@hominis/fireforge 0.55.0 → 0.60.0

package/dist/src/core/git-diff.js

@@ -1,539 +0,0 @@
-// SPDX-License-Identifier: EUPL-1.2
-import { stat } from 'node:fs/promises';
-import { join } from 'node:path';
-import { GitError } from '../errors/git.js';
-import { toError } from '../utils/errors.js';
-import { pathExists } from '../utils/fs.js';
-import { verbose } from '../utils/logger.js';
-import { chunkPathspecs, git } from './git-base.js';
-import { readNewFileContent, withForcedBinaryAttribute } from './git-diff-new-file.js';
-import { fileExistsInHead, hashObjectBatch, listTrackedInHead } from './git-file-ops.js';
-import { mintDisposableGitIndex } from './git-readonly-index.js';
-import { getUntrackedFiles, getUntrackedFilesInDir } from './git-status.js';
-/**
- * Gets the diff for a specific file.
- *
- * `--binary` is load-bearing: without it git degrades a binary file's section
- * to the informational `Binary files a/x and b/x differ`, which carries none
- * of the bytes and cannot be replayed by `git apply`. It only expands the
- * binary section's index line to full hashes (text sections keep their
- * abbreviated form), so text output stays byte-identical.
- * @param repoDir - Repository directory
- * @param filePath - Path to the file (relative to repo)
- * @returns Diff content
- */
-export async function getFileDiff(repoDir, filePath) {
-    return git(['diff', '--binary', 'HEAD', '--', filePath], repoDir);
-}
-/**
- * Abbreviates a full git blob hash to the 10-character form used in the
- * synthesized `index 0000000000..<hash>` line, falling back to the all-zero
- * placeholder when no usable hash is available. Mirrors the original per-file
- * truncation exactly.
- * @param fullHash - Full blob hash, or undefined when hashing failed
- * @returns 10-character abbreviated hash or the zero placeholder
- */
-function abbreviateBlobHash(fullHash) {
-    if (fullHash !== undefined && fullHash.length >= 10) {
-        return fullHash.slice(0, 10);
-    }
-    return '0000000000';
-}
-/**
- * Pure formatter for a new (untracked) file's unified diff. Shared by the
- * batched cold-run path in {@link getDiffForFilesAgainstHead} and the
- * standalone {@link generateNewFileDiff}. The only thing that differs is
- * where `blobHash` comes from (a single batched `git hash-object` vs a
- * per-file one), never the formatting. Preserves the empty-file form, the
- * trailing-newline handling, and the "No newline at end of file" marker
- * byte-for-byte.
- *
- * @param filePath - Path to the file (relative to repo)
- * @param content - File content
- * @param blobHash - Abbreviated blob hash for the index line
- * @returns Diff content in unified diff format
- */
-function buildNewFileDiffBody(filePath, content, blobHash) {
-    // Handle empty files
-    if (content.length === 0) {
-        return [
-            `diff --git a/${filePath} b/${filePath}`,
-            'new file mode 100644',
-            `index 0000000000..${blobHash}`,
-            '--- /dev/null',
-            `+++ b/${filePath}`,
-            '',
-        ].join('\n');
-    }
-    const lines = content.split('\n');
-    // Handle files that don't end with newline
-    const hasTrailingNewline = content.endsWith('\n');
-    const lineCount = hasTrailingNewline ? lines.length - 1 : lines.length;
-    // Build the unified diff format for a new file
-    const diffLines = [
-        `diff --git a/${filePath} b/${filePath}`,
-        'new file mode 100644',
-        `index 0000000000..${blobHash}`,
-        '--- /dev/null',
-        `+++ b/${filePath}`,
-        `@@ -0,0 +1,${lineCount} @@`,
-    ];
-    // Add each line with a + prefix
-    for (let i = 0; i < lineCount; i++) {
-        diffLines.push(`+${lines[i]}`);
-    }
-    // Add "No newline at end of file" marker if needed
-    if (!hasTrailingNewline && lineCount > 0) {
-        diffLines.push('\');
-    }
-    return diffLines.join('\n') + '\n';
-}
-/**
- * Defensive check shared by the new-file readers: a directory here means a
- * caller bypassed the expansion layers and handed the leaf reader a path it
- * cannot read. Surfaces an actionable message naming the offending path
- * rather than the raw `EISDIR` the read would throw. This is a recurring bug
- * class, hence the guard at every leaf.
- * @param repoDir - Repository directory
- * @param filePath - Path to the file (relative to repo)
- */
-async function rejectDirectory(repoDir, filePath) {
-    const fileStat = await stat(join(repoDir, filePath));
-    if (fileStat.isDirectory()) {
-        throw new GitError(`expected a file but found a directory at '${filePath}' — caller must expand directory entries before diffing`, `hash-object ${filePath}`);
-    }
-}
-/**
- * Generates a unified diff for a new (untracked) file. A file that is binary
- * or not valid UTF-8 (see {@link readNewFileContent}) is routed through
- * {@link generateBinaryFilePatch} instead, because a text body cannot carry
- * its bytes faithfully.
- * @param repoDir - Repository directory
- * @param filePath - Path to the file (relative to repo)
- * @returns Diff content in unified diff format
- */
-export async function generateNewFileDiff(repoDir, filePath) {
-    const fullPath = join(repoDir, filePath);
-    await rejectDirectory(repoDir, filePath);
-    const file = await readNewFileContent(repoDir, filePath);
-    if (file.binary) {
-        return generateBinaryFilePatch(repoDir, filePath);
-    }
-    // Compute the abbreviated git blob hash for the index line
-    let blobHash = '0000000000';
-    try {
-        const fullHash = (await git(['hash-object', fullPath], repoDir)).trim();
-        blobHash = abbreviateBlobHash(fullHash);
-    }
-    catch (error) {
-        verbose(`git hash-object failed for ${filePath}; falling back to zero blob hash: ${toError(error).message}`);
-    }
-    return buildNewFileDiffBody(filePath, file.content, blobHash);
-}
-/**
- * Generates a patch for a file.
- * If the file is tracked in HEAD, it generates a standard contextual diff.
- * If the file is untracked (new), it generates a "new file" format patch (snapshot).
- * This ensures standard 3-way mergeable context diffs for existing Mozilla files.
- * @param repoDir - Repository directory
- * @param filePath - Path to the file (relative to repo)
- * @returns Diff content in unified diff format
- */
-export async function generateFullFilePatch(repoDir, filePath) {
-    // If file exists in HEAD, use standard git diff HEAD -- <file>
-    // This generates a contextual diff that is safer for rebasing
-    if (await fileExistsInHead(repoDir, filePath)) {
-        return getFileDiff(repoDir, filePath);
-    }
-    // If file is new/untracked, use the full-file "new file" format
-    return generateNewFileDiff(repoDir, filePath);
-}
-/**
- * Gets the diff for all modified files, including untracked (new) files.
- *
- * Tracked binary files round-trip through `git diff --binary`. Untracked ones
- * go through {@link generateBinaryFilePatch} below. Both arms must stay
- * binary-aware. See {@link getFileDiff} for why.
- * @param repoDir - Repository directory
- * @returns Diff content
- */
-export async function getAllDiff(repoDir) {
-    // Get diff for tracked files
-    const trackedDiff = await git(['diff', '--binary', 'HEAD'], repoDir);
-    // Get untracked files (properly expanded, not directories)
-    const untrackedFiles = await getUntrackedFiles(repoDir);
-    // Classify first. `readNewFileContent` reads the file directly and spawns
-    // nothing, so the only per-file process in this loop would be the
-    // `git hash-object` inside `generateNewFileDiff`: one spawn per untracked
-    // text file, uncapped, against a Firefox tree where "untracked" can mean a
-    // very large number. Hence classify → one batched hash → render. The
-    // classification carries each text file's content so it is read once.
-    //
-    // Everything else here is deliberately untouched: `git diff HEAD` (with
-    // renames, unlike the `--no-renames` scoped sibling), the `ls-files`
-    // ordering, tracked-block-first emission, and the `'\n'` empty-result
-    // sentinel are all observable. `getAllDiff`'s output is SHA-256'd into
-    // every tree fingerprint (tree-store.ts) and written verbatim into `.patch`
-    // files (export-all.ts), so its bytes are a contract.
-    const classified = new Map();
-    for (const file of untrackedFiles) {
-        await rejectDirectory(repoDir, file);
-        classified.set(file, await readNewFileContent(repoDir, file));
-    }
-    const textHashes = await hashObjectBatch(repoDir, untrackedFiles
-        .filter((file) => classified.get(file)?.binary === false)
-        .map((file) => join(repoDir, file)));
-    // Binary patches stage into the index and must stay serial. Text patches
-    // are pure formatting over the already-batched hashes.
-    const binaryDiffs = new Map();
-    for (const file of untrackedFiles) {
-        if (classified.get(file)?.binary === true) {
-            binaryDiffs.set(file, await generateBinaryFilePatch(repoDir, file));
-        }
-    }
-    const untrackedDiffs = [];
-    for (const file of untrackedFiles) {
-        const entry = classified.get(file);
-        if (entry === undefined || entry.binary) {
-            const binary = binaryDiffs.get(file);
-            if (binary !== undefined)
-                untrackedDiffs.push(binary);
-            continue;
-        }
-        const fullPath = join(repoDir, file);
-        untrackedDiffs.push(buildNewFileDiffBody(file, entry.content, abbreviateBlobHash(textHashes.get(fullPath))));
-    }
-    // Combine all diffs. Each already ends with \n, so concatenate directly
-    // to avoid inserting blank lines between diff sections.
-    const allDiffs = [trackedDiff, ...untrackedDiffs].filter((d) => d.trim().length > 0);
-    const combined = allDiffs.join('');
-    return combined.endsWith('\n') ? combined : combined + '\n';
-}
-/**
- * Splits a combined `git diff` body into one string per file section,
- * preserving exact bytes. A section runs from a column-0 `diff --git ` line up
- * to (but not including) the next such line. Boundary detection is anchored to
- * column 0 because every diff *body* line is prefixed by a space, `+`, `-`,
- * `\`, or `@`, so a context or added line that merely contains the text
- * `diff --git` can never be mistaken for a header. Paths are not parsed out
- * of the header here on purpose. See {@link buildTrackedSections}.
- * @param combined - Combined `git diff` stdout
- * @returns File sections in git's emission order
- */
-function splitDiffSections(combined) {
-    const marker = 'diff --git ';
-    const sections = [];
-    let start = -1;
-    for (let i = 0; i < combined.length; i++) {
-        if ((i === 0 || combined[i - 1] === '\n') && combined.startsWith(marker, i)) {
-            if (start !== -1)
-                sections.push(combined.slice(start, i));
-            start = i;
-        }
-    }
-    if (start !== -1)
-        sections.push(combined.slice(start, combined.length));
-    return sections;
-}
-/**
- * Runs one `git diff --binary --no-renames HEAD` over the tracked files
- * (chunked under ARG_MAX) and returns a `Map<path, section>` whose sections are
- * byte-identical to the per-file `git diff --binary HEAD -- <file>` they
- * replace.
- *
- * `--binary` is load-bearing: a tracked binary file would otherwise degrade to
- * the un-appliable `Binary files a/x and b/x differ` stub. It expands only the
- * binary section's index line to full hashes, so text sections are
- * unaffected. See {@link getFileDiff}.
- *
- * `--no-renames` is load-bearing: a multi-path diff under a user's
- * `diff.renames=true`/`=copies` could otherwise emit a single 2-path rename
- * section (`a/<old> b/<new>`) that a single-path `git diff HEAD -- <file>` can
- * never produce. `--no-renames` re-splits it into the same delete + add bytes
- * the per-file loop emitted.
- *
- * Sections are attributed to paths by position against a companion
- * `git diff --no-renames HEAD -z --name-only` (raw, unquoted, NUL-delimited
- * paths, emitted in the same order as the sections), never by parsing the
- * `diff --git` header, which is ambiguous or unparseable under `core.quotePath`
- * (non-ASCII paths are C-quoted), paths containing spaces, or
- * `diff.noprefix`/`diff.mnemonicPrefix`. If the section and name counts ever
- * disagree (an unmodeled config), that chunk falls back to per-file
- * {@link getFileDiff} so no file's diff is ever silently dropped.
- * @param repoDir - Repository directory
- * @param trackedFiles - Repo-relative files known to exist in HEAD
- * @returns Map from path to its exact diff section (changed files only)
- */
-async function buildTrackedSections(repoDir, trackedFiles) {
-    const sectionsByPath = new Map();
-    for (const chunk of chunkPathspecs(trackedFiles)) {
-        const combined = await git(['diff', '--binary', '--no-renames', 'HEAD', '--', ...chunk], repoDir);
-        const namesOutput = await git(['diff', '--no-renames', 'HEAD', '-z', '--name-only', '--', ...chunk], repoDir);
-        const names = namesOutput.split('\0').filter((name) => name.length > 0);
-        const sections = splitDiffSections(combined);
-        if (sections.length === names.length) {
-            for (let i = 0; i < names.length; i++) {
-                // Exact-path keys (raw git bytes, same encoding as the inputs), never
-                // a substring/startsWith match, so `foo.txt` cannot capture
-                // `foo.txt.bak`'s section.
-                sectionsByPath.set(names[i], sections[i]);
-            }
-            continue;
-        }
-        // Counts disagree, so recover the exact pre-batch per-file bytes for this
-        // chunk rather than risk dropping or mis-keying a section.
-        for (const file of chunk) {
-            const diff = await getFileDiff(repoDir, file);
-            if (diff.trim())
-                sectionsByPath.set(file, diff);
-        }
-    }
-    return sectionsByPath;
-}
-/**
- * Builds a combined diff against HEAD for the provided files without touching
- * the real git index. Tracked files use `git diff --binary HEAD` (binary
- * sections included). Untracked files use synthesized new-file diffs, or
- * {@link generateBinaryFilePatch} when the file is binary.
- *
- * Performance: the work is batched into a handful of `git` invocations (one
- * `ls-tree` to classify, one `diff` over all tracked files, one `hash-object`
- * over all new text files) rather than ~2 spawns per file, which dominates
- * the cold-run cost on a Firefox-sized checkout. Binary, directory, and
- * recursion paths stay per-file because they are rare and (for binary)
- * mutate the index.
- *
- * @param repoDir - Repository directory
- * @param files - File paths to diff (relative to repo root)
- * @returns Combined diff content
- */
-export async function getDiffForFilesAgainstHead(repoDir, files) {
-    // Expand any directory entries (paths ending with `/`) into their
-    // individual untracked files before diffing. `git status --porcelain=v1`
-    // reports collapsed untracked directories as `?? dir/`, and every caller
-    // that feeds the aggregate working-tree state into this function must not
-    // trigger an EISDIR when the diff pass reads `dir/` as if it were a file.
-    // The caller-side expansion in `lint.ts` and `export-all.ts` covers the
-    // common path. Guarding here makes a single bad call site harmless.
-    const expandedFiles = [];
-    for (const file of files) {
-        if (file.endsWith('/')) {
-            const inner = await getUntrackedFilesInDir(repoDir, file);
-            for (const entry of inner)
-                expandedFiles.push(entry);
-            continue;
-        }
-        expandedFiles.push(file);
-    }
-    const uniqueFiles = [...new Set(expandedFiles)].sort();
-    if (uniqueFiles.length === 0)
-        return '';
-    // Batch 1: classify tracked-vs-new for the whole set in one `ls-tree` pass,
-    // replacing one `fileExistsInHead` spawn per file.
-    const tracked = await listTrackedInHead(repoDir, uniqueFiles);
-    // Batch 2: one diff over every tracked file, split back to exact per-file
-    // sections keyed by path.
-    const trackedSections = await buildTrackedSections(repoDir, uniqueFiles.filter((file) => tracked.has(file)));
-    // Classify the non-tracked files. Directory and binary entries keep their
-    // per-file handling (rare, and binary patches mutate the index so they must
-    // stay serial). Plain new text files are collected for a single batched
-    // `git hash-object`. Results land in `sectionByFile`, keyed by path, to be
-    // emitted in sorted order below.
-    const sectionByFile = new Map();
-    const newTextFiles = [];
-    for (const file of uniqueFiles) {
-        if (tracked.has(file))
-            continue;
-        const fullPath = join(repoDir, file);
-        if (!(await pathExists(fullPath))) {
-            continue;
-        }
-        // Second defence against EISDIR: a non-HEAD path that exists on disk is
-        // usually a new file, but can also be a directory that arrived without
-        // the trailing slash `expandUntrackedDirectoryEntries` would have
-        // produced (caller stripped it, submodule entry, tracked-file-replaced-
-        // by-dir). Expand it via the same helper used by the slash branch and
-        // recurse so each contained file is diffed individually. Fail loud when
-        // the directory has no readable content rather than silently skipping it.
-        const fileStat = await stat(fullPath);
-        if (fileStat.isDirectory()) {
-            const innerFiles = await getUntrackedFilesInDir(repoDir, file);
-            if (innerFiles.length === 0) {
-                throw new GitError(`'${file}' is a directory with no untracked content (submodule or gitignored?) — cannot diff as a file`, `ls-files --others -- ${file}`);
-            }
-            const innerDiff = await getDiffForFilesAgainstHead(repoDir, innerFiles);
-            if (innerDiff.trim())
-                sectionByFile.set(file, innerDiff);
-            continue;
-        }
-        // Binary (by git's NUL heuristic or because the bytes are not valid
-        // UTF-8) takes the byte-faithful binary-patch arm.
-        const newFile = await readNewFileContent(repoDir, file);
-        if (newFile.binary) {
-            const diff = await generateBinaryFilePatch(repoDir, file);
-            if (diff.trim())
-                sectionByFile.set(file, diff);
-            continue;
-        }
-        newTextFiles.push({ file, content: newFile.content });
-    }
-    // Batch 3: blob hashes for every new text file in one `git hash-object`.
-    // A miss (rare: a path became unreadable after the stat above) falls back to
-    // the zero hash with the same verbose log the per-file path emitted.
-    const blobHashes = await hashObjectBatch(repoDir, newTextFiles.map(({ file }) => join(repoDir, file)));
-    for (const { file, content } of newTextFiles) {
-        const fullHash = blobHashes.get(join(repoDir, file));
-        if (fullHash === undefined) {
-            verbose(`git hash-object failed for ${file}; falling back to zero blob hash`);
-        }
-        const body = buildNewFileDiffBody(file, content, abbreviateBlobHash(fullHash));
-        if (body.trim())
-            sectionByFile.set(file, body);
-    }
-    // Reassemble in the sorted `uniqueFiles` order, not git's section order,
-    // which uses git's own collation and diverges from JS `.sort()` for
-    // non-ASCII paths. Driving emission off `uniqueFiles` (as the previous
-    // per-file loop did) keeps the combined output byte-identical. Do not change
-    // this to push sections in git's emission order.
-    const diffs = [];
-    for (const file of uniqueFiles) {
-        const section = trackedSections.get(file) ?? sectionByFile.get(file);
-        if (section && section.trim())
-            diffs.push(section);
-    }
-    if (diffs.length === 0) {
-        return '';
-    }
-    // Each diff from git already ends with \n. Concatenate directly to
-    // preserve context lines (including trailing whitespace-only context)
-    // and avoid inserting blank lines between diff sections.
-    const combined = diffs.join('');
-    return combined.endsWith('\n') ? combined : combined + '\n';
-}
-/**
- * Generates a combined diff for staged files against HEAD.
- * @param repoDir - Repository directory
- * @param files - File paths to diff (relative to repo)
- * @returns Diff content for the staged files
- */
-export async function getStagedDiffForFiles(repoDir, files) {
-    return git(['diff', '--binary', '--cached', 'HEAD', '--', ...files], repoDir);
-}
-/**
- * Serializes the index-mutating untracked-binary path. The bounded per-patch
- * lint pool can call {@link getDiffForFilesAgainstHead} (and thus this) for
- * several patches at once. Two concurrent `git add --intent-to-add` / `git
- * reset` sequences would collide on `.git/index.lock` (a hard failure) or
- * interleave one file's stage with another's unstage. This process-level
- * promise chain runs the staging sequences one at a time. Read-only callers
- * (`git diff --binary HEAD`) do not need it. Binary patches are rare, so the
- * serialization cost is negligible.
- */
-let binaryStagingLock = Promise.resolve();
-function runWithBinaryStagingLock(task) {
-    const result = binaryStagingLock.then(task, task);
-    binaryStagingLock = result.then(() => undefined, () => undefined);
-    return result;
-}
-/**
- * Generates a GIT binary patch for a binary file.
- * For tracked files, uses `git diff --binary HEAD`.
- * For untracked files, temporarily stages with `--intent-to-add` to produce a diff.
- *
- * Every `git diff` here runs under {@link withForcedBinaryAttribute}: a
- * caller sends a file down this arm because a text body cannot carry its
- * bytes (NUL, or not valid UTF-8), and git's own NUL-only heuristic would
- * otherwise still render the non-UTF-8 case as text.
- * @param repoDir - Repository directory
- * @param filePath - File path (relative to repo root)
- * @returns The binary diff string, or empty string if no diff
- */
-export async function generateBinaryFilePatch(repoDir, filePath) {
-    return withForcedBinaryAttribute(filePath, (gitConfigArgs) => generateBinaryFilePatchWithConfig(repoDir, filePath, gitConfigArgs));
-}
-/**
- * The body of {@link generateBinaryFilePatch}, with the forced-binary config
- * arguments already resolved.
- * @param repoDir - Repository directory
- * @param filePath - File path (relative to repo root)
- * @param gitConfigArgs - `-c` arguments spliced ahead of every `diff`
- * @returns The binary diff string, or empty string if no diff
- */
-async function generateBinaryFilePatchWithConfig(repoDir, filePath, gitConfigArgs) {
-    // Try tracked file diff first (read-only, so no index lock needed)
-    const tracked = await git([...gitConfigArgs, 'diff', '--binary', 'HEAD', '--', filePath], repoDir);
-    if (tracked.trim())
-        return tracked;
-    // The diff for an untracked file has to come from an index entry, so one
-    // has to be written. Writing it to a private index is what keeps that
-    // write invisible: the primary `.git/index` never moves, so a concurrent
-    // `fireforge test` fingerprinting `engine/` with `git status` cannot
-    // observe the transient `--intent-to-add` entry and void its own verdict
-    // on our bookkeeping. That was a real failure mode: parallel gate lanes
-    // killed three runs with `FAIL reason=inconclusive` on branding PNGs
-    // flapping between `A` and `??`.
-    //
-    // Nothing is restored afterwards because nothing shared was changed, and
-    // no lock is taken because two concurrent callers get two indexes. Both
-    // the restore dance and the in-process lock survive below purely for the
-    // fallback, where the primary index really is the one being written.
-    const disposable = await mintDisposableGitIndex(repoDir);
-    if (disposable !== undefined) {
-        try {
-            await git(['add', '--intent-to-add', '--', filePath], repoDir, { env: disposable.env });
-            return await git([...gitConfigArgs, 'diff', '--binary', '--', filePath], repoDir, {
-                env: disposable.env,
-            });
-        }
-        finally {
-            await disposable.dispose();
-        }
-    }
-    // Fallback: no private index could be minted (a non-git directory, an
-    // unreadable index, no writable tmpdir). Degrading to the shared index is
-    // the pre-existing behaviour rather than a new risk, and refusing instead
-    // would fail an export over an optimisation. The stage/unstage pair
-    // mutates the index, so it must not interleave with another concurrent
-    // binary patch (see runWithBinaryStagingLock).
-    verbose(`Binary patch for ${filePath} is staging on the shared index (no private index).`);
-    return runWithBinaryStagingLock(async () => {
-        // Snapshot the path's exact index entry before touching it. A blanket
-        // `git reset HEAD -- <file>` cleanup restores the path to its HEAD state.
-        // For a path absent from HEAD that means *removing* whatever entry the
-        // index carried, silently discarding any staged state that arrived
-        // between the read-only diff above and this staging block (a concurrent
-        // writer, or a staged-add whose worktree file was deleted). Restoring
-        // the captured entry is exact regardless of how the entry got there.
-        const priorEntry = await git(['ls-files', '--stage', '--', filePath], repoDir);
-        try {
-            await git(['add', '--intent-to-add', '--', filePath], repoDir);
-            return await git([...gitConfigArgs, 'diff', '--binary', '--', filePath], repoDir);
-        }
-        finally {
-            await restoreIndexEntry(repoDir, filePath, priorEntry);
-        }
-    });
-}
-/**
- * Restores a path's index entry to a previously captured `ls-files --stage`
- * snapshot after a temporary `--intent-to-add` staging. No prior entry means
- * the temporary entry is dropped via `reset HEAD` (the path's HEAD state is
- * "absent", and this branch only runs for files untracked in HEAD). A prior
- * stage-0 entry is re-pointed at its recorded mode and blob. Unmerged entries
- * (stages 1 to 3) cannot be rebuilt through `--cacheinfo`. That degenerate
- * state falls back to the reset-to-HEAD restore rather than corrupting the
- * conflict.
- * @param repoDir - Repository directory
- * @param filePath - File path (relative to repo root)
- * @param priorEntry - Raw `git ls-files --stage -- <file>` output from before staging
- */
-async function restoreIndexEntry(repoDir, filePath, priorEntry) {
-    const entryMatch = /^(\d{6}) ([0-9a-f]{4,64}) 0\t/.exec(priorEntry.trim());
-    const mode = entryMatch?.[1];
-    const oid = entryMatch?.[2];
-    if (mode === undefined || oid === undefined) {
-        await git(['reset', 'HEAD', '--', filePath], repoDir);
-        return;
-    }
-    await git(['update-index', '--add', '--cacheinfo', `${mode},${oid},${filePath}`], repoDir);
-}

package/dist/src/core/git-file-ops.d.ts

@@ -1,108 +0,0 @@
-import type { GitStatusEntry } from './git-base.js';
-/**
- * Restores a tracked path from HEAD, including staged changes.
- * @param repoDir - Repository directory
- * @param filePath - Path to the file (relative to repo)
- */
-export declare function restoreTrackedPath(repoDir: string, filePath: string): Promise<void>;
-/**
- * Discards a status entry according to its git state.
- * @param repoDir - Repository directory
- * @param entry - Parsed git status entry
- */
-export declare function discardStatusEntry(repoDir: string, entry: GitStatusEntry): Promise<void>;
-/**
- * Stages specific files in the repository.
- * @param repoDir - Repository directory
- * @param files - File paths to stage (relative to repo)
- */
-export declare function stageFiles(repoDir: string, files: string[]): Promise<void>;
-/**
- * Unstages specific files from the index.
- * @param repoDir - Repository directory
- * @param files - File paths to unstage (relative to repo)
- */
-export declare function unstageFiles(repoDir: string, files: string[]): Promise<void>;
-/**
- * Checks if a file exists in the HEAD commit.
- * @param repoDir - Repository directory
- * @param filePath - Path to the file (relative to repo)
- * @returns true if file exists in HEAD
- */
-export declare function fileExistsInHead(repoDir: string, filePath: string): Promise<boolean>;
-/**
- * Batched equivalent of {@link fileExistsInHead}: returns the subset of
- * `files` that are tracked in HEAD, using a single `git ls-tree` per ARG_MAX
- * chunk instead of one spawn per file. This is the cold-run hot path, and
- * a Firefox-sized checkout has hundreds of affected files.
- *
- * `-r` lists nested blobs by full repo-relative path. `--name-only -z` makes
- * the output a trivial NUL-split with no quoting to undo. Membership in the
- * returned Set is exactly `await fileExistsInHead(repoDir, file)` for any
- * non-directory `file`. Throws (via {@link git}) when HEAD itself is
- * unresolvable, matching the per-file helper's failure mode.
- *
- * @param repoDir - Repository directory
- * @param files - Repo-relative paths to classify
- * @returns The subset of `files` present in HEAD
- */
-export declare function listTrackedInHead(repoDir: string, files: string[]): Promise<Set<string>>;
-/**
- * Batched equivalent of the per-file `git hash-object` in
- * {@link import('./git-diff.js').generateNewFileDiff}: computes the git blob
- * hash for every path in one spawn per ARG_MAX chunk and returns a
- * `Map<fullPath, fullHash>`.
- *
- * Uses {@link import('../utils/process.js').exec} rather than {@link git}
- * (which throws on a non-zero exit) because `git hash-object f1 f2 …` is
- * all-or-nothing: it aborts at the first unreadable path and emits nothing
- * for the rest. To keep the per-file contract, where one bad path zeroes
- * only its own index line, a chunk that does not return exactly one hash
- * per input falls back to hashing that chunk's paths individually. A path
- * that is still unhashable is left out of the map. The caller applies the
- * `0000000000` zero-hash fallback for any miss. Hashing stays in git rather
- * than in-process so filters and `.gitattributes` are applied per path and
- * the result cannot diverge under `core.autocrlf`/`text` attributes.
- *
- * @param repoDir - Repository directory
- * @param fullPaths - Absolute file paths to hash
- * @returns Map from each input path to its full blob hash (misses omitted)
- */
-export declare function hashObjectBatch(repoDir: string, fullPaths: string[]): Promise<Map<string, string>>;
-/**
- * Gets the content of a file at a specific git ref (HEAD by default).
- * @param repoDir - Repository directory
- * @param filePath - Path to the file (relative to repo)
- * @param ref - Git ref to read from (commit hash, branch, tag). Defaults to HEAD.
- * @returns File content or null if file doesn't exist at that ref
- */
-export declare function getFileContentAtRef(repoDir: string, filePath: string, ref?: string): Promise<string | null>;
-/**
- * Batched equivalent of {@link getFileContentAtRef}: reads every path at
- * `ref` through one `git cat-file --batch` process instead of one
- * `git show` per file. A Firefox-sized ownership scan paid 1,493 of those,
- * about 18 ms each and almost all of it spawn overhead.
- *
- * The first request line is `ref` itself, so an unresolvable ref throws a
- * {@link GitError} exactly as the per-file helper does, rather than being
- * read as "every path missing". A path reported `missing` maps to `null`,
- * as the per-file helper returns for a path absent at that ref. Anything
- * the batch cannot answer as a plain blob (a path containing a newline,
- * which the line protocol cannot carry, or an object that is not a blob)
- * is fetched through {@link getFileContentAtRef}, so the result for every
- * path is what that helper returns.
- *
- * @param repoDir - Repository directory
- * @param filePaths - Paths relative to the repository root
- * @param ref - Git ref to read from. Defaults to HEAD.
- * @returns Content (or `null` when absent at `ref`) for every input path
- */
-export declare function getFilesContentAtRef(repoDir: string, filePaths: readonly string[], ref?: string): Promise<Map<string, string | null>>;
-/**
- * Checks if a file is binary by looking for NUL bytes in the first 8KB.
- * Uses the same heuristic as git.
- * @param repoDir - Repository directory
- * @param filePath - File path (relative to repo root)
- * @returns true if the file appears to be binary
- */
-export declare function isBinaryFile(repoDir: string, filePath: string): Promise<boolean>;