quote paths so names cannot forge headers or terminal control sequences.
|
|
patch, err := difflib.GetUnifiedDiffString(difflib.UnifiedDiff{A: diffLines(a.data), B: diffLines(b.data), FromFile: fmt.Sprintf("%q", from), ToFile: fmt.Sprintf("%q", to), Context: 3})
|
|
if err != nil {
|
|
change.Truncated = true
|
|
} else {
|
|
limit := min(remaining, maxFilePatchBytes)
|
|
if len(patch) > limit {
|
|
patch = patch[:limit]
|
|
for !utf8.ValidString(patch) && len(patch) > 0 {
|
|
patch = patch[:len(patch)-1]
|
|
}
|
|
change.Truncated = true
|
|
}
|
|
change.Patch = patch
|
|
remaining -= len(patch)
|
|
}
|
|
}
|
|
}
|
|
add(change)
|
|
}
|
|
// In a Git working tree, paths the content snapshot could not settle
|
|
// (skipped as unreadable/oversized, or beyond the time/file limit) are still
|
|
// listed when their Git status changed during the visit. A path dirty with
|
|
// the same status before and after cannot be proven changed and stays
|
|
// counted as skipped. Protected paths are never listed.
|
|
if before.gitOK && after.gitOK {
|
|
gitNames := map[string]bool{}
|
|
for name := range before.gitStatus {
|
|
gitNames[name] = true
|
|
}
|
|
for name := range after.gitStatus {
|
|
gitNames[name] = true
|
|
}
|
|
ordered := make([]string, 0, len(gitNames))
|
|
for name := range gitNames {
|
|
if !decided[name] && fs.ValidPath(name) && !protected(name) {
|
|
ordered = append(ordered, name)
|
|
}
|
|
}
|
|
sort.Strings(ordered)
|
|
for _, name := range ordered {
|
|
was, now := before.gitStatus[name], after.gitStatus[name]
|
|
if was == now {
|
|
continue
|
|
}
|
|
kind := "modified"
|
|
switch {
|
|
case now == "" && was == "??", strings.Contains(now, "D"):
|
|
kind = "deleted"
|
|
case was == "" && (now == "??" || strings.HasPrefix(now, "A")):
|
|
kind = "added"
|
|
}
|
|
add(protocol.FileChange{Path: name, Kind: kind, Truncated: true, NoPatch: true})
|
|
listed++
|
|
}
|
|
}
|
|
if listed > 0 && result.Status == "partial" && result.Omitted == 0 {
|
|
result.Diagnostic = "Some changed files exceeded capture limits and are listed without patches; this is not a complete workspace diff."
|
|
}
|
|
return result
|
|
}
|
|
|
|
func diffLines(data []byte) []string {
|
|
if len(data) == 0 {
|
|
return nil
|
|
}
|
|
lines := strings.SplitAfter(string(data), "\n")
|
|
if lines[len(lines)-1] == "" {
|
|
return lines[:len(lines)-1]
|
|
}
|
|
lines[len(lines)-1] += "\n\\ No newline at end of file\n"
|
|
return lines
|
|
}
|