vole
Clone
Read only, no credentials:
https://rodent.software/vole.git
Write over SSH with a registered key:
git clone ssh://git@rodent.software:2222/vole.git
A free and open-source forge that lives in your Git repo.
| 1 | // Package git wraps the git binary. Vole shells out rather than reimplementing |
| 2 | // git: hooks, pack handling, and compare-and-swap ref updates all come from the |
| 3 | // real thing. |
| 4 | package git |
| 5 | |
| 6 | import ( |
| 7 | "bufio" |
| 8 | "bytes" |
| 9 | "context" |
| 10 | "errors" |
| 11 | "fmt" |
| 12 | "io" |
| 13 | "os" |
| 14 | "os/exec" |
| 15 | "path/filepath" |
| 16 | "sort" |
| 17 | "strconv" |
| 18 | "strings" |
| 19 | ) |
| 20 | |
| 21 | const ZeroSHA = "0000000000000000000000000000000000000000" |
| 22 | |
| 23 | // Entry is one record from `git ls-tree` or one line of `git mktree` input. |
| 24 | type Entry struct { |
| 25 | Mode string |
| 26 | Type string |
| 27 | SHA string |
| 28 | Name string |
| 29 | } |
| 30 | |
| 31 | type Repo struct { |
| 32 | Dir string |
| 33 | // ExtraEnv is appended to every git invocation. It exists for the one case |
| 34 | // where a repository-scoped variable must survive the scrub below: a hook |
| 35 | // inspecting the very repository being pushed, whose new objects are only |
| 36 | // visible through GIT_QUARANTINE_PATH. |
| 37 | ExtraEnv []string |
| 38 | // ctx, when set, bounds every git invocation this handle starts. A web page |
| 39 | // that shells out to git needs one: a repository can be slow, and the process |
| 40 | // is the thing that has to stop. |
| 41 | ctx context.Context |
| 42 | } |
| 43 | |
| 44 | func Open(dir string) *Repo { return &Repo{Dir: dir} } |
| 45 | |
| 46 | // WithContext returns a handle whose git invocations are killed when ctx ends. It |
| 47 | // is a copy, so a caller cannot change the deadline of a handle somebody else |
| 48 | // holds. |
| 49 | func (r *Repo) WithContext(ctx context.Context) *Repo { |
| 50 | cp := *r |
| 51 | cp.ctx = ctx |
| 52 | return &cp |
| 53 | } |
| 54 | |
| 55 | func (r *Repo) run(stdin []byte, args ...string) ([]byte, string, error) { |
| 56 | return r.runEnv(nil, stdin, args...) |
| 57 | } |
| 58 | |
| 59 | // runEnv is run with extra environment entries appended, so a single call can |
| 60 | // carry an author without mutating process state. os/exec keeps the last value |
| 61 | // for a duplicated key, which is what makes the override reliable. |
| 62 | func (r *Repo) runEnv(extra []string, stdin []byte, args ...string) ([]byte, string, error) { |
| 63 | cmd := r.command(args...) |
| 64 | cmd.Env = append(cmd.Env, extra...) |
| 65 | if stdin != nil { |
| 66 | cmd.Stdin = bytes.NewReader(stdin) |
| 67 | } |
| 68 | var out, errb bytes.Buffer |
| 69 | cmd.Stdout = &out |
| 70 | cmd.Stderr = &errb |
| 71 | err := cmd.Run() |
| 72 | return out.Bytes(), strings.TrimSpace(errb.String()), err |
| 73 | } |
| 74 | |
| 75 | // Run exposes raw git invocation for callers that need it. |
| 76 | func (r *Repo) Run(stdin []byte, args ...string) ([]byte, error) { |
| 77 | out, errs, err := r.run(stdin, args...) |
| 78 | if err != nil { |
| 79 | return out, fmt.Errorf("git %s: %w: %s", strings.Join(args, " "), err, errs) |
| 80 | } |
| 81 | return out, nil |
| 82 | } |
| 83 | |
| 84 | // command builds a git invocation with the environment every call shares. |
| 85 | func (r *Repo) command(args ...string) *exec.Cmd { |
| 86 | var cmd *exec.Cmd |
| 87 | if r.ctx != nil { |
| 88 | cmd = exec.CommandContext(r.ctx, "git", args...) |
| 89 | } else { |
| 90 | cmd = exec.Command("git", args...) |
| 91 | } |
| 92 | cmd.Env = append(scrubbedEnv(), |
| 93 | "GIT_DIR="+r.Dir, |
| 94 | "GIT_TERMINAL_PROMPT=0", |
| 95 | "GIT_AUTHOR_NAME=vole", "GIT_AUTHOR_EMAIL=vole@localhost", |
| 96 | "GIT_COMMITTER_NAME=vole", "GIT_COMMITTER_EMAIL=vole@localhost", |
| 97 | ) |
| 98 | cmd.Env = append(cmd.Env, r.ExtraEnv...) |
| 99 | return cmd |
| 100 | } |
| 101 | |
| 102 | // runHead returns at most n bytes of a git command's stdout. `cat-file blob` |
| 103 | // writes the whole object, and the caller only wants to sniff the first bytes of |
| 104 | // a file that is about to be accepted, so the process is stopped once the |
| 105 | // sample is read rather than after a needless 100 MiB copy. |
| 106 | func (r *Repo) runHead(n int, args ...string) ([]byte, string, error) { |
| 107 | cmd := r.command(args...) |
| 108 | var errb bytes.Buffer |
| 109 | cmd.Stderr = &errb |
| 110 | stdout, err := cmd.StdoutPipe() |
| 111 | if err != nil { |
| 112 | return nil, "", err |
| 113 | } |
| 114 | if err := cmd.Start(); err != nil { |
| 115 | return nil, "", err |
| 116 | } |
| 117 | buf := make([]byte, n) |
| 118 | got, rerr := io.ReadFull(stdout, buf) |
| 119 | if rerr != nil && !errors.Is(rerr, io.ErrUnexpectedEOF) && !errors.Is(rerr, io.EOF) { |
| 120 | cmd.Process.Kill() |
| 121 | cmd.Wait() |
| 122 | return nil, strings.TrimSpace(errb.String()), rerr |
| 123 | } |
| 124 | // Short blob: the process has already finished. Otherwise stop it; git's |
| 125 | // exit status for a closed pipe is not a failure of this call. |
| 126 | if got == n { |
| 127 | cmd.Process.Kill() |
| 128 | } |
| 129 | cmd.Wait() |
| 130 | return buf[:got], strings.TrimSpace(errb.String()), nil |
| 131 | } |
| 132 | |
| 133 | // InitBare creates a bare repo at dir with the given default branch. |
| 134 | func InitBare(dir, branch string) error { |
| 135 | if err := os.MkdirAll(dir, 0o755); err != nil { |
| 136 | return err |
| 137 | } |
| 138 | cmd := exec.Command("git", "init", "--bare", "-q", "--initial-branch="+branch, dir) |
| 139 | var errb bytes.Buffer |
| 140 | cmd.Stderr = &errb |
| 141 | if err := cmd.Run(); err != nil { |
| 142 | return fmt.Errorf("git init: %w: %s", err, strings.TrimSpace(errb.String())) |
| 143 | } |
| 144 | return nil |
| 145 | } |
| 146 | |
| 147 | // RevParse resolves a revision to a commit SHA, returning "" when it does not |
| 148 | // exist (an unborn branch is not an error). |
| 149 | func (r *Repo) RevParse(rev string) (string, error) { |
| 150 | out, _, err := r.run(nil, "rev-parse", "--verify", "--quiet", rev+"^{commit}") |
| 151 | if err != nil { |
| 152 | return "", nil |
| 153 | } |
| 154 | return strings.TrimSpace(string(out)), nil |
| 155 | } |
| 156 | |
| 157 | // RefCommit resolves a branch ref to its commit. A loose ref is read directly, |
| 158 | // which is how git stores every branch it updates, and anything else — a packed |
| 159 | // ref, a symbolic ref, a ref that is not there — falls back to rev-parse. The |
| 160 | // fast path exists because the index checks the metadata ref on every page and |
| 161 | // every search refresh; a file read is far cheaper than a process. |
| 162 | // |
| 163 | // It is for branch refs. A tag ref can point at a tag object, which RevParse |
| 164 | // peels and this does not. |
| 165 | func (r *Repo) RefCommit(ref string) (string, error) { |
| 166 | name := strings.TrimPrefix(ref, "refs/heads/") |
| 167 | if name != ref && looseRefName(name) { |
| 168 | if data, err := os.ReadFile(filepath.Join(r.Dir, "refs", "heads", name)); err == nil { |
| 169 | if sha := strings.TrimSpace(string(data)); isObjectID(sha) { |
| 170 | return sha, nil |
| 171 | } |
| 172 | } |
| 173 | } |
| 174 | return r.RevParse(ref) |
| 175 | } |
| 176 | |
| 177 | // looseRefName rejects a name that cannot be a branch path segment, so the fast |
| 178 | // path never reads outside the ref store. |
| 179 | func looseRefName(name string) bool { |
| 180 | if name == "" || strings.HasPrefix(name, "/") || strings.HasSuffix(name, "/") || |
| 181 | strings.Contains(name, "..") || strings.ContainsAny(name, "\\\x00\n") { |
| 182 | return false |
| 183 | } |
| 184 | return true |
| 185 | } |
| 186 | |
| 187 | // isObjectID reports the form of a git object id: 40 hex digits for SHA-1, 64 |
| 188 | // for SHA-256. A ref file holding anything else (a symbolic ref, a comment) is |
| 189 | // not answered by the fast path. |
| 190 | func isObjectID(s string) bool { |
| 191 | if len(s) != 40 && len(s) != 64 { |
| 192 | return false |
| 193 | } |
| 194 | for _, r := range s { |
| 195 | if (r < '0' || r > '9') && (r < 'a' || r > 'f') { |
| 196 | return false |
| 197 | } |
| 198 | } |
| 199 | return true |
| 200 | } |
| 201 | |
| 202 | // RevParseRaw resolves a rev without peeling to a commit, returning "" when it |
| 203 | // does not exist. Use this to detect a ref that points at the wrong object type. |
| 204 | func (r *Repo) RevParseRaw(rev string) (string, error) { |
| 205 | out, errs, err := r.run(nil, "rev-parse", "--verify", "--quiet", rev) |
| 206 | if err != nil { |
| 207 | return "", nil |
| 208 | } |
| 209 | sha := strings.TrimSpace(string(out)) |
| 210 | if sha == "" { |
| 211 | return "", fmt.Errorf("rev-parse %q: empty result: %s", rev, errs) |
| 212 | } |
| 213 | return sha, nil |
| 214 | } |
| 215 | |
| 216 | // ResolveAny resolves a rev to an object of any type (tree, blob, commit). |
| 217 | func (r *Repo) ResolveAny(rev string) (string, string, error) { |
| 218 | out, errs, err := r.run(nil, "cat-file", "-t", rev) |
| 219 | if err != nil { |
| 220 | return "", "", fmt.Errorf("resolve %q: %w: %s", rev, err, errs) |
| 221 | } |
| 222 | sha, err := r.RevParseAny(rev) |
| 223 | if err != nil { |
| 224 | return "", "", err |
| 225 | } |
| 226 | return sha, strings.TrimSpace(string(out)), nil |
| 227 | } |
| 228 | |
| 229 | func (r *Repo) RevParseAny(rev string) (string, error) { |
| 230 | out, errs, err := r.run(nil, "rev-parse", "--verify", "--quiet", rev) |
| 231 | if err != nil { |
| 232 | return "", fmt.Errorf("rev-parse %q: %w: %s", rev, err, errs) |
| 233 | } |
| 234 | return strings.TrimSpace(string(out)), nil |
| 235 | } |
| 236 | |
| 237 | // TreeOf resolves a revision to the tree it names. It is how a merge commit is |
| 238 | // compared with the tree a clean merge of its parents would produce, without |
| 239 | // reading either tree into memory. |
| 240 | func (r *Repo) TreeOf(rev string) (string, error) { |
| 241 | out, errs, err := r.run(nil, "rev-parse", "--verify", "--quiet", rev+"^{tree}") |
| 242 | if err != nil { |
| 243 | return "", fmt.Errorf("rev-parse %q^{tree}: %w: %s", rev, err, errs) |
| 244 | } |
| 245 | return strings.TrimSpace(string(out)), nil |
| 246 | } |
| 247 | |
| 248 | func (r *Repo) Exists() bool { |
| 249 | st, err := os.Stat(r.Dir) |
| 250 | return err == nil && st.IsDir() |
| 251 | } |
| 252 | |
| 253 | // LsTree lists the entries at treeish (a commit, a tree, or commit:path). |
| 254 | func (r *Repo) LsTree(treeish string) ([]Entry, error) { |
| 255 | return r.lsTree(false, false, treeish) |
| 256 | } |
| 257 | |
| 258 | // LsTreeR lists every blob under treeish recursively, with full paths. |
| 259 | func (r *Repo) LsTreeR(treeish string) ([]Entry, error) { |
| 260 | return r.lsTree(true, false, treeish) |
| 261 | } |
| 262 | |
| 263 | // LsTreeRT lists every entry under treeish recursively, including the trees |
| 264 | // themselves, with full paths. A writer uses it to find the subtree ids it can |
| 265 | // reuse; a reader that only wants files uses LsTreeR. |
| 266 | func (r *Repo) LsTreeRT(treeish string) ([]Entry, error) { |
| 267 | return r.lsTree(true, true, treeish) |
| 268 | } |
| 269 | |
| 270 | func (r *Repo) lsTree(recursive, withTrees bool, treeish string) ([]Entry, error) { |
| 271 | args := []string{"ls-tree"} |
| 272 | if recursive { |
| 273 | args = append(args, "-r") |
| 274 | } |
| 275 | if withTrees { |
| 276 | args = append(args, "-t") |
| 277 | } |
| 278 | args = append(args, "-z", treeish) |
| 279 | out, errs, err := r.run(nil, args...) |
| 280 | if err != nil { |
| 281 | return nil, fmt.Errorf("ls-tree %q: %w: %s", treeish, err, errs) |
| 282 | } |
| 283 | var es []Entry |
| 284 | for _, rec := range bytes.Split(out, []byte{0}) { |
| 285 | if len(rec) == 0 { |
| 286 | continue |
| 287 | } |
| 288 | tab := bytes.IndexByte(rec, '\t') |
| 289 | if tab < 0 { |
| 290 | return nil, fmt.Errorf("bad ls-tree record %q", rec) |
| 291 | } |
| 292 | f := strings.Fields(string(rec[:tab])) |
| 293 | if len(f) != 3 { |
| 294 | return nil, fmt.Errorf("bad ls-tree fields %q", rec) |
| 295 | } |
| 296 | es = append(es, Entry{Mode: f[0], Type: f[1], SHA: f[2], Name: string(rec[tab+1:])}) |
| 297 | } |
| 298 | return es, nil |
| 299 | } |
| 300 | |
| 301 | // BlobBySHA returns a blob's contents by object id. |
| 302 | func (r *Repo) BlobBySHA(sha string) ([]byte, error) { |
| 303 | out, errs, err := r.run(nil, "cat-file", "blob", sha) |
| 304 | if err != nil { |
| 305 | return nil, fmt.Errorf("cat-file blob %s: %w: %s", sha, err, errs) |
| 306 | } |
| 307 | return out, nil |
| 308 | } |
| 309 | |
| 310 | // BlobsBySHA returns the contents of every named blob in one git call. A caller |
| 311 | // that is about to read a whole tree uses it so the cost is one process, not one |
| 312 | // per file. The result is keyed by the object id git reports, which is the id |
| 313 | // that was asked for. |
| 314 | func (r *Repo) BlobsBySHA(shas []string) (map[string][]byte, error) { |
| 315 | if len(shas) == 0 { |
| 316 | return map[string][]byte{}, nil |
| 317 | } |
| 318 | var in bytes.Buffer |
| 319 | for _, sha := range shas { |
| 320 | if strings.ContainsAny(sha, "\n\r\x00\t ") { |
| 321 | return nil, fmt.Errorf("cat-file: malformed object id %q", sha) |
| 322 | } |
| 323 | in.WriteString(sha + "\n") |
| 324 | } |
| 325 | out, errs, err := r.run(in.Bytes(), "cat-file", "--batch") |
| 326 | if err != nil { |
| 327 | return nil, fmt.Errorf("cat-file --batch: %w: %s", err, errs) |
| 328 | } |
| 329 | blobs := make(map[string][]byte, len(shas)) |
| 330 | // The contents are sliced out of out rather than copied, so the map shares |
| 331 | // one buffer with the read: a large tree costs one copy, not two. |
| 332 | rest := out |
| 333 | for len(rest) > 0 { |
| 334 | i := bytes.IndexByte(rest, '\n') |
| 335 | if i < 0 { |
| 336 | return nil, fmt.Errorf("cat-file: unterminated object header") |
| 337 | } |
| 338 | header := string(rest[:i]) |
| 339 | rest = rest[i+1:] |
| 340 | fields := strings.Fields(header) |
| 341 | if len(fields) == 2 && fields[1] == "missing" { |
| 342 | return nil, fmt.Errorf("cat-file: %s is missing", fields[0]) |
| 343 | } |
| 344 | if len(fields) != 3 { |
| 345 | return nil, fmt.Errorf("cat-file: bad object header %q", header) |
| 346 | } |
| 347 | size, err := strconv.ParseInt(fields[2], 10, 64) |
| 348 | if err != nil || size < 0 || int64(len(rest)) < size+1 { |
| 349 | return nil, fmt.Errorf("cat-file: bad object size %q", fields[2]) |
| 350 | } |
| 351 | data := rest[:size] |
| 352 | // Every object is followed by one newline. It is framing, not content: a |
| 353 | // blob that does not end in a newline still gets one here. |
| 354 | if rest[size] != '\n' { |
| 355 | return nil, fmt.Errorf("cat-file: object %s is not newline-terminated", fields[0]) |
| 356 | } |
| 357 | blobs[fields[0]] = data |
| 358 | rest = rest[size+1:] |
| 359 | } |
| 360 | return blobs, nil |
| 361 | } |
| 362 | |
| 363 | // SetConfig writes a repo-local git config value. |
| 364 | func (r *Repo) SetConfig(key, value string) error { |
| 365 | _, err := r.Run(nil, "config", key, value) |
| 366 | return err |
| 367 | } |
| 368 | |
| 369 | func (r *Repo) MkTree(es []Entry) (string, error) { |
| 370 | sorted := make([]Entry, len(es)) |
| 371 | copy(sorted, es) |
| 372 | sort.Slice(sorted, func(i, j int) bool { return sorted[i].Name < sorted[j].Name }) |
| 373 | var buf bytes.Buffer |
| 374 | for _, e := range sorted { |
| 375 | fmt.Fprintf(&buf, "%s %s %s\t%s\x00", e.Mode, e.Type, e.SHA, e.Name) |
| 376 | } |
| 377 | out, errs, err := r.run(buf.Bytes(), "mktree", "-z") |
| 378 | if err != nil { |
| 379 | return "", fmt.Errorf("mktree: %w: %s", err, errs) |
| 380 | } |
| 381 | return strings.TrimSpace(string(out)), nil |
| 382 | } |
| 383 | |
| 384 | func (r *Repo) HashObject(payload []byte) (string, error) { |
| 385 | out, errs, err := r.run(payload, "hash-object", "-w", "--stdin") |
| 386 | if err != nil { |
| 387 | return "", fmt.Errorf("hash-object: %w: %s", err, errs) |
| 388 | } |
| 389 | return strings.TrimSpace(string(out)), nil |
| 390 | } |
| 391 | |
| 392 | // CommitTreeAs commits with a specific author, which is the audit record for a |
| 393 | // metadata change. |
| 394 | func (r *Repo) CommitTreeAs(tree, parent, msg, author string) (string, error) { |
| 395 | return r.commitTree(tree, parent, msg, author) |
| 396 | } |
| 397 | |
| 398 | func (r *Repo) commitTree(tree, parent, msg, author string) (string, error) { |
| 399 | parents := []string{} |
| 400 | if parent != "" { |
| 401 | parents = append(parents, parent) |
| 402 | } |
| 403 | return r.commitTreeParents(tree, parents, msg, author) |
| 404 | } |
| 405 | |
| 406 | // CommitMergeTree writes a merge commit over an already-built tree with the given |
| 407 | // parents, in order: the base tip first, then the approved head. The order matters |
| 408 | // to the gate, which accepts a merge whose parents are the old tip and the approved |
| 409 | // head in either order but nothing else. |
| 410 | func (r *Repo) CommitMergeTree(tree string, parents []string, msg, author string) (string, error) { |
| 411 | if len(parents) != 2 { |
| 412 | return "", fmt.Errorf("commit-tree: a merge commit needs two parents, got %d", len(parents)) |
| 413 | } |
| 414 | return r.commitTreeParents(tree, parents, msg, author) |
| 415 | } |
| 416 | |
| 417 | func (r *Repo) commitTreeParents(tree string, parents []string, msg, author string) (string, error) { |
| 418 | args := []string{"commit-tree", tree} |
| 419 | for _, p := range parents { |
| 420 | if p == "" { |
| 421 | continue |
| 422 | } |
| 423 | args = append(args, "-p", p) |
| 424 | } |
| 425 | env := []string{} |
| 426 | if strings.TrimSpace(author) != "" { |
| 427 | name := strings.TrimSpace(author) |
| 428 | env = append(env, |
| 429 | "GIT_AUTHOR_NAME="+name, "GIT_AUTHOR_EMAIL="+name+"@vole", |
| 430 | "GIT_COMMITTER_NAME="+name, "GIT_COMMITTER_EMAIL="+name+"@vole") |
| 431 | } |
| 432 | out, errs, err := r.runEnv(env, []byte(msg), args...) |
| 433 | if err != nil { |
| 434 | return "", fmt.Errorf("commit-tree: %w: %s", err, errs) |
| 435 | } |
| 436 | return strings.TrimSpace(string(out)), nil |
| 437 | } |
| 438 | |
| 439 | func (r *Repo) CommitTree(tree, parent, msg string) (string, error) { |
| 440 | return r.commitTree(tree, parent, msg, "") |
| 441 | } |
| 442 | |
| 443 | // LastTouchedByPath reports, for every path under dir reachable from rev, the newest |
| 444 | // commit that touched it. That commit is the one whose signature covers the bytes at |
| 445 | // the tip, which is what makes it the right commit to verify. |
| 446 | // |
| 447 | // The first answer wins, not the last: `git log` walks newest first, and a path added, |
| 448 | // deleted and re-added inside one push has several commits that "added" it. Verifying |
| 449 | // the oldest of those would check a signature over bytes that are no longer there. |
| 450 | // |
| 451 | // One `git log` answers for a whole subtree, so a record's attestation costs one |
| 452 | // process for all of them rather than one per file. |
| 453 | func (r *Repo) LastTouchedByPath(rev, dir string) (map[string]string, error) { |
| 454 | return r.touchedByPath(rev, dir, "%H") |
| 455 | } |
| 456 | |
| 457 | // TouchTimes is LastTouchedByPath for the question a list asks: when was each |
| 458 | // path under dir last changed? Rev may be a commit range, so an incremental |
| 459 | // reader only walks the commits it has not seen. |
| 460 | // |
| 461 | // The metadata format deliberately has no `updated` field — the commit that |
| 462 | // changed a file already carries the time, and writing it back would add a |
| 463 | // changed line to every edit. This is how the index turns one `git log` into a |
| 464 | // per-file time, rather than one process per entity. |
| 465 | func (r *Repo) TouchTimes(rev, dir string) (map[string]int64, error) { |
| 466 | touched, err := r.touchedByPath(rev, dir, "%at") |
| 467 | if err != nil { |
| 468 | return nil, err |
| 469 | } |
| 470 | if len(touched) == 0 { |
| 471 | return nil, nil |
| 472 | } |
| 473 | times := make(map[string]int64, len(touched)) |
| 474 | for path, when := range touched { |
| 475 | t, err := strconv.ParseInt(when, 10, 64) |
| 476 | if err != nil { |
| 477 | continue |
| 478 | } |
| 479 | times[path] = t |
| 480 | } |
| 481 | return times, nil |
| 482 | } |
| 483 | |
| 484 | // touchedByPath attributes every path under dir reachable from rev to the newest |
| 485 | // commit that touched it, reading the commit field named by format. |
| 486 | func (r *Repo) touchedByPath(rev, dir, format string) (map[string]string, error) { |
| 487 | if rev == "" || rev == ZeroSHA { |
| 488 | return nil, nil |
| 489 | } |
| 490 | out, errs, err := r.run(nil, "log", "--format=%x00"+format, "--name-only", rev, "--", dir) |
| 491 | if err != nil { |
| 492 | return nil, fmt.Errorf("log %s: %w: %s", dir, err, errs) |
| 493 | } |
| 494 | touched := map[string]string{} |
| 495 | // Each record starts with a NUL and the commit field, so paths that share a |
| 496 | // commit are attributed without guessing where one commit's list ends. |
| 497 | for _, chunk := range bytes.Split(out, []byte{0})[1:] { |
| 498 | lines := strings.Split(strings.TrimSpace(string(chunk)), "\n") |
| 499 | if len(lines) == 0 { |
| 500 | continue |
| 501 | } |
| 502 | value := strings.TrimSpace(lines[0]) |
| 503 | if value == "" { |
| 504 | continue |
| 505 | } |
| 506 | for _, line := range lines[1:] { |
| 507 | path := strings.TrimSpace(line) |
| 508 | if path == "" { |
| 509 | continue |
| 510 | } |
| 511 | if _, seen := touched[path]; !seen { |
| 512 | touched[path] = value |
| 513 | } |
| 514 | } |
| 515 | } |
| 516 | return touched, nil |
| 517 | } |
| 518 | |
| 519 | // UpdateRefCAS moves ref from old to new. Returns false when another writer won |
| 520 | // the race, which the caller retries. An empty old means "must not exist". |
| 521 | func (r *Repo) UpdateRefCAS(ref, new, old string) (bool, error) { |
| 522 | if old == "" { |
| 523 | old = ZeroSHA |
| 524 | } |
| 525 | _, errs, err := r.run(nil, "update-ref", ref, new, old) |
| 526 | if err == nil { |
| 527 | return true, nil |
| 528 | } |
| 529 | if strings.Contains(errs, "cannot lock ref") || |
| 530 | strings.Contains(errs, "is at") || |
| 531 | strings.Contains(errs, "unable to resolve") || |
| 532 | strings.Contains(errs, "reference already exists") { |
| 533 | return false, nil |
| 534 | } |
| 535 | return false, fmt.Errorf("update-ref %s: %w: %s", ref, err, errs) |
| 536 | } |
| 537 | |
| 538 | func (r *Repo) UpdateRef(ref, new string) error { |
| 539 | _, err := r.Run(nil, "update-ref", ref, new) |
| 540 | return err |
| 541 | } |
| 542 | |
| 543 | func (r *Repo) DeleteRef(ref string) error { |
| 544 | _, err := r.Run(nil, "update-ref", "-d", ref) |
| 545 | return err |
| 546 | } |
| 547 | |
| 548 | // MaxRefsListed bounds what a listing returns. A page that lists refs walks every |
| 549 | // one of them, and a repository's ref count is something its contributors choose; |
| 550 | // the *quota* check uses RefCount, which must see all of them, but nothing a |
| 551 | // browser renders needs more than this. |
| 552 | const MaxRefsListed = 500 |
| 553 | |
| 554 | func (r *Repo) ListRefs(prefix string) (map[string]string, error) { |
| 555 | out, err := r.Run(nil, "for-each-ref", "--count="+strconv.Itoa(MaxRefsListed), |
| 556 | "--format=%(refname)%1f%(objectname)", prefix) |
| 557 | if err != nil { |
| 558 | return nil, err |
| 559 | } |
| 560 | m := map[string]string{} |
| 561 | for _, line := range strings.Split(strings.TrimSpace(string(out)), "\n") { |
| 562 | if line == "" { |
| 563 | continue |
| 564 | } |
| 565 | f := strings.SplitN(line, "\x1f", 2) |
| 566 | if len(f) == 2 { |
| 567 | m[f[0]] = f[1] |
| 568 | } |
| 569 | } |
| 570 | return m, nil |
| 571 | } |
| 572 | |
| 573 | // RefEntry is one ref and the creation time git reports for it. A branch's time |
| 574 | // is the committer date of its tip, a tag's is its tagger or commit date, which |
| 575 | // is what makes one listing enough to sort both. |
| 576 | type RefEntry struct { |
| 577 | Name string |
| 578 | When int64 |
| 579 | } |
| 580 | |
| 581 | // Refs lists the refs under every given prefix in one call. The ref picker shows |
| 582 | // branches and tags together, and asking for them separately is a second process |
| 583 | // per page for no reason. Like ListRefs, the listing stops at MaxRefsListed; the |
| 584 | // cap is shared by the prefixes, which is the same page-render bound the callers |
| 585 | // applied one list at a time. |
| 586 | func (r *Repo) Refs(prefixes ...string) ([]RefEntry, error) { |
| 587 | if len(prefixes) == 0 { |
| 588 | return nil, nil |
| 589 | } |
| 590 | args := make([]string, 0, len(prefixes)+3) |
| 591 | args = append(args, "for-each-ref", "--count="+strconv.Itoa(MaxRefsListed), |
| 592 | "--format=%(refname)%1f%(creatordate:unix)") |
| 593 | args = append(args, prefixes...) |
| 594 | out, err := r.Run(nil, args...) |
| 595 | if err != nil { |
| 596 | return nil, err |
| 597 | } |
| 598 | var refs []RefEntry |
| 599 | for _, line := range strings.Split(strings.TrimSpace(string(out)), "\n") { |
| 600 | name, when, ok := strings.Cut(line, "\x1f") |
| 601 | if !ok || name == "" { |
| 602 | continue |
| 603 | } |
| 604 | n, _ := strconv.ParseInt(when, 10, 64) |
| 605 | refs = append(refs, RefEntry{Name: name, When: n}) |
| 606 | } |
| 607 | return refs, nil |
| 608 | } |
| 609 | |
| 610 | // BlobAt returns the contents of a blob at commit:path. |
| 611 | func (r *Repo) BlobAt(commit, path string) ([]byte, error) { |
| 612 | out, errs, err := r.run(nil, "cat-file", "blob", commit+":"+path) |
| 613 | if err != nil { |
| 614 | return nil, fmt.Errorf("cat-file %s:%s: %w: %s", commit, path, err, errs) |
| 615 | } |
| 616 | return out, nil |
| 617 | } |
| 618 | |
| 619 | // BlobSize returns the length of the object at rev without reading it. It is |
| 620 | // what lets a reader decide to stream a 100 MiB file or to render only its first |
| 621 | // page, instead of allocating the whole thing and measuring it afterwards. |
| 622 | func (r *Repo) BlobSize(rev string) (int64, error) { |
| 623 | out, errs, err := r.run(nil, "cat-file", "-s", rev) |
| 624 | if err != nil { |
| 625 | return 0, fmt.Errorf("cat-file -s %s: %w: %s", rev, err, errs) |
| 626 | } |
| 627 | n, err := strconv.ParseInt(strings.TrimSpace(string(out)), 10, 64) |
| 628 | if err != nil { |
| 629 | return 0, fmt.Errorf("cat-file -s %s: bad size %q", rev, out) |
| 630 | } |
| 631 | return n, nil |
| 632 | } |
| 633 | |
| 634 | // BlobSizes returns the size of every named blob in one git call. A loader that |
| 635 | // is about to read a whole tree of files uses it to refuse the oversized ones |
| 636 | // before any of them becomes a []byte. |
| 637 | func (r *Repo) BlobSizes(shas []string) (map[string]int64, error) { |
| 638 | for _, sha := range shas { |
| 639 | if strings.ContainsAny(sha, "\n\r\x00\t ") { |
| 640 | return nil, fmt.Errorf("cat-file: malformed object id %q", sha) |
| 641 | } |
| 642 | } |
| 643 | return r.blobSizes(shas) |
| 644 | } |
| 645 | |
| 646 | // ErrObjectTooLarge reports that a caller asked for an object that exceeds the |
| 647 | // cap it named. It is a policy answer, not a broken repository. |
| 648 | var ErrObjectTooLarge = errors.New("git: object exceeds size limit") |
| 649 | |
| 650 | // BlobStream is one object read from git as a stream, with its header already |
| 651 | // parsed. It exists so a download can be copied to the wire without the object |
| 652 | // first becoming a []byte: serving a 100 MiB blob costs one 64 KiB buffer no |
| 653 | // matter how many readers ask for it at once. |
| 654 | // |
| 655 | // A stream holds a git process open, so every one of them must be closed. Close |
| 656 | // on a partly read stream kills the process instead of draining the rest: the |
| 657 | // caller stopped caring about the bytes. |
| 658 | type BlobStream struct { |
| 659 | SHA string |
| 660 | Type string |
| 661 | Size int64 |
| 662 | |
| 663 | br *bufio.Reader |
| 664 | cmd *exec.Cmd |
| 665 | in io.WriteCloser |
| 666 | left int64 |
| 667 | done bool |
| 668 | closed bool |
| 669 | } |
| 670 | |
| 671 | // OpenBlob resolves rev (a revision, a commit:path pair, or an object id) and |
| 672 | // returns its header with the reader positioned at the first payload byte. |
| 673 | func (r *Repo) OpenBlob(rev string) (*BlobStream, error) { |
| 674 | if rev == "" || strings.ContainsAny(rev, "\n\r\x00") { |
| 675 | return nil, fmt.Errorf("cat-file: malformed revision %q", rev) |
| 676 | } |
| 677 | cmd := r.command("cat-file", "--batch") |
| 678 | in, err := cmd.StdinPipe() |
| 679 | if err != nil { |
| 680 | return nil, err |
| 681 | } |
| 682 | out, err := cmd.StdoutPipe() |
| 683 | if err != nil { |
| 684 | in.Close() |
| 685 | return nil, err |
| 686 | } |
| 687 | cmd.Stderr = io.Discard |
| 688 | if err := cmd.Start(); err != nil { |
| 689 | in.Close() |
| 690 | return nil, err |
| 691 | } |
| 692 | kill := func() { |
| 693 | in.Close() |
| 694 | cmd.Process.Kill() |
| 695 | cmd.Wait() |
| 696 | } |
| 697 | if _, err := io.WriteString(in, rev+"\n"); err != nil { |
| 698 | kill() |
| 699 | return nil, fmt.Errorf("cat-file %s: %w", rev, err) |
| 700 | } |
| 701 | br := bufio.NewReaderSize(out, 64<<10) |
| 702 | line, err := br.ReadString('\n') |
| 703 | if err != nil { |
| 704 | kill() |
| 705 | return nil, fmt.Errorf("cat-file %s: %w", rev, err) |
| 706 | } |
| 707 | fields := strings.Fields(strings.TrimRight(line, "\n")) |
| 708 | if len(fields) == 2 && fields[1] == "missing" { |
| 709 | kill() |
| 710 | return nil, fmt.Errorf("cat-file %s: object not found", rev) |
| 711 | } |
| 712 | if len(fields) != 3 { |
| 713 | kill() |
| 714 | return nil, fmt.Errorf("cat-file %s: unexpected header %q", rev, line) |
| 715 | } |
| 716 | size, err := strconv.ParseInt(fields[2], 10, 64) |
| 717 | if err != nil || size < 0 { |
| 718 | kill() |
| 719 | return nil, fmt.Errorf("cat-file %s: bad size in %q", rev, line) |
| 720 | } |
| 721 | return &BlobStream{SHA: fields[0], Type: fields[1], Size: size, br: br, cmd: cmd, in: in, left: size}, nil |
| 722 | } |
| 723 | |
| 724 | // OpenBlobAt is OpenBlob for commit:path. |
| 725 | func (r *Repo) OpenBlobAt(commit, path string) (*BlobStream, error) { |
| 726 | return r.OpenBlob(commit + ":" + path) |
| 727 | } |
| 728 | |
| 729 | // Read implements io.Reader over the object's payload. It stops at the object's |
| 730 | // own length; the trailing newline git writes after each object is eaten by the |
| 731 | // last read. |
| 732 | func (b *BlobStream) Read(p []byte) (int, error) { |
| 733 | if b.closed || b.left <= 0 { |
| 734 | return 0, io.EOF |
| 735 | } |
| 736 | if int64(len(p)) > b.left { |
| 737 | p = p[:b.left] |
| 738 | } |
| 739 | n, err := b.br.Read(p) |
| 740 | b.left -= int64(n) |
| 741 | if b.left == 0 { |
| 742 | b.finish() |
| 743 | } |
| 744 | return n, err |
| 745 | } |
| 746 | |
| 747 | // finish eats the object's trailing newline and reaps the process. A failure |
| 748 | // here is not the reader's business: the payload is already delivered. |
| 749 | func (b *BlobStream) finish() { |
| 750 | if b.done { |
| 751 | return |
| 752 | } |
| 753 | b.done = true |
| 754 | b.br.Discard(1) |
| 755 | b.in.Close() |
| 756 | b.cmd.Wait() |
| 757 | } |
| 758 | |
| 759 | // Close releases the process. It is safe to call more than once and after a |
| 760 | // full read. |
| 761 | func (b *BlobStream) Close() error { |
| 762 | if b.closed { |
| 763 | return nil |
| 764 | } |
| 765 | b.closed = true |
| 766 | if !b.done { |
| 767 | b.done = true |
| 768 | b.in.Close() |
| 769 | if b.cmd.Process != nil { |
| 770 | b.cmd.Process.Kill() |
| 771 | } |
| 772 | b.cmd.Wait() |
| 773 | } |
| 774 | return nil |
| 775 | } |
| 776 | |
| 777 | // ReadPrefix returns at most max bytes of the stream and reports whether more |
| 778 | // bytes were left behind. The rest is not read: the process is stopped. |
| 779 | func (b *BlobStream) ReadPrefix(max int64) ([]byte, bool, error) { |
| 780 | if max < 0 { |
| 781 | max = 0 |
| 782 | } |
| 783 | take := b.left |
| 784 | if take > max { |
| 785 | take = max |
| 786 | } |
| 787 | buf := make([]byte, take) |
| 788 | if _, err := io.ReadFull(b, buf); err != nil && !errors.Is(err, io.EOF) && !errors.Is(err, io.ErrUnexpectedEOF) { |
| 789 | return nil, false, err |
| 790 | } |
| 791 | return buf, b.left > 0, nil |
| 792 | } |
| 793 | |
| 794 | // BlobPrefixAt returns at most max bytes of the blob at commit:path and whether |
| 795 | // there were more. It is the read for anything that renders a file inline: a |
| 796 | // caller that wants a preview never needs more than the preview. |
| 797 | func (r *Repo) BlobPrefixAt(commit, path string, max int64) ([]byte, bool, error) { |
| 798 | stream, err := r.OpenBlobAt(commit, path) |
| 799 | if err != nil { |
| 800 | return nil, false, err |
| 801 | } |
| 802 | defer stream.Close() |
| 803 | return stream.ReadPrefix(max) |
| 804 | } |
| 805 | |
| 806 | func (r *Repo) ObjectType(rev string) (string, error) { |
| 807 | out, _, err := r.run(nil, "cat-file", "-t", rev) |
| 808 | if err != nil { |
| 809 | return "", err |
| 810 | } |
| 811 | return strings.TrimSpace(string(out)), nil |
| 812 | } |
| 813 | |
| 814 | // Symref returns the branch HEAD points at, if any. |
| 815 | func (r *Repo) Symref() string { |
| 816 | out, _, err := r.run(nil, "symbolic-ref", "--short", "HEAD") |
| 817 | if err != nil { |
| 818 | return "" |
| 819 | } |
| 820 | return strings.TrimSpace(string(out)) |
| 821 | } |
| 822 | |
| 823 | // SetHEAD points HEAD at an existing branch. It does not create the branch; the |
| 824 | // caller decides whether an unborn HEAD is acceptable. |
| 825 | func (r *Repo) SetHEAD(branch string) error { |
| 826 | _, errs, err := r.run(nil, "symbolic-ref", "HEAD", "refs/heads/"+branch) |
| 827 | if err != nil { |
| 828 | return fmt.Errorf("set HEAD to %s: %w: %s", branch, err, errs) |
| 829 | } |
| 830 | return nil |
| 831 | } |
| 832 | |
| 833 | // ValidTag reports whether name is a syntactically valid tag. The leading-dash |
| 834 | // check keeps a name from being read as an option by git. |
| 835 | func (r *Repo) ValidTag(name string) bool { |
| 836 | if name == "" || strings.HasPrefix(name, "-") || strings.Contains(name, "\x00") { |
| 837 | return false |
| 838 | } |
| 839 | _, _, err := r.run(nil, "check-ref-format", "refs/tags/"+name) |
| 840 | return err == nil |
| 841 | } |
| 842 | |
| 843 | // WriteAnnotatedTag creates, or with force moves, an annotated tag. The tagger is |
| 844 | // the account that asked for it, not the server process, so the tag explains who |
| 845 | // made the release. |
| 846 | func (r *Repo) WriteAnnotatedTag(name, target, message, tagger string, force bool) error { |
| 847 | if !r.ValidTag(name) { |
| 848 | return fmt.Errorf("%q is not a valid tag name", name) |
| 849 | } |
| 850 | args := []string{"tag", "-a"} |
| 851 | if force { |
| 852 | args = append(args, "-f") |
| 853 | } |
| 854 | args = append(args, name, "-m", message, target) |
| 855 | _, errs, err := r.runEnv([]string{ |
| 856 | "GIT_COMMITTER_NAME=" + tagger, |
| 857 | "GIT_COMMITTER_EMAIL=" + tagger + "@vole.local", |
| 858 | }, nil, args...) |
| 859 | if err != nil { |
| 860 | return fmt.Errorf("tag %s: %w: %s", name, err, errs) |
| 861 | } |
| 862 | return nil |
| 863 | } |
| 864 | |
| 865 | // ValidBranch reports whether name is a syntactically valid branch. It checks |
| 866 | // the full ref rather than using --branch, which would expand @{-1} and similar |
| 867 | // history syntax instead of validating the bytes the caller wrote. |
| 868 | func (r *Repo) ValidBranch(name string) bool { |
| 869 | if name == "" || strings.HasPrefix(name, "-") || strings.Contains(name, "\x00") { |
| 870 | return false |
| 871 | } |
| 872 | _, _, err := r.run(nil, "check-ref-format", "refs/heads/"+name) |
| 873 | return err == nil |
| 874 | } |
| 875 | |
| 876 | // commitFormat is one line per commit: sha, author, email, epoch, subject. |
| 877 | // %x1f is a unit separator, which cannot appear in a subject the way a tab or a |
| 878 | // newline can. |
| 879 | const commitFormat = "%H%x1f%an%x1f%ae%x1f%at%x1f%s" |
| 880 | |
| 881 | type Commit struct { |
| 882 | SHA string |
| 883 | Author string |
| 884 | Email string |
| 885 | When int64 |
| 886 | Subject string |
| 887 | } |
| 888 | |
| 889 | // repositoryEnv lists the variables git sets to point at a specific repository. |
| 890 | // They are inherited by hooks: pre-receive runs with GIT_QUARANTINE_PATH and |
| 891 | // friends set for the repository being pushed, so a hook that shells out to git |
| 892 | // for a *different* repository (the roster, for instance) would silently read the |
| 893 | // pushed one instead. Every command we run names its repository explicitly, so |
| 894 | // these are always dropped. |
| 895 | var repositoryEnv = map[string]bool{ |
| 896 | "GIT_DIR": true, |
| 897 | "GIT_WORK_TREE": true, |
| 898 | "GIT_COMMON_DIR": true, |
| 899 | "GIT_INDEX_FILE": true, |
| 900 | "GIT_OBJECT_DIRECTORY": true, |
| 901 | "GIT_ALTERNATE_OBJECT_DIRECTORIES": true, |
| 902 | "GIT_QUARANTINE_PATH": true, |
| 903 | "GIT_NAMESPACE": true, |
| 904 | "GIT_PREFIX": true, |
| 905 | } |
| 906 | |
| 907 | func scrubbedEnv() []string { |
| 908 | env := os.Environ() |
| 909 | out := make([]string, 0, len(env)) |
| 910 | for _, kv := range env { |
| 911 | name := kv |
| 912 | if i := strings.IndexByte(kv, '='); i >= 0 { |
| 913 | name = kv[:i] |
| 914 | } |
| 915 | if repositoryEnv[name] { |
| 916 | continue |
| 917 | } |
| 918 | out = append(out, kv) |
| 919 | } |
| 920 | return out |
| 921 | } |
| 922 | |
| 923 | // Log returns up to n commits from rev. |
| 924 | func (r *Repo) Log(rev string, n int) ([]Commit, error) { |
| 925 | args := []string{"log", "--format=" + commitFormat} |
| 926 | if n > 0 { |
| 927 | args = append(args, "-n", strconv.Itoa(n)) |
| 928 | } |
| 929 | args = append(args, rev, "--") |
| 930 | out, errs, err := r.run(nil, args...) |
| 931 | if err != nil { |
| 932 | return nil, fmt.Errorf("log %q: %w: %s", rev, err, errs) |
| 933 | } |
| 934 | return parseCommits(out), nil |
| 935 | } |
| 936 | |
| 937 | // LogRange returns the commits reachable from to but not from, oldest first. It is |
| 938 | // what a reconciler reads to apply every step of a branch advance it missed, rather |
| 939 | // than only the net difference between two tips. |
| 940 | func (r *Repo) LogRange(from, to string) ([]Commit, error) { |
| 941 | args := []string{"log", "--reverse", "--format=" + commitFormat} |
| 942 | if from == "" || from == ZeroSHA { |
| 943 | args = append(args, to, "--") |
| 944 | } else { |
| 945 | args = append(args, from+".."+to, "--") |
| 946 | } |
| 947 | out, errs, err := r.run(nil, args...) |
| 948 | if err != nil { |
| 949 | return nil, fmt.Errorf("log %s..%s: %w: %s", from, to, err, errs) |
| 950 | } |
| 951 | return parseCommits(out), nil |
| 952 | } |
| 953 | |
| 954 | // RevParseFull resolves a revision to a full commit id. It refuses ambiguous |
| 955 | // short ids rather than picking one. |
| 956 | func (r *Repo) RevParseFull(rev string) (string, error) { |
| 957 | out, errs, err := r.run(nil, "rev-parse", "--verify", "--quiet", "--end-of-options", rev+"^{commit}") |
| 958 | if err != nil { |
| 959 | return "", fmt.Errorf("rev-parse %s: %w: %s", rev, err, errs) |
| 960 | } |
| 961 | sha := strings.TrimSpace(string(out)) |
| 962 | if len(sha) != 40 { |
| 963 | return "", fmt.Errorf("%q is not a commit", rev) |
| 964 | } |
| 965 | return sha, nil |
| 966 | } |
| 967 | |
| 968 | // CurrentBranch returns the checked-out branch name, or "" when detached. |
| 969 | func (r *Repo) CurrentBranch() (string, error) { |
| 970 | out, _, err := r.run(nil, "symbolic-ref", "--short", "-q", "HEAD") |
| 971 | if err != nil { |
| 972 | return "", nil // detached or unborn: nothing is checked out by name |
| 973 | } |
| 974 | return strings.TrimSpace(string(out)), nil |
| 975 | } |
| 976 | |
| 977 | // DiffNameStatus lists the paths a branch changes relative to from, using the |
| 978 | // merge base so unrelated history on from does not show up as changes. |
| 979 | func (r *Repo) DiffNameStatus(from, to string) ([]string, error) { |
| 980 | out, errs, err := r.run(nil, "diff", "--name-status", from+"..."+to) |
| 981 | if err != nil { |
| 982 | return nil, fmt.Errorf("diff %s...%s: %w: %s", from, to, err, errs) |
| 983 | } |
| 984 | var lines []string |
| 985 | for _, line := range strings.Split(strings.TrimSpace(string(out)), "\n") { |
| 986 | if strings.TrimSpace(line) != "" { |
| 987 | lines = append(lines, strings.Replace(line, "\t", " ", -1)) |
| 988 | } |
| 989 | } |
| 990 | return lines, nil |
| 991 | } |
| 992 | |
| 993 | // DiffStatus returns the change status of each path between two commits, as the |
| 994 | // single letter git uses: A added, M modified, D deleted, T type changed, R |
| 995 | // renamed, C copied. Callers that guard a path must treat every status except D |
| 996 | // as a write, or a check that only looks at additions can be walked past by |
| 997 | // editing a file that already exists. |
| 998 | func (r *Repo) DiffStatus(old, new string) (map[string]string, error) { |
| 999 | return r.DiffStatusPaths(old, new) |
| 1000 | } |
| 1001 | |
| 1002 | // DiffStatusPaths is DiffStatus limited to the given pathspecs. An incremental |
| 1003 | // refresh only cares about the subtrees it indexes, so a range that also carries |
| 1004 | // unrelated changes does not scan them. |
| 1005 | func (r *Repo) DiffStatusPaths(old, new string, pathspecs ...string) (map[string]string, error) { |
| 1006 | args := []string{"diff-tree", "-r", "--no-commit-id", "--name-status", old, new} |
| 1007 | if len(pathspecs) > 0 { |
| 1008 | args = append(args, "--") |
| 1009 | args = append(args, pathspecs...) |
| 1010 | } |
| 1011 | out, errs, err := r.run(nil, args...) |
| 1012 | if err != nil { |
| 1013 | return nil, fmt.Errorf("diff-tree %s..%s: %w: %s", old, new, err, errs) |
| 1014 | } |
| 1015 | status := map[string]string{} |
| 1016 | for _, line := range strings.Split(strings.TrimSpace(string(out)), "\n") { |
| 1017 | if line == "" { |
| 1018 | continue |
| 1019 | } |
| 1020 | parts := strings.SplitN(line, "\t", 2) |
| 1021 | if len(parts) != 2 { |
| 1022 | continue |
| 1023 | } |
| 1024 | // Rename and copy entries carry two paths; the destination is the one that |
| 1025 | // matters for a write check. |
| 1026 | status[strings.TrimSpace(parts[1])] = strings.TrimSpace(parts[0])[:1] |
| 1027 | } |
| 1028 | return status, nil |
| 1029 | } |
| 1030 | |
| 1031 | // CommitParents returns a commit's parents, in order. |
| 1032 | func (r *Repo) CommitParents(sha string) ([]string, error) { |
| 1033 | out, errs, err := r.run(nil, "rev-list", "--parents", "-n", "1", sha) |
| 1034 | if err != nil { |
| 1035 | return nil, fmt.Errorf("rev-list --parents %s: %w: %s", sha, err, errs) |
| 1036 | } |
| 1037 | f := strings.Fields(strings.TrimSpace(string(out))) |
| 1038 | if len(f) == 0 { |
| 1039 | return nil, fmt.Errorf("no such commit %s", sha) |
| 1040 | } |
| 1041 | return f[1:], nil |
| 1042 | } |
| 1043 | |
| 1044 | // IsAncestor reports whether ancestor is reachable from descendant. Used to tell |
| 1045 | // a fast-forward push from a rewrite. |
| 1046 | func (r *Repo) IsAncestor(ancestor, descendant string) (bool, error) { |
| 1047 | if ancestor == "" || ancestor == ZeroSHA { |
| 1048 | return true, nil // a new branch cannot rewrite anything |
| 1049 | } |
| 1050 | _, _, err := r.run(nil, "merge-base", "--is-ancestor", ancestor, descendant) |
| 1051 | if err == nil { |
| 1052 | return true, nil |
| 1053 | } |
| 1054 | if ee, ok := err.(*exec.ExitError); ok && ee.ExitCode() == 1 { |
| 1055 | return false, nil |
| 1056 | } |
| 1057 | return false, fmt.Errorf("merge-base --is-ancestor %s %s: %w", ancestor, descendant, err) |
| 1058 | } |
| 1059 | |
| 1060 | // CountMerges counts merge commits in old..new, or in all of new when old is the |
| 1061 | // zero SHA. |
| 1062 | func (r *Repo) CountMerges(old, new string) (int, error) { |
| 1063 | args := []string{"rev-list", "--count", "--merges", new} |
| 1064 | if old != "" && old != ZeroSHA { |
| 1065 | args = []string{"rev-list", "--count", "--merges", old + ".." + new} |
| 1066 | } |
| 1067 | out, errs, err := r.run(nil, args...) |
| 1068 | if err != nil { |
| 1069 | if strings.Contains(errs, "bad revision") || strings.Contains(errs, "unknown revision") { |
| 1070 | return 0, nil // history we cannot read is not a merge we can prove |
| 1071 | } |
| 1072 | return 0, fmt.Errorf("rev-list --merges: %w: %s", err, errs) |
| 1073 | } |
| 1074 | n, err := strconv.Atoi(strings.TrimSpace(string(out))) |
| 1075 | if err != nil { |
| 1076 | return 0, nil |
| 1077 | } |
| 1078 | return n, nil |
| 1079 | } |
| 1080 | |
| 1081 | // LogGrep returns commits whose message mentions pattern. Used for issue |
| 1082 | // backlinks, so the search happens in git rather than in a second index. |
| 1083 | func (r *Repo) LogGrep(rev, pattern string, n int) ([]Commit, error) { |
| 1084 | out, errs, err := r.run(nil, "log", |
| 1085 | "--grep="+pattern, "--fixed-strings", "--format="+commitFormat, |
| 1086 | "-n", strconv.Itoa(n), rev, "--") |
| 1087 | if err != nil { |
| 1088 | return nil, fmt.Errorf("log --grep %q: %w: %s", pattern, err, errs) |
| 1089 | } |
| 1090 | return parseCommits(out), nil |
| 1091 | } |
| 1092 | |
| 1093 | func parseCommits(out []byte) []Commit { |
| 1094 | var cs []Commit |
| 1095 | for _, line := range strings.Split(strings.TrimSpace(string(out)), "\n") { |
| 1096 | if line == "" { |
| 1097 | continue |
| 1098 | } |
| 1099 | f := strings.SplitN(line, "\x1f", 5) |
| 1100 | if len(f) != 5 { |
| 1101 | continue |
| 1102 | } |
| 1103 | ts, _ := strconv.ParseInt(f[3], 10, 64) |
| 1104 | cs = append(cs, Commit{SHA: f[0], Author: f[1], Email: f[2], When: ts, Subject: f[4]}) |
| 1105 | } |
| 1106 | return cs |
| 1107 | } |
| 1108 | |
| 1109 | // Grep returns fixed-string, case-insensitive, non-binary matches as raw |
| 1110 | // path:line:text records. It stops reading after maxBytes, so a broad query |
| 1111 | // cannot make the server hold an unbounded result set. |
| 1112 | func (r *Repo) Grep(rev, pattern string, maxBytes int) ([]byte, bool, error) { |
| 1113 | if maxBytes < 1 { |
| 1114 | maxBytes = 1 |
| 1115 | } |
| 1116 | out, _, err := r.runHead(maxBytes+1, |
| 1117 | "grep", "-n", "-I", "-i", "-F", "-e", pattern, rev, "--") |
| 1118 | if err != nil { |
| 1119 | return nil, false, fmt.Errorf("grep %q: %w", pattern, err) |
| 1120 | } |
| 1121 | truncated := len(out) > maxBytes |
| 1122 | if truncated { |
| 1123 | out = out[:maxBytes] |
| 1124 | } |
| 1125 | return out, truncated, nil |
| 1126 | } |
| 1127 | |
| 1128 | // LogPath lists commits that touched one path. follow asks git to follow a |
| 1129 | // rename; it is only valid for a single file, so directory history uses false. |
| 1130 | func (r *Repo) LogPath(rev, path string, n int, follow bool) ([]Commit, error) { |
| 1131 | args := []string{"log", "--format=" + commitFormat, "-n", strconv.Itoa(n)} |
| 1132 | if follow { |
| 1133 | args = append(args, "--follow") |
| 1134 | } |
| 1135 | args = append(args, rev, "--", path) |
| 1136 | out, errs, err := r.run(nil, args...) |
| 1137 | if err != nil { |
| 1138 | return nil, fmt.Errorf("log %q -- %q: %w: %s", rev, path, err, errs) |
| 1139 | } |
| 1140 | return parseCommits(out), nil |
| 1141 | } |
| 1142 | |
| 1143 | // addMarker starts each commit's block in AddedPathsPerCommit's output. It is a |
| 1144 | // control character no path in a metadata tree can contain, so a path can never |
| 1145 | // be mistaken for the separator. git takes it as a %x escape because a pretty |
| 1146 | // format rejects a literal control byte. |
| 1147 | const ( |
| 1148 | addMarker = "\x01" |
| 1149 | addMarkerSpec = "%x01" |
| 1150 | ) |
| 1151 | |
| 1152 | // AddedPathsPerCommit lists, oldest first, the paths each commit added, limited |
| 1153 | // to the given pathspecs. A commit that added nothing under them is absent from |
| 1154 | // the result, so a caller counting blocks is counting only commits that changed |
| 1155 | // the tree it asked about. |
| 1156 | func (r *Repo) AddedPathsPerCommit(rev string, pathspecs ...string) ([][]string, error) { |
| 1157 | args := []string{"log", "--reverse", "--diff-filter=A", "--name-only", "--no-renames", |
| 1158 | "--format=" + addMarkerSpec, rev, "--"} |
| 1159 | args = append(args, pathspecs...) |
| 1160 | out, errs, err := r.run(nil, args...) |
| 1161 | if err != nil { |
| 1162 | return nil, fmt.Errorf("log %q added paths: %w: %s", rev, err, errs) |
| 1163 | } |
| 1164 | var groups [][]string |
| 1165 | var cur []string |
| 1166 | flush := func() { |
| 1167 | if len(cur) > 0 { |
| 1168 | groups = append(groups, cur) |
| 1169 | cur = nil |
| 1170 | } |
| 1171 | } |
| 1172 | for _, line := range strings.Split(string(out), "\n") { |
| 1173 | line = strings.TrimRight(line, "\r") |
| 1174 | switch line { |
| 1175 | case addMarker: |
| 1176 | flush() |
| 1177 | case "": |
| 1178 | // git separates the format line and the names with blank lines. |
| 1179 | default: |
| 1180 | cur = append(cur, line) |
| 1181 | } |
| 1182 | } |
| 1183 | flush() |
| 1184 | return groups, nil |
| 1185 | } |
| 1186 | |
| 1187 | // BlameLine is one line of `git blame` output with the commit that last touched |
| 1188 | // it. Line is the line number in the blamed revision. |
| 1189 | type BlameLine struct { |
| 1190 | SHA string |
| 1191 | Author string |
| 1192 | When int64 |
| 1193 | Line int |
| 1194 | Text string |
| 1195 | } |
| 1196 | |
| 1197 | // Blame reads line attribution for one file, bounded by maxBytes. A cut result |
| 1198 | // is reported rather than silently shown as the whole file. |
| 1199 | func (r *Repo) Blame(rev, path string, maxBytes int) ([]BlameLine, bool, error) { |
| 1200 | if maxBytes < 1 { |
| 1201 | maxBytes = 1 |
| 1202 | } |
| 1203 | out, _, err := r.runHead(maxBytes+1, |
| 1204 | "blame", "--line-porcelain", rev, "--", path) |
| 1205 | if err != nil { |
| 1206 | return nil, false, fmt.Errorf("blame %q -- %q: %w", rev, path, err) |
| 1207 | } |
| 1208 | truncated := len(out) > maxBytes |
| 1209 | if truncated { |
| 1210 | out = out[:maxBytes] |
| 1211 | } |
| 1212 | var lines []BlameLine |
| 1213 | var cur BlameLine |
| 1214 | have := false |
| 1215 | for _, line := range strings.Split(string(out), "\n") { |
| 1216 | switch { |
| 1217 | case strings.HasPrefix(line, "\t"): |
| 1218 | if have { |
| 1219 | cur.Text = line[1:] |
| 1220 | lines = append(lines, cur) |
| 1221 | have = false |
| 1222 | } |
| 1223 | case strings.HasPrefix(line, "author "): |
| 1224 | cur.Author = strings.TrimPrefix(line, "author ") |
| 1225 | case strings.HasPrefix(line, "author-time "): |
| 1226 | cur.When, _ = strconv.ParseInt(strings.TrimPrefix(line, "author-time "), 10, 64) |
| 1227 | default: |
| 1228 | fields := strings.Fields(line) |
| 1229 | if len(fields) >= 3 { |
| 1230 | sha := strings.TrimPrefix(fields[0], "^") |
| 1231 | if n, err := strconv.Atoi(fields[2]); err == nil && len(sha) == 40 { |
| 1232 | cur = BlameLine{SHA: sha, Line: n} |
| 1233 | have = true |
| 1234 | } |
| 1235 | } |
| 1236 | } |
| 1237 | } |
| 1238 | return lines, truncated, nil |
| 1239 | } |
| 1240 | |
| 1241 | func (r *Repo) CountCommits(rev string) (int, error) { |
| 1242 | out, _, err := r.run(nil, "rev-list", "--count", rev) |
| 1243 | if err != nil { |
| 1244 | return 0, nil // unborrn branch |
| 1245 | } |
| 1246 | n, _ := strconv.Atoi(strings.TrimSpace(string(out))) |
| 1247 | return n, nil |
| 1248 | } |
| 1249 | |
| 1250 | // NewBlobSizes returns the size of every blob reachable from new but not from |
| 1251 | // old. Used by the pre-receive hook to enforce object size limits without |
| 1252 | // walking the entire history. |
| 1253 | func (r *Repo) NewBlobSizes(old, new string) (map[string]int64, error) { |
| 1254 | blobs, err := r.NewBlobs(old, new) |
| 1255 | if err != nil { |
| 1256 | return nil, err |
| 1257 | } |
| 1258 | out := make(map[string]int64, len(blobs)) |
| 1259 | for sha, b := range blobs { |
| 1260 | out[sha] = b.Size |
| 1261 | } |
| 1262 | return out, nil |
| 1263 | } |
| 1264 | |
| 1265 | // Blob is one object a push introduces, with the path git reported for it. |
| 1266 | // An object reachable from two paths is reported once, under one of them: the |
| 1267 | // path is for the message, not for authorization or naming anything durable. |
| 1268 | type Blob struct { |
| 1269 | SHA string |
| 1270 | Path string |
| 1271 | Size int64 |
| 1272 | } |
| 1273 | |
| 1274 | // NewBlobs lists the blobs a push adds, with sizes. rev-list --objects is the |
| 1275 | // only git command here that reports path and object id together, so it is the |
| 1276 | // one used for both the size check and LFS pointer scanning. |
| 1277 | func (r *Repo) NewBlobs(old, new string) (map[string]Blob, error) { |
| 1278 | args := []string{"rev-list", "--objects", new} |
| 1279 | // A brand-new ref reports the zero SHA as its old value; passing that to |
| 1280 | // --not would make git error out and silently skip the check. |
| 1281 | if old != "" && old != ZeroSHA { |
| 1282 | args = append(args, "--not", old) |
| 1283 | } |
| 1284 | out, errs, err := r.run(nil, args...) |
| 1285 | if err != nil { |
| 1286 | return nil, fmt.Errorf("rev-list: %w: %s", err, errs) |
| 1287 | } |
| 1288 | paths := map[string]string{} |
| 1289 | var shas []string |
| 1290 | for _, line := range strings.Split(strings.TrimSpace(string(out)), "\n") { |
| 1291 | if line == "" { |
| 1292 | continue |
| 1293 | } |
| 1294 | sha, path, _ := strings.Cut(line, " ") |
| 1295 | shas = append(shas, sha) |
| 1296 | paths[sha] = path |
| 1297 | } |
| 1298 | if len(shas) == 0 { |
| 1299 | return map[string]Blob{}, nil |
| 1300 | } |
| 1301 | sizes, err := r.blobSizes(shas) |
| 1302 | if err != nil { |
| 1303 | return nil, err |
| 1304 | } |
| 1305 | blobs := make(map[string]Blob, len(sizes)) |
| 1306 | for sha, n := range sizes { |
| 1307 | blobs[sha] = Blob{SHA: sha, Path: paths[sha], Size: n} |
| 1308 | } |
| 1309 | return blobs, nil |
| 1310 | } |
| 1311 | |
| 1312 | // TreeBlobs lists every blob in a commit's tree with its size. This is what the |
| 1313 | // LFS check scans: what a ref serves after the push, not every object the push |
| 1314 | // happens to carry. A commit that deletes a broken pointer must stay pushable, |
| 1315 | // or there is no way out of the mistake. |
| 1316 | func (r *Repo) TreeBlobs(treeish string) (map[string]Blob, error) { |
| 1317 | entries, err := r.LsTreeR(treeish) |
| 1318 | if err != nil { |
| 1319 | return nil, err |
| 1320 | } |
| 1321 | paths := map[string]string{} |
| 1322 | var shas []string |
| 1323 | for _, e := range entries { |
| 1324 | if e.Type != "blob" { |
| 1325 | continue |
| 1326 | } |
| 1327 | paths[e.SHA] = e.Name |
| 1328 | shas = append(shas, e.SHA) |
| 1329 | } |
| 1330 | sizes, err := r.blobSizes(shas) |
| 1331 | if err != nil { |
| 1332 | return nil, err |
| 1333 | } |
| 1334 | blobs := make(map[string]Blob, len(sizes)) |
| 1335 | for sha, n := range sizes { |
| 1336 | blobs[sha] = Blob{SHA: sha, Path: paths[sha], Size: n} |
| 1337 | } |
| 1338 | return blobs, nil |
| 1339 | } |
| 1340 | |
| 1341 | // Diff returns the unified diff git prints between two revisions. A range may |
| 1342 | // be written with git's own syntax ("main...feature"), so the caller decides |
| 1343 | // between two-dot and merge-base semantics. |
| 1344 | // |
| 1345 | // The patch is complete. A view that renders one file asks for that file |
| 1346 | // (DiffPaths) and a view that lists files asks for their names and counts |
| 1347 | // (DiffInventory), so no caller has to cut a patch to stay small: the size of a |
| 1348 | // page follows the file it shows, not the size of the change. |
| 1349 | func (r *Repo) Diff(rangeOrFrom, to string) (string, error) { |
| 1350 | args := append([]string{"diff", "--no-color", "--no-ext-diff", "--find-renames", "--unified=3"}, diffRange(rangeOrFrom, to)...) |
| 1351 | out, errs, err := r.run(nil, args...) |
| 1352 | if err != nil { |
| 1353 | return "", fmt.Errorf("diff: %w: %s", err, errs) |
| 1354 | } |
| 1355 | return string(out), nil |
| 1356 | } |
| 1357 | |
| 1358 | // DiffPaths is Diff limited to the paths given, which is what a per-file view |
| 1359 | // reads. A rename needs both of its paths named: filtered to the destination |
| 1360 | // alone, git reports what is left as a new file and the rename is lost. |
| 1361 | func (r *Repo) DiffPaths(rangeOrFrom, to string, paths []string) (string, error) { |
| 1362 | args := append([]string{"diff", "--no-color", "--no-ext-diff", "--find-renames", "--unified=3"}, diffRange(rangeOrFrom, to)...) |
| 1363 | args = append(append(args, "--"), paths...) |
| 1364 | out, errs, err := r.run(nil, args...) |
| 1365 | if err != nil { |
| 1366 | return "", fmt.Errorf("diff %s: %w: %s", strings.Join(paths, " "), err, errs) |
| 1367 | } |
| 1368 | return string(out), nil |
| 1369 | } |
| 1370 | |
| 1371 | // DiffInventory lists what a change touches without generating its patch: the |
| 1372 | // status, both paths, and the line counts of every file. It is what a changed- |
| 1373 | // file tree, a landing summary and a file count read, so a change touching a |
| 1374 | // thousand files costs one walk of the trees rather than a patch large enough to |
| 1375 | // hold every hunk — and nothing is cut, so the inventory is the whole change. |
| 1376 | // |
| 1377 | // `--raw --numstat` prints the raw records first and the numstat records second, |
| 1378 | // both NUL-separated because a path may hold a tab, a quote or a non-ASCII byte. |
| 1379 | // The two sections are printed in the same order, which is what pairs a status |
| 1380 | // with its counts. |
| 1381 | func (r *Repo) DiffInventory(rangeOrFrom, to string) ([]PathChange, error) { |
| 1382 | args := append([]string{"diff", "--raw", "--numstat", "--find-renames", "-z"}, diffRange(rangeOrFrom, to)...) |
| 1383 | out, errs, err := r.run(nil, args...) |
| 1384 | if err != nil { |
| 1385 | return nil, fmt.Errorf("diff --raw --numstat: %w: %s", err, errs) |
| 1386 | } |
| 1387 | changes, err := parseInventory(string(out)) |
| 1388 | if err != nil { |
| 1389 | return nil, err |
| 1390 | } |
| 1391 | return changes, nil |
| 1392 | } |
| 1393 | |
| 1394 | // ShowPatchPath is a commit's patch against its parent (or the empty tree for a |
| 1395 | // root commit), limited to the paths given. `git show` gets the root case right, |
| 1396 | // which is why it is used here instead of assembling a range by hand. |
| 1397 | // |
| 1398 | // It is what a per-file commit view reads, so opening a broad commit costs the |
| 1399 | // hunks of one file rather than the whole diff. |
| 1400 | func (r *Repo) ShowPatchPath(rev string, paths []string) (string, error) { |
| 1401 | args := []string{"show", "--format=", "--patch", "--no-color", "--no-ext-diff", "--find-renames", "--unified=3", rev, "--"} |
| 1402 | args = append(args, paths...) |
| 1403 | out, errs, err := r.run(nil, args...) |
| 1404 | if err != nil { |
| 1405 | return "", fmt.Errorf("show %s (%s): %w: %s", rev, strings.Join(paths, " "), err, errs) |
| 1406 | } |
| 1407 | return string(out), nil |
| 1408 | } |
| 1409 | |
| 1410 | // diffRange returns the revision arguments a diff call takes. One range string |
| 1411 | // is git's own "a...b" syntax; two revisions are the endpoints git diffs when a |
| 1412 | // landing record names the tip it landed on rather than a range. |
| 1413 | func diffRange(rangeOrFrom, to string) []string { |
| 1414 | if to == "" { |
| 1415 | return []string{rangeOrFrom} |
| 1416 | } |
| 1417 | return []string{rangeOrFrom, to} |
| 1418 | } |
| 1419 | |
| 1420 | // parseInventory pairs the two sections `git diff --raw --numstat -z` prints. |
| 1421 | func parseInventory(out string) ([]PathChange, error) { |
| 1422 | changes, stats, err := parseInventorySections(strings.Split(out, "\x00")) |
| 1423 | if err != nil { |
| 1424 | return nil, err |
| 1425 | } |
| 1426 | if len(stats) != len(changes) { |
| 1427 | return nil, fmt.Errorf("diff --raw --numstat: %d files but %d count records", len(changes), len(stats)) |
| 1428 | } |
| 1429 | for i := range changes { |
| 1430 | // The sections are printed in the same order; checking that they agree |
| 1431 | // is what keeps a count from landing on the wrong file. |
| 1432 | if pathLabel(changes[i]) != pathLabel(stats[i]) { |
| 1433 | return nil, fmt.Errorf("diff --raw --numstat: %q and %q disagree", pathLabel(changes[i]), pathLabel(stats[i])) |
| 1434 | } |
| 1435 | changes[i].Added, changes[i].Removed = stats[i].Added, stats[i].Removed |
| 1436 | changes[i].Binary = stats[i].Binary |
| 1437 | } |
| 1438 | return changes, nil |
| 1439 | } |
| 1440 | |
| 1441 | // parseInventorySections reads the raw records, then returns them beside the |
| 1442 | // numstat records that follow. |
| 1443 | func parseInventorySections(fields []string) ([]PathChange, []PathChange, error) { |
| 1444 | var changes []PathChange |
| 1445 | // A raw record is ":<oldmode> <newmode> <oldsha> <newsha> <status>" and then |
| 1446 | // one path, or two for a rename or a copy. The status carries a similarity |
| 1447 | // score for those (R073), so only its first byte names the change. |
| 1448 | for len(fields) > 0 && strings.HasPrefix(fields[0], ":") { |
| 1449 | record := fields[0] |
| 1450 | fields = fields[1:] |
| 1451 | status := record[strings.LastIndexByte(record, ' ')+1:] |
| 1452 | if status == "" { |
| 1453 | return nil, nil, fmt.Errorf("diff --raw: %q has no status", record) |
| 1454 | } |
| 1455 | change := PathChange{Status: status[:1]} |
| 1456 | if change.Status == "R" || change.Status == "C" { |
| 1457 | if len(fields) < 2 { |
| 1458 | return nil, nil, fmt.Errorf("diff --raw: %q is missing a path", record) |
| 1459 | } |
| 1460 | change.Old, change.New = fields[0], fields[1] |
| 1461 | fields = fields[2:] |
| 1462 | } else { |
| 1463 | if len(fields) < 1 { |
| 1464 | return nil, nil, fmt.Errorf("diff --raw: %q is missing a path", record) |
| 1465 | } |
| 1466 | change.Old, change.New = fields[0], fields[0] |
| 1467 | fields = fields[1:] |
| 1468 | switch change.Status { |
| 1469 | case "A": |
| 1470 | change.Old = "" |
| 1471 | case "D": |
| 1472 | change.New = "" |
| 1473 | } |
| 1474 | } |
| 1475 | changes = append(changes, change) |
| 1476 | } |
| 1477 | |
| 1478 | var stats []PathChange |
| 1479 | // A numstat record is "<added>\t<removed>\t<path>". Binary files have no |
| 1480 | // lines, and a rename leaves the path empty and prints its two paths as the |
| 1481 | // two records that follow the counts. |
| 1482 | for len(fields) > 0 { |
| 1483 | record := fields[0] |
| 1484 | fields = fields[1:] |
| 1485 | if record == "" { |
| 1486 | continue |
| 1487 | } |
| 1488 | added, rest, ok := strings.Cut(record, "\t") |
| 1489 | if !ok { |
| 1490 | return nil, nil, fmt.Errorf("diff --numstat: %q is not counts and a path", record) |
| 1491 | } |
| 1492 | removed, path, ok := strings.Cut(rest, "\t") |
| 1493 | if !ok { |
| 1494 | return nil, nil, fmt.Errorf("diff --numstat: %q is not counts and a path", record) |
| 1495 | } |
| 1496 | stat := PathChange{} |
| 1497 | if added == "-" || removed == "-" { |
| 1498 | stat.Binary = true |
| 1499 | } else { |
| 1500 | var err error |
| 1501 | if stat.Added, err = strconv.Atoi(added); err != nil { |
| 1502 | return nil, nil, fmt.Errorf("diff --numstat: %q has no added count", record) |
| 1503 | } |
| 1504 | if stat.Removed, err = strconv.Atoi(removed); err != nil { |
| 1505 | return nil, nil, fmt.Errorf("diff --numstat: %q has no removed count", record) |
| 1506 | } |
| 1507 | } |
| 1508 | if path == "" { |
| 1509 | if len(fields) < 2 { |
| 1510 | return nil, nil, fmt.Errorf("diff --numstat: %q names no paths", record) |
| 1511 | } |
| 1512 | stat.Old, stat.New = fields[0], fields[1] |
| 1513 | fields = fields[2:] |
| 1514 | } else { |
| 1515 | stat.New = path |
| 1516 | } |
| 1517 | stats = append(stats, stat) |
| 1518 | } |
| 1519 | return changes, stats, nil |
| 1520 | } |
| 1521 | |
| 1522 | // pathLabel is the path a change is named by: the new one, or the old one for a |
| 1523 | // deletion. |
| 1524 | func pathLabel(c PathChange) string { |
| 1525 | if c.New != "" { |
| 1526 | return c.New |
| 1527 | } |
| 1528 | return c.Old |
| 1529 | } |
| 1530 | |
| 1531 | // PathChange is one path a commit touched, as git's name-status reports it. Old |
| 1532 | // is the path before the commit and New is the path after it, so Old is empty for |
| 1533 | // a file the commit added and New is empty for a deletion. They differ only for a |
| 1534 | // rename or a copy, and the patch for such a path is complete only when git is |
| 1535 | // told both. |
| 1536 | // |
| 1537 | // Status is git's single letter: A added, M modified, D deleted, R renamed, |
| 1538 | // C copied, T type changed. |
| 1539 | type PathChange struct { |
| 1540 | Status string |
| 1541 | Old string |
| 1542 | New string |
| 1543 | // Added and Removed are the line counts of the change to this path, and |
| 1544 | // Binary says there are no lines to count. They are zero for a listing that |
| 1545 | // did not ask for them (ShowNameStatus). |
| 1546 | Added int |
| 1547 | Removed int |
| 1548 | Binary bool |
| 1549 | } |
| 1550 | |
| 1551 | // ShowNameStatus lists the paths one commit touched without generating its |
| 1552 | // patch. It is the cheap index a per-file viewer renders first, so the cost of |
| 1553 | // opening a commit grows with the number of files it touched, not with the size |
| 1554 | // of their diffs. |
| 1555 | // |
| 1556 | // The output is read NUL-separated because git otherwise quotes a path holding a |
| 1557 | // quote, a tab or a non-ASCII byte, and these paths are what the view builds |
| 1558 | // links and pathspecs from. |
| 1559 | func (r *Repo) ShowNameStatus(rev string) ([]PathChange, error) { |
| 1560 | out, errs, err := r.run(nil, "show", "--format=", "--name-status", "--no-color", "--find-renames", "-z", rev) |
| 1561 | if err != nil { |
| 1562 | return nil, fmt.Errorf("show --name-status %s: %w: %s", rev, err, errs) |
| 1563 | } |
| 1564 | fields := strings.Split(string(out), "\x00") |
| 1565 | changes := make([]PathChange, 0, len(fields)/2) |
| 1566 | for i := 0; i < len(fields); { |
| 1567 | status := strings.TrimSpace(fields[i]) |
| 1568 | i++ |
| 1569 | if status == "" { |
| 1570 | continue |
| 1571 | } |
| 1572 | if i >= len(fields) { |
| 1573 | break |
| 1574 | } |
| 1575 | first := fields[i] |
| 1576 | i++ |
| 1577 | c := PathChange{Status: status[:1], Old: first, New: first} |
| 1578 | if c.Status == "R" || c.Status == "C" { |
| 1579 | if i >= len(fields) { |
| 1580 | break |
| 1581 | } |
| 1582 | c.New = fields[i] |
| 1583 | i++ |
| 1584 | } |
| 1585 | switch c.Status { |
| 1586 | case "A": |
| 1587 | c.Old = "" |
| 1588 | case "D": |
| 1589 | c.New = "" |
| 1590 | } |
| 1591 | changes = append(changes, c) |
| 1592 | } |
| 1593 | return changes, nil |
| 1594 | } |
| 1595 | |
| 1596 | // CommitBody is a commit's full message, for the commit page. |
| 1597 | func (r *Repo) CommitBody(rev string) (string, error) { |
| 1598 | out, errs, err := r.run(nil, "show", "-s", "--format=%B", rev) |
| 1599 | if err != nil { |
| 1600 | return "", fmt.Errorf("show -s %s: %w: %s", rev, err, errs) |
| 1601 | } |
| 1602 | return strings.TrimSpace(string(out)), nil |
| 1603 | } |
| 1604 | |
| 1605 | // AllBlobs lists every blob reachable from any ref, with its size. It is the |
| 1606 | // marking step of garbage collection, so it walks history rather than tips: an |
| 1607 | // old commit still serves its files. |
| 1608 | func (r *Repo) AllBlobs() (map[string]int64, error) { |
| 1609 | out, errs, err := r.run(nil, "rev-list", "--objects", "--all") |
| 1610 | if err != nil { |
| 1611 | return nil, fmt.Errorf("rev-list --all: %w: %s", err, errs) |
| 1612 | } |
| 1613 | var shas []string |
| 1614 | for _, line := range strings.Split(strings.TrimSpace(string(out)), "\n") { |
| 1615 | if line == "" { |
| 1616 | continue |
| 1617 | } |
| 1618 | sha, _, _ := strings.Cut(line, " ") |
| 1619 | shas = append(shas, sha) |
| 1620 | } |
| 1621 | return r.blobSizes(shas) |
| 1622 | } |
| 1623 | |
| 1624 | func (r *Repo) blobSizes(shas []string) (map[string]int64, error) { |
| 1625 | if len(shas) == 0 { |
| 1626 | return map[string]int64{}, nil |
| 1627 | } |
| 1628 | var in bytes.Buffer |
| 1629 | for _, s := range shas { |
| 1630 | in.WriteString(s + "\n") |
| 1631 | } |
| 1632 | bo, errs, err := r.run(in.Bytes(), "cat-file", "--batch-check=%(objectname) %(objecttype) %(objectsize)") |
| 1633 | if err != nil { |
| 1634 | return nil, fmt.Errorf("cat-file --batch-check: %w: %s", err, errs) |
| 1635 | } |
| 1636 | sizes := map[string]int64{} |
| 1637 | for _, line := range strings.Split(strings.TrimSpace(string(bo)), "\n") { |
| 1638 | f := strings.Fields(line) |
| 1639 | if len(f) != 3 || f[1] != "blob" { |
| 1640 | continue |
| 1641 | } |
| 1642 | n, err := strconv.ParseInt(f[2], 10, 64) |
| 1643 | if err != nil { |
| 1644 | continue |
| 1645 | } |
| 1646 | sizes[f[0]] = n |
| 1647 | } |
| 1648 | return sizes, nil |
| 1649 | } |
| 1650 | |
| 1651 | // BlobPrefix reads at most n bytes from a blob. Sniffing for a pointer or for |
| 1652 | // binary content should not cost a 100 MiB read of a blob that is about to be |
| 1653 | // accepted anyway. |
| 1654 | func (r *Repo) BlobPrefix(sha string, n int) ([]byte, error) { |
| 1655 | out, errs, err := r.runHead(n, "cat-file", "blob", sha) |
| 1656 | if err != nil { |
| 1657 | return nil, fmt.Errorf("cat-file blob %s: %w: %s", sha, err, errs) |
| 1658 | } |
| 1659 | return out, nil |
| 1660 | } |
| 1661 |