git-fs-s3 0.9.0 → 0.11.0

package/README.md

@@ -153,6 +153,19 @@ The returned fs also exposes:
 - `detectLooseObjects(gitdir)` — detect the loose-object hint up front (otherwise done on the first loose read).
 - `prefetchPacks(gitdir, { maxPacks? })` — warm every `.pack`/`.idx` in parallel before a sequential walk (skipped past `maxPacks * 2` entries).
 - `invalidate(pathPrefix)` — drop hints and registered gitdirs under the prefix and forward to the store's `invalidate` if it has one.
+- `getRange(filepath, start, end)` — present only when the backing store supports range reads (see `ObjectStore.getRange`); `getObjectSize` is its only caller.
+
+### `getObjectSize(repo, oid)`
+
+An object's size in bytes, **without reading it** — or `null` when the size
+isn't available (a loose object, an object no pack indexes, a store without
+range reads, or a delta prefix too short to yield its header). Built for one
+question: *is this object too large to render or diff?* A git object's length
+lives in its pack entry, so without this the only way to ask is to inflate the
+object — which is the cost a size limit exists to avoid. The answer is exact
+(whole objects from the entry header, deltified ones from the delta header that
+opens their deflated stream) and costs one `.idx` read plus one or two small
+range reads of the pack.
 
 ### Git layout
 
@@ -160,7 +173,7 @@ The returned fs also exposes:
 
 ### `S3ObjectStore({ client, bucket, prefix?, contentType? })`
 
-`ObjectStore` over `@aws-sdk/client-s3` (optional peer dependency, loaded only via the `/s3` subpath). Reuse one `S3Client` across stores so HTTP connections are pooled — a busy repo page can mean hundreds of object reads. `contentType(key)` optionally derives a `Content-Type` header for uploads.
+`ObjectStore` over `@aws-sdk/client-s3` (optional peer dependency, loaded only via the `/s3` subpath). Implements the optional `getRange(key, start, end)` (S3's `Range` header, inclusive at both ends), which `getObjectSize` uses. Reuse one `S3Client` across stores so HTTP connections are pooled — a busy repo page can mean hundreds of object reads. `contentType(key)` optionally derives a `Content-Type` header for uploads.
 
 ### `MemoryObjectStore()`
 
@@ -248,7 +261,7 @@ Every client-supplied ref name in a push is validated with `isSafeFullRefName` (
 - **Commits** — `commitFilesToBare` (write one or more files as a single commit against a branch, creating it if new), `deleteFileFromBare`, `authorNow`.
 - **Trees** — `upsertTree`/`deleteFromTree` (overlay blobs onto a tree, return the new root oid), `listTreeEntries`, `findTreeEntry`.
 - **History** — `getCommitLog`/`getCommitHistory` (memoized commit-chain walks; pages are slices of the deepest walk), `getFileContent`/`getTreeFromRef`, `getFileHistory` (per-file commit history), `getLastCommitsForTree` (the "last commit" column on a directory listing), `summarizeCommit`.
-- **Diff** — `getCommitDiff` (a commit against its first parent) and `getDiffBetweenRefs` (a pull request's three-dot diff, against the merge base); unified patches via the `diff` package (optional peer dependency).
+- **Diff** — `getCommitDiff` (a commit against its first parent) and `getDiffBetweenRefs` (a pull request's three-dot diff, against the merge base); unified patches via the `diff` package (optional peer dependency). A file over `MAX_DIFF_TEXT_BYTES` is reported as `kind: "too-large"` **without being read** — the size comes from its pack entry (see `getObjectSize`) — and a diff's blob reads are bounded (`MAX_CONCURRENT_READS`), so one wide commit can't fan out into a request per entry.
 - **Merge** — `analyzeMerge` (fast-forward/diverged pre-check — not a real content-conflict check, see its doc comment) and `fastForwardMerge`. Neither does a real (non-fast-forward) content merge — for that, see [`git-edge`](https://www.npmjs.com/package/git-edge)'s `threeWayMerge`, an object-level three-way merge with no worktree needed, built on this package (same `{ fs, gitdir, cache? }` shape; it stores merges through `withPackedWrites`).
 
 ```typescript
@@ -266,6 +279,25 @@ const { entries } = await getCommitLog(repo, { ref: "main", depth: 50 });
 
 `resultKeyPrefixes` (from `/ops`) lists the `ResultCache` key prefixes these functions write under — evict them after rewriting a repo's storage out of band (bulk sync, rename), or a cached walk result outlives the history it describes.
 
+## Submodules
+
+Tree entries retain Git's `type: "commit"` and mode `160000` for submodules.
+HTTP reachability, receive-pack validation, upload-pack and repack treat these
+as external pointers and never try to read the submodule commit in the parent.
+`getFileHistory` tracks pointer changes; `getFileContent` only reads blobs.
+
+`getCommitDiff` and `getDiffBetweenRefs` emit `kind: "submodule"` for additions,
+removals, updates and conversions to/from blobs or trees. `oldEntry`/`newEntry`
+carry the object ID, type and mode. Pointer changes do not count as text line
+additions/deletions. Diff results also include the exact `oldRef` (parent or
+merge base, absent for an empty base) and `newRef` used for navigation.
+
+`getSubmodules(repo, ref, hooks?)` reads root `.gitmodules` at that revision;
+`parseGitmodules(text)` parses its named path/URL records without I/O. Neither
+expands includes nor follows URLs. Missing/oversized config returns no records;
+malformed records and ambiguous duplicate paths are ignored. Browser URL
+validation and authorization to other repositories belong to the application.
+
 ## Semantics & limitations
 
 - **Bare repositories are the target.** Plumbing (`writeBlob`/`writeTree`/`writeCommit`/`readCommit`/`log`/refs/branches) is covered by the test suite. Worktree operations (`checkout`, `add`, `status`) want a real disk — hydrate to `/tmp` for those.

package/dist/chunk-G636FB7W.js

@@ -0,0 +1,143 @@
+import {
+  OBJ_OFS_DELTA,
+  OBJ_REF_DELTA,
+  OID_BYTES,
+  findPackIndexOffset,
+  inflatePrefix,
+  packFileName,
+  parsePackDirectory,
+  parsePackFileName,
+  rawFs,
+  readDeltaHeader,
+  readEntryHeader,
+  readOfsDeltaOffset
+} from "./chunk-HO6OPXUS.js";
+
+// src/git-layout.ts
+var HEX = new Set("0123456789abcdef");
+function isHex(value, length) {
+  if (value.length !== length) return false;
+  for (const char of value) if (!HEX.has(char)) return false;
+  return true;
+}
+function classifyGitdirPath(path) {
+  const segments = path.split("/");
+  const [first, second, third, ...rest] = segments;
+  if (segments.length === 1) {
+    if (first === "HEAD") return { kind: "head" };
+    if (first === "config") return { kind: "config" };
+    if (first === "refs") return { kind: "refs", path };
+    return { kind: "other" };
+  }
+  if (first === "refs") return { kind: "refs", path };
+  if (first !== "objects" || second === void 0) return { kind: "other" };
+  if (segments.length === 2 && second === "") return { kind: "objects-dir" };
+  if (second === "pack" && third !== void 0 && rest.length === 0) {
+    if (third === "") return { kind: "pack-dir" };
+    const file = parsePackFileName(third);
+    return file ? { kind: "pack-file", file } : { kind: "other" };
+  }
+  if (third !== void 0 && rest.length === 0 && isHex(second, 2) && isHex(third, 38)) {
+    return { kind: "loose-object", oid: second + third };
+  }
+  return { kind: "other" };
+}
+function isMutableGitdirEntry(entry) {
+  switch (entry.kind) {
+    case "head":
+    case "refs":
+    case "objects-dir":
+    case "pack-dir":
+      return true;
+    default:
+      return false;
+  }
+}
+function parseLooseObjectPath(path) {
+  const segments = path.split("/");
+  if (segments.length < 3) return null;
+  const gitdirSegments = segments.slice(0, -3);
+  const entry = classifyGitdirPath(segments.slice(-3).join("/"));
+  if (entry.kind !== "loose-object") return null;
+  return { gitdir: gitdirSegments.join("/"), oid: entry.oid };
+}
+
+// src/object-size.ts
+var ENTRY_PROBE_BYTES = 64;
+var DELTA_PROBE_BYTES = 4 * 1024;
+var DELTA_HEADER_MAX_BYTES = 20;
+async function locatePackedObject(repo, oid) {
+  const fsp = rawFs(repo);
+  const packDir = `${repo.gitdir}/objects/pack`;
+  const indexes = parsePackDirectory(
+    await fsp.readdir(packDir).catch(() => [])
+  ).filter((file) => file.kind === "idx");
+  for (const { base } of indexes) {
+    const idx = await fsp.readFile(`${packDir}/${packFileName(base, "idx")}`);
+    if (!(idx instanceof Uint8Array)) continue;
+    const offset = findPackIndexOffset(idx, oid);
+    if (offset !== null) {
+      return { packName: packFileName(base, "pack"), offset };
+    }
+  }
+  return null;
+}
+function rangeReader(repo) {
+  const promises = rawFs(repo);
+  if (!promises.getRange) return null;
+  return (key, start, end) => {
+    if (!promises.getRange) return Promise.resolve(null);
+    return promises.getRange(key, start, end);
+  };
+}
+async function deltaResultSize(read, packKey, dataStart) {
+  const compressed = await read(
+    packKey,
+    dataStart,
+    dataStart + DELTA_PROBE_BYTES - 1
+  );
+  if (!compressed || compressed.byteLength === 0) return null;
+  const inflated = await inflatePrefix(
+    new Uint8Array(compressed),
+    DELTA_HEADER_MAX_BYTES
+  );
+  if (!inflated) return null;
+  const header = readDeltaHeader(inflated, 0);
+  return header.length <= inflated.byteLength ? header.resultSize : null;
+}
+async function getObjectSize(repo, oid) {
+  const read = rangeReader(repo);
+  if (!read) return null;
+  const located = await locatePackedObject(repo, oid);
+  if (!located) return null;
+  const packKey = `${repo.gitdir}/objects/pack/${located.packName}`;
+  const probe = await read(
+    packKey,
+    located.offset,
+    located.offset + ENTRY_PROBE_BYTES - 1
+  );
+  if (!probe || probe.byteLength === 0) return null;
+  const header = readEntryHeader(probe, 0);
+  if (header.length > probe.byteLength) return null;
+  if (header.type !== OBJ_OFS_DELTA && header.type !== OBJ_REF_DELTA) {
+    return header.size;
+  }
+  const bodyStart = located.offset + header.length;
+  let dataStart;
+  if (header.type === OBJ_REF_DELTA) {
+    dataStart = bodyStart + OID_BYTES;
+  } else {
+    const base = readOfsDeltaOffset(probe, header.length);
+    if (header.length + base.length > probe.byteLength) return null;
+    dataStart = bodyStart + base.length;
+  }
+  return deltaResultSize(read, packKey, dataStart);
+}
+
+export {
+  classifyGitdirPath,
+  isMutableGitdirEntry,
+  parseLooseObjectPath,
+  getObjectSize
+};
+//# sourceMappingURL=chunk-G636FB7W.js.map

package/dist/chunk-G636FB7W.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/git-layout.ts","../src/object-size.ts"],"sourcesContent":["/**\n * Typed classification of paths inside a git directory, so callers can\n * reason about \"a ref\", \"a loose object\" or \"the pack listing\" as values\n * instead of matching path strings against patterns.\n *\n * Edge-compatible: no node: imports.\n */\n\nimport { type PackFileName, parsePackFileName } from \"./pack-files.js\";\n\n/** What a gitdir-relative path refers to. Listing prefixes end in `/`. */\nexport type GitdirEntry =\n\t| { readonly kind: \"head\" }\n\t| { readonly kind: \"config\" }\n\t/** A ref file (`refs/heads/main`) or a refs listing (`refs/heads/`). */\n\t| { readonly kind: \"refs\"; readonly path: string }\n\t/** The `objects/` listing. */\n\t| { readonly kind: \"objects-dir\" }\n\t/** The `objects/pack/` listing. */\n\t| { readonly kind: \"pack-dir\" }\n\t| { readonly kind: \"pack-file\"; readonly file: PackFileName }\n\t| { readonly kind: \"loose-object\"; readonly oid: string }\n\t| { readonly kind: \"other\" };\n\nconst HEX = new Set(\"0123456789abcdef\");\n\nfunction isHex(value: string, length: number): boolean {\n\tif (value.length !== length) return false;\n\tfor (const char of value) if (!HEX.has(char)) return false;\n\treturn true;\n}\n\n/** Classify a path relative to a gitdir (no leading slash). */\nexport function classifyGitdirPath(path: string): GitdirEntry {\n\tconst segments = path.split(\"/\");\n\tconst [first, second, third, ...rest] = segments;\n\tif (segments.length === 1) {\n\t\tif (first === \"HEAD\") return { kind: \"head\" };\n\t\tif (first === \"config\") return { kind: \"config\" };\n\t\tif (first === \"refs\") return { kind: \"refs\", path };\n\t\treturn { kind: \"other\" };\n\t}\n\tif (first === \"refs\") return { kind: \"refs\", path };\n\tif (first !== \"objects\" || second === undefined) return { kind: \"other\" };\n\tif (segments.length === 2 && second === \"\") return { kind: \"objects-dir\" };\n\tif (second === \"pack\" && third !== undefined && rest.length === 0) {\n\t\tif (third === \"\") return { kind: \"pack-dir\" };\n\t\tconst file = parsePackFileName(third);\n\t\treturn file ? { kind: \"pack-file\", file } : { kind: \"other\" };\n\t}\n\tif (\n\t\tthird !== undefined &&\n\t\trest.length === 0 &&\n\t\tisHex(second, 2) &&\n\t\tisHex(third, 38)\n\t) {\n\t\treturn { kind: \"loose-object\", oid: second + third };\n\t}\n\treturn { kind: \"other\" };\n}\n\n/**\n * Whether reads of this entry can change between pushes: refs, `HEAD`, and\n * the object/pack listings. Everything else is content-addressed (objects)\n * or written once (config, a given pack file name).\n */\nexport function isMutableGitdirEntry(entry: GitdirEntry): boolean {\n\tswitch (entry.kind) {\n\t\tcase \"head\":\n\t\tcase \"refs\":\n\t\tcase \"objects-dir\":\n\t\tcase \"pack-dir\":\n\t\t\treturn true;\n\t\tdefault:\n\t\t\treturn false;\n\t}\n}\n\n/**\n * A loose-object path anywhere in the fs, split into its gitdir and oid —\n * for callers that don't know where the gitdir boundary is: a loose object\n * is always `<gitdir>/objects/<2 hex>/<38 hex>`.\n */\nexport function parseLooseObjectPath(\n\tpath: string,\n): { readonly gitdir: string; readonly oid: string } | null {\n\tconst segments = path.split(\"/\");\n\tif (segments.length < 3) return null;\n\tconst gitdirSegments = segments.slice(0, -3);\n\tconst entry = classifyGitdirPath(segments.slice(-3).join(\"/\"));\n\tif (entry.kind !== \"loose-object\") return null;\n\treturn { gitdir: gitdirSegments.join(\"/\"), oid: entry.oid };\n}\n","/**\n * An object's size without reading (or inflating) the object.\n *\n * Why this exists: a git object's length lives in its pack entry, but the only\n * way isomorphic-git will hand you an object's bytes is by inflating the whole\n * thing — so \"is this blob too large to render or diff\" could only be asked\n * *after* loading it, which is exactly the cost a size limit is meant to avoid.\n * A 500 MB blob committed to a public repository would be materialised in\n * memory by a single page view before any limit could reject it.\n *\n * Reading the answer is cheap but not trivial, because git stores an object two\n * ways:\n *\n * - **whole** — the entry header records the size, so a small range read at the\n *   entry's offset answers it;\n * - **as a delta** — the entry header records the size of the *delta stream*,\n *   and the reconstructed size is the second varint inside the delta data, which\n *   is itself deflated. Only a bounded prefix of that stream is needed, so\n *   {@link inflatePrefix} recovers it without inflating the object.\n *\n * Either way the answer is exact, which matters because it decides whether a\n * caller may load the object at all. The cost is one `.idx` read (served from\n * the fs cache after the first) plus one or two small range reads of the pack,\n * so a caller can ask about a hundred objects without fetching a pack.\n *\n * `null` means \"not a packed object whose size could be read\" — a loose object,\n * a genuinely absent one, an unreadable index, a store without range reads, or\n * a delta prefix too short to yield its header. A caller must then read the\n * object rather than treating `null` as \"too large\"; see the call sites.\n */\nimport { inflatePrefix } from \"./edge-utils.js\";\nimport { rawFs } from \"./http/types.js\";\nimport type { Repo } from \"./ops/types.js\";\nimport {\n\tOBJ_OFS_DELTA,\n\tOBJ_REF_DELTA,\n\tOID_BYTES,\n\treadDeltaHeader,\n\treadEntryHeader,\n\treadOfsDeltaOffset,\n} from \"./pack-entry.js\";\nimport { packFileName, parsePackDirectory } from \"./pack-files.js\";\nimport { findPackIndexOffset } from \"./pack-index.js\";\n\n/**\n * Bytes fetched at an entry's offset. The entry header itself is at most 10\n * bytes for a 64-bit length, a `REF_DELTA`'s base id is 20, and an `OFS_DELTA`'s\n * base offset at most 10 — 40 bytes, so 64 answers the layout without a second\n * request.\n */\nconst ENTRY_PROBE_BYTES = 64;\n\n/**\n * Compressed bytes fetched from a delta's data. A delta header is two varints\n * (at most 10 bytes each) whose compressed form is smaller still, so this is\n * generous by orders of magnitude while staying a bounded read.\n */\nconst DELTA_PROBE_BYTES = 4 * 1024;\n\n/** The most a delta header can occupy once inflated: two 10-byte varints. */\nconst DELTA_HEADER_MAX_BYTES = 20;\n\n/** The pack holding `oid`, with the offset of its entry. */\nasync function locatePackedObject(\n\trepo: Repo,\n\toid: string,\n): Promise<{ packName: string; offset: number } | null> {\n\tconst fsp = rawFs(repo);\n\tconst packDir = `${repo.gitdir}/objects/pack`;\n\tconst indexes = parsePackDirectory(\n\t\tawait fsp.readdir(packDir).catch(() => [] as string[]),\n\t).filter((file) => file.kind === \"idx\");\n\n\tfor (const { base } of indexes) {\n\t\tconst idx = await fsp.readFile(`${packDir}/${packFileName(base, \"idx\")}`);\n\t\tif (!(idx instanceof Uint8Array)) continue;\n\t\tconst offset = findPackIndexOffset(idx, oid);\n\t\tif (offset !== null) {\n\t\t\treturn { packName: packFileName(base, \"pack\"), offset };\n\t\t}\n\t}\n\treturn null;\n}\n\n/** The store's range reader, or `null` when it has none. */\nfunction rangeReader(repo: Repo) {\n\tconst promises = rawFs(repo) as {\n\t\tgetRange?: (\n\t\t\tkey: string,\n\t\t\tstart: number,\n\t\t\tend: number,\n\t\t) => Promise<Uint8Array | null>;\n\t};\n\tif (!promises.getRange) return null;\n\treturn (key: string, start: number, end: number) => {\n\t\tif (!promises.getRange) return Promise.resolve(null);\n\t\treturn promises.getRange(key, start, end);\n\t};\n}\n\n/**\n * A delta entry's reconstructed size, read from the header its deflated data\n * opens with. `null` when the bounded prefix doesn't yield a complete header.\n */\nasync function deltaResultSize(\n\tread: NonNullable<ReturnType<typeof rangeReader>>,\n\tpackKey: string,\n\tdataStart: number,\n): Promise<number | null> {\n\tconst compressed = await read(\n\t\tpackKey,\n\t\tdataStart,\n\t\tdataStart + DELTA_PROBE_BYTES - 1,\n\t);\n\tif (!compressed || compressed.byteLength === 0) return null;\n\n\tconst inflated = await inflatePrefix(\n\t\tnew Uint8Array(compressed),\n\t\tDELTA_HEADER_MAX_BYTES,\n\t);\n\tif (!inflated) return null;\n\n\tconst header = readDeltaHeader(inflated, 0);\n\treturn header.length <= inflated.byteLength ? header.resultSize : null;\n}\n\n/**\n * `oid`'s size in bytes, or `null` when it isn't a packed object this can read\n * (see the module comment for what that means to a caller).\n */\nexport async function getObjectSize(\n\trepo: Repo,\n\toid: string,\n): Promise<number | null> {\n\t// Without range reads the pack would have to be fetched whole, which is the\n\t// cost this exists to avoid — so report \"unknown\" and let the caller read.\n\tconst read = rangeReader(repo);\n\tif (!read) return null;\n\n\tconst located = await locatePackedObject(repo, oid);\n\tif (!located) return null;\n\n\tconst packKey = `${repo.gitdir}/objects/pack/${located.packName}`;\n\tconst probe = await read(\n\t\tpackKey,\n\t\tlocated.offset,\n\t\tlocated.offset + ENTRY_PROBE_BYTES - 1,\n\t);\n\tif (!probe || probe.byteLength === 0) return null;\n\n\tconst header = readEntryHeader(probe, 0);\n\t// A header that ran off the end of the probe is a header this read, not one\n\t// it can interpret.\n\tif (header.length > probe.byteLength) return null;\n\tif (header.type !== OBJ_OFS_DELTA && header.type !== OBJ_REF_DELTA) {\n\t\treturn header.size;\n\t}\n\n\t// The base reference sits between the entry header and the deflated delta\n\t// data: 20 bytes of oid for `REF_DELTA`, or a backwards offset for\n\t// `OFS_DELTA`.\n\tconst bodyStart = located.offset + header.length;\n\tlet dataStart: number;\n\tif (header.type === OBJ_REF_DELTA) {\n\t\tdataStart = bodyStart + OID_BYTES;\n\t} else {\n\t\tconst base = readOfsDeltaOffset(probe, header.length);\n\t\tif (header.length + base.length > probe.byteLength) return null;\n\t\tdataStart = bodyStart + base.length;\n\t}\n\treturn deltaResultSize(read, packKey, dataStart);\n}\n"],"mappings":";;;;;;;;;;;;;;;;AAwBA,IAAM,MAAM,IAAI,IAAI,kBAAkB;AAEtC,SAAS,MAAM,OAAe,QAAyB;AACtD,MAAI,MAAM,WAAW,OAAQ,QAAO;AACpC,aAAW,QAAQ,MAAO,KAAI,CAAC,IAAI,IAAI,IAAI,EAAG,QAAO;AACrD,SAAO;AACR;AAGO,SAAS,mBAAmB,MAA2B;AAC7D,QAAM,WAAW,KAAK,MAAM,GAAG;AAC/B,QAAM,CAAC,OAAO,QAAQ,OAAO,GAAG,IAAI,IAAI;AACxC,MAAI,SAAS,WAAW,GAAG;AAC1B,QAAI,UAAU,OAAQ,QAAO,EAAE,MAAM,OAAO;AAC5C,QAAI,UAAU,SAAU,QAAO,EAAE,MAAM,SAAS;AAChD,QAAI,UAAU,OAAQ,QAAO,EAAE,MAAM,QAAQ,KAAK;AAClD,WAAO,EAAE,MAAM,QAAQ;AAAA,EACxB;AACA,MAAI,UAAU,OAAQ,QAAO,EAAE,MAAM,QAAQ,KAAK;AAClD,MAAI,UAAU,aAAa,WAAW,OAAW,QAAO,EAAE,MAAM,QAAQ;AACxE,MAAI,SAAS,WAAW,KAAK,WAAW,GAAI,QAAO,EAAE,MAAM,cAAc;AACzE,MAAI,WAAW,UAAU,UAAU,UAAa,KAAK,WAAW,GAAG;AAClE,QAAI,UAAU,GAAI,QAAO,EAAE,MAAM,WAAW;AAC5C,UAAM,OAAO,kBAAkB,KAAK;AACpC,WAAO,OAAO,EAAE,MAAM,aAAa,KAAK,IAAI,EAAE,MAAM,QAAQ;AAAA,EAC7D;AACA,MACC,UAAU,UACV,KAAK,WAAW,KAChB,MAAM,QAAQ,CAAC,KACf,MAAM,OAAO,EAAE,GACd;AACD,WAAO,EAAE,MAAM,gBAAgB,KAAK,SAAS,MAAM;AAAA,EACpD;AACA,SAAO,EAAE,MAAM,QAAQ;AACxB;AAOO,SAAS,qBAAqB,OAA6B;AACjE,UAAQ,MAAM,MAAM;AAAA,IACnB,KAAK;AAAA,IACL,KAAK;AAAA,IACL,KAAK;AAAA,IACL,KAAK;AACJ,aAAO;AAAA,IACR;AACC,aAAO;AAAA,EACT;AACD;AAOO,SAAS,qBACf,MAC2D;AAC3D,QAAM,WAAW,KAAK,MAAM,GAAG;AAC/B,MAAI,SAAS,SAAS,EAAG,QAAO;AAChC,QAAM,iBAAiB,SAAS,MAAM,GAAG,EAAE;AAC3C,QAAM,QAAQ,mBAAmB,SAAS,MAAM,EAAE,EAAE,KAAK,GAAG,CAAC;AAC7D,MAAI,MAAM,SAAS,eAAgB,QAAO;AAC1C,SAAO,EAAE,QAAQ,eAAe,KAAK,GAAG,GAAG,KAAK,MAAM,IAAI;AAC3D;;;AC1CA,IAAM,oBAAoB;AAO1B,IAAM,oBAAoB,IAAI;AAG9B,IAAM,yBAAyB;AAG/B,eAAe,mBACd,MACA,KACuD;AACvD,QAAM,MAAM,MAAM,IAAI;AACtB,QAAM,UAAU,GAAG,KAAK,MAAM;AAC9B,QAAM,UAAU;AAAA,IACf,MAAM,IAAI,QAAQ,OAAO,EAAE,MAAM,MAAM,CAAC,CAAa;AAAA,EACtD,EAAE,OAAO,CAAC,SAAS,KAAK,SAAS,KAAK;AAEtC,aAAW,EAAE,KAAK,KAAK,SAAS;AAC/B,UAAM,MAAM,MAAM,IAAI,SAAS,GAAG,OAAO,IAAI,aAAa,MAAM,KAAK,CAAC,EAAE;AACxE,QAAI,EAAE,eAAe,YAAa;AAClC,UAAM,SAAS,oBAAoB,KAAK,GAAG;AAC3C,QAAI,WAAW,MAAM;AACpB,aAAO,EAAE,UAAU,aAAa,MAAM,MAAM,GAAG,OAAO;AAAA,IACvD;AAAA,EACD;AACA,SAAO;AACR;AAGA,SAAS,YAAY,MAAY;AAChC,QAAM,WAAW,MAAM,IAAI;AAO3B,MAAI,CAAC,SAAS,SAAU,QAAO;AAC/B,SAAO,CAAC,KAAa,OAAe,QAAgB;AACnD,QAAI,CAAC,SAAS,SAAU,QAAO,QAAQ,QAAQ,IAAI;AACnD,WAAO,SAAS,SAAS,KAAK,OAAO,GAAG;AAAA,EACzC;AACD;AAMA,eAAe,gBACd,MACA,SACA,WACyB;AACzB,QAAM,aAAa,MAAM;AAAA,IACxB;AAAA,IACA;AAAA,IACA,YAAY,oBAAoB;AAAA,EACjC;AACA,MAAI,CAAC,cAAc,WAAW,eAAe,EAAG,QAAO;AAEvD,QAAM,WAAW,MAAM;AAAA,IACtB,IAAI,WAAW,UAAU;AAAA,IACzB;AAAA,EACD;AACA,MAAI,CAAC,SAAU,QAAO;AAEtB,QAAM,SAAS,gBAAgB,UAAU,CAAC;AAC1C,SAAO,OAAO,UAAU,SAAS,aAAa,OAAO,aAAa;AACnE;AAMA,eAAsB,cACrB,MACA,KACyB;AAGzB,QAAM,OAAO,YAAY,IAAI;AAC7B,MAAI,CAAC,KAAM,QAAO;AAElB,QAAM,UAAU,MAAM,mBAAmB,MAAM,GAAG;AAClD,MAAI,CAAC,QAAS,QAAO;AAErB,QAAM,UAAU,GAAG,KAAK,MAAM,iBAAiB,QAAQ,QAAQ;AAC/D,QAAM,QAAQ,MAAM;AAAA,IACnB;AAAA,IACA,QAAQ;AAAA,IACR,QAAQ,SAAS,oBAAoB;AAAA,EACtC;AACA,MAAI,CAAC,SAAS,MAAM,eAAe,EAAG,QAAO;AAE7C,QAAM,SAAS,gBAAgB,OAAO,CAAC;AAGvC,MAAI,OAAO,SAAS,MAAM,WAAY,QAAO;AAC7C,MAAI,OAAO,SAAS,iBAAiB,OAAO,SAAS,eAAe;AACnE,WAAO,OAAO;AAAA,EACf;AAKA,QAAM,YAAY,QAAQ,SAAS,OAAO;AAC1C,MAAI;AACJ,MAAI,OAAO,SAAS,eAAe;AAClC,gBAAY,YAAY;AAAA,EACzB,OAAO;AACN,UAAM,OAAO,mBAAmB,OAAO,OAAO,MAAM;AACpD,QAAI,OAAO,SAAS,KAAK,SAAS,MAAM,WAAY,QAAO;AAC3D,gBAAY,YAAY,KAAK;AAAA,EAC9B;AACA,SAAO,gBAAgB,MAAM,SAAS,SAAS;AAChD;","names":[]}

package/dist/{chunk-PASVUPKR.js → chunk-HO6OPXUS.js}

@@ -54,6 +54,28 @@ async function deflate(data) {
   );
   return new Uint8Array(await new Response(stream).arrayBuffer());
 }
+async function inflatePrefix(compressed, enough) {
+  const stream = new Response(compressed).body.pipeThrough(
+    new DecompressionStream("deflate")
+  );
+  const reader = stream.getReader();
+  const chunks = [];
+  let total = 0;
+  try {
+    while (total < enough) {
+      const { done, value } = await reader.read();
+      if (done) break;
+      chunks.push(value);
+      total += value.byteLength;
+    }
+  } catch {
+  } finally {
+    await reader.cancel().catch(() => {
+    });
+  }
+  if (total === 0) return null;
+  return concat(...chunks.map((chunk) => new Uint8Array(chunk)));
+}
 async function inflate(data) {
   const stream = new Response(data).body.pipeThrough(
     new DecompressionStream("deflate")
@@ -216,6 +238,18 @@ function parsePackDirectory(entries) {
   return entries.map(parsePackFileName).filter((file) => file !== null);
 }
 
+// src/http/types.ts
+var runStep = (hooks, label, fn) => hooks?.step ? hooks.step(label, fn) : fn();
+function rawFs(repo) {
+  const fs = repo.fs;
+  if (!fs?.promises) {
+    throw new TypeError(
+      "This operation needs a promise fs (`fs.promises`) \u2014 node:fs and createGitFs both provide one"
+    );
+  }
+  return fs.promises;
+}
+
 // src/refs.ts
 var BAD_REF_COMPONENT = (
   // biome-ignore lint/suspicious/noControlCharactersInRegex: control chars are exactly what git's own ref-name rules reject — this needs to match the same range.
@@ -259,6 +293,98 @@ function qualifyBranchRef(ref) {
   return `refs/heads/${ref}`;
 }
 
+// src/pack-entry.ts
+var OBJ_OFS_DELTA = 6;
+var OBJ_REF_DELTA = 7;
+var OID_BYTES = 20;
+function readEntryHeader(bytes, start) {
+  let byte = bytes[start];
+  const type = byte >> 4 & 7;
+  let size = byte & 15;
+  let shift = 4;
+  let at = start + 1;
+  while (byte & 128) {
+    byte = bytes[at++];
+    size += (byte & 127) * 2 ** shift;
+    shift += 7;
+  }
+  return { type, size, length: at - start };
+}
+function encodeEntryHeader(type, size) {
+  const bytes = [];
+  let rest = Math.floor(size / 16);
+  bytes.push((rest > 0 ? 128 : 0) | type << 4 | size % 16);
+  while (rest > 0) {
+    const low = rest % 128;
+    rest = Math.floor(rest / 128);
+    bytes.push((rest > 0 ? 128 : 0) | low);
+  }
+  return new Uint8Array(bytes);
+}
+function readOfsDeltaOffset(bytes, start) {
+  let at = start;
+  let byte = bytes[at++];
+  let distance = byte & 127;
+  while (byte & 128) {
+    byte = bytes[at++];
+    distance = (distance + 1) * 128 + (byte & 127);
+  }
+  return { distance, length: at - start };
+}
+function readVarint(bytes, start) {
+  let at = start;
+  let byte = bytes[at++];
+  let value = byte & 127;
+  let shift = 7;
+  while (byte & 128) {
+    byte = bytes[at++];
+    value += (byte & 127) * 2 ** shift;
+    shift += 7;
+  }
+  return { value, length: at - start };
+}
+function readDeltaHeader(bytes, start) {
+  const base = readVarint(bytes, start);
+  const result = readVarint(bytes, start + base.length);
+  return {
+    baseSize: base.value,
+    resultSize: result.value,
+    length: base.length + result.length
+  };
+}
+
+// src/pack-index.ts
+var IDX_V2 = {
+  magic: 4285812579,
+  version: 2,
+  headerBytes: 8,
+  fanoutEntries: 256,
+  /** Offsets with this bit set index the 64-bit large-offset table. */
+  largeOffsetFlag: 2147483648
+};
+function readPackIndex(idx) {
+  const view = new DataView(idx.buffer, idx.byteOffset, idx.byteLength);
+  const namesStart = IDX_V2.headerBytes + IDX_V2.fanoutEntries * 4;
+  if (idx.byteLength < namesStart || view.getUint32(0) !== IDX_V2.magic || view.getUint32(4) !== IDX_V2.version) {
+    return null;
+  }
+  const count = view.getUint32(namesStart - 4);
+  const crcStart = namesStart + count * OID_BYTES;
+  const offsetsStart = crcStart + count * 4;
+  if (idx.byteLength < offsetsStart + count * 4) return null;
+  const entries = [];
+  for (let i = 0; i < count; i++) {
+    const offset = view.getUint32(offsetsStart + i * 4);
+    if (offset & IDX_V2.largeOffsetFlag) return null;
+    const at = namesStart + i * OID_BYTES;
+    entries.push({ oid: toHex(idx.subarray(at, at + OID_BYTES)), offset });
+  }
+  return entries;
+}
+function findPackIndexOffset(idx, oid) {
+  return readPackIndex(idx)?.find((entry) => entry.oid === oid)?.offset ?? null;
+}
+
 export {
   encodeUtf8,
   decodeUtf8,
@@ -269,6 +395,7 @@ export {
   fromHex,
   sha1,
   deflate,
+  inflatePrefix,
   inflate,
   hasNullByte,
   readBlobContent,
@@ -289,6 +416,17 @@ export {
   parsePackFileName,
   packFileName,
   parsePackDirectory,
+  runStep,
+  rawFs,
+  OBJ_OFS_DELTA,
+  OBJ_REF_DELTA,
+  OID_BYTES,
+  readEntryHeader,
+  encodeEntryHeader,
+  readOfsDeltaOffset,
+  readDeltaHeader,
+  readPackIndex,
+  findPackIndexOffset,
   isSafeFullRefName,
   isSafeBranchName,
   isFullSha,
@@ -297,4 +435,4 @@ export {
   isSafeTreePath,
   qualifyBranchRef
 };
-//# sourceMappingURL=chunk-PASVUPKR.js.map
+//# sourceMappingURL=chunk-HO6OPXUS.js.map

package/dist/chunk-HO6OPXUS.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/edge-utils.ts","../src/git-errors.ts","../src/pack-files.ts","../src/http/types.ts","../src/refs.ts","../src/pack-entry.ts","../src/pack-index.ts"],"sourcesContent":["/**\n * Edge-compatible utilities replacing node:crypto, node:zlib, and Buffer.\n *\n * Every function here uses only Web APIs (SubtleCrypto, CompressionStream,\n * TextEncoder/TextDecoder) — no Node built-ins. They work on Cloudflare\n * Workers, Vercel Edge, Deno Deploy, and Node >= 18.\n */\n\nconst textEncoder = new TextEncoder();\nconst textDecoder = new TextDecoder();\n\n// ---------------------------------------------------------------------------\n// Text\n// ---------------------------------------------------------------------------\n\n/**\n * Encode a UTF-8 string to bytes.\n *\n * Return type pinned to `Uint8Array<ArrayBuffer>` (not the bare `Uint8Array`,\n * whose default type argument differs across TypeScript versions) so it's\n * always assignable to Fetch API `BodyInit` regardless of a consumer's own\n * TypeScript/lib version.\n */\nexport function encodeUtf8(data: string): Uint8Array<ArrayBuffer> {\n\treturn textEncoder.encode(data);\n}\n\n/** Decode bytes as UTF-8. */\nexport function decodeUtf8(data: Uint8Array): string {\n\treturn textDecoder.decode(data);\n}\n\n/** Decode bytes as ASCII. */\nexport function decodeAscii(data: Uint8Array): string {\n\tlet s = \"\";\n\tfor (let i = 0; i < data.length; i++)\n\t\ts += String.fromCharCode(data[i] as number);\n\treturn s;\n}\n\n// ---------------------------------------------------------------------------\n// Array manipulation\n// ---------------------------------------------------------------------------\n\n/** Concatenate any number of Uint8Arrays into one. */\nexport function concat(...parts: Uint8Array[]): Uint8Array<ArrayBuffer> {\n\tlet total = 0;\n\tfor (const p of parts) total += p.length;\n\tconst out = new Uint8Array(total);\n\tlet offset = 0;\n\tfor (const p of parts) {\n\t\tout.set(p, offset);\n\t\toffset += p.length;\n\t}\n\treturn out;\n}\n\n// ---------------------------------------------------------------------------\n// Encoding\n// ---------------------------------------------------------------------------\n\n/** Uint8Array → lowercase hex string. */\nexport function toHex(data: Uint8Array): string {\n\tlet hex = \"\";\n\tfor (let i = 0; i < data.length; i++)\n\t\thex += (data[i] as number).toString(16).padStart(2, \"0\");\n\treturn hex;\n}\n\n/** Uint8Array → base64 string. */\nexport function toBase64(data: Uint8Array): string {\n\tlet binary = \"\";\n\tfor (let i = 0; i < data.length; i++)\n\t\tbinary += String.fromCharCode(data[i] as number);\n\treturn btoa(binary);\n}\n\n/** Hex string → Uint8Array. */\nexport function fromHex(hex: string): Uint8Array<ArrayBuffer> {\n\tconst bytes = new Uint8Array(hex.length / 2);\n\tfor (let i = 0; i < bytes.length; i++) {\n\t\tbytes[i] = Number.parseInt(hex.slice(i * 2, i * 2 + 2), 16);\n\t}\n\treturn bytes;\n}\n\n// ---------------------------------------------------------------------------\n// Crypto\n// ---------------------------------------------------------------------------\n\n/** SHA-1 hash via Web Crypto API. Returns a hex string. */\nexport async function sha1(\n\tdata: Uint8Array<ArrayBuffer> | string,\n): Promise<string> {\n\tconst bytes = typeof data === \"string\" ? encodeUtf8(data) : data;\n\tconst hash = await globalThis.crypto.subtle.digest(\"SHA-1\", bytes);\n\treturn toHex(new Uint8Array(hash));\n}\n\n// ---------------------------------------------------------------------------\n// Compression\n// ---------------------------------------------------------------------------\n\n/** Deflate (zlib) via the CompressionStream Web API. */\nexport async function deflate(\n\tdata: Uint8Array<ArrayBuffer>,\n): Promise<Uint8Array<ArrayBuffer>> {\n\t// Response, not Blob.stream(): some Blob implementations (jsdom's) have\n\t// no stream().\n\tconst stream = (new Response(data).body as ReadableStream).pipeThrough(\n\t\tnew CompressionStream(\"deflate\"),\n\t);\n\treturn new Uint8Array(await new Response(stream).arrayBuffer());\n}\n\n/**\n * The first `enough` bytes of a deflate stream, given a prefix of that stream.\n *\n * Deflate emits output as it consumes input, so a bounded compressed prefix\n * yields a bounded uncompressed prefix — which is how a delta entry's header\n * (the only place its reconstructed size is recorded) can be read without\n * inflating the whole object. The prefix ends mid-symbol by construction, so\n * the stream errors after the bytes we wanted have already arrived; that error\n * carries no information and is not propagated.\n *\n * Returns `null` when fewer than `enough` bytes could be recovered, so a caller\n * distinguishes \"not readable from this prefix\" from a zero-length answer.\n */\nexport async function inflatePrefix(\n\tcompressed: Uint8Array<ArrayBuffer>,\n\tenough: number,\n): Promise<Uint8Array<ArrayBuffer> | null> {\n\tconst stream = (new Response(compressed).body as ReadableStream).pipeThrough(\n\t\tnew DecompressionStream(\"deflate\"),\n\t);\n\tconst reader = stream.getReader();\n\tconst chunks: Uint8Array[] = [];\n\tlet total = 0;\n\ttry {\n\t\twhile (total < enough) {\n\t\t\tconst { done, value } = await reader.read();\n\t\t\tif (done) break;\n\t\t\tchunks.push(value);\n\t\t\ttotal += value.byteLength;\n\t\t}\n\t} catch {\n\t\t// Expected: the compressed prefix is truncated. Whatever was emitted\n\t\t// before that is what this returns.\n\t} finally {\n\t\tawait reader.cancel().catch(() => {});\n\t}\n\tif (total === 0) return null;\n\treturn concat(...chunks.map((chunk) => new Uint8Array(chunk)));\n}\n\n/** Inflate (zlib) via the DecompressionStream Web API. */\nexport async function inflate(\n\tdata: Uint8Array<ArrayBuffer>,\n): Promise<Uint8Array<ArrayBuffer>> {\n\tconst stream = (new Response(data).body as ReadableStream).pipeThrough(\n\t\tnew DecompressionStream(\"deflate\"),\n\t);\n\treturn new Uint8Array(await new Response(stream).arrayBuffer());\n}\n\n// ---------------------------------------------------------------------------\n// Binary detection (replaces Buffer.includes(0) pattern)\n// ---------------------------------------------------------------------------\n\n/** Check if a Uint8Array contains a null byte. */\nexport function hasNullByte(data: Uint8Array): boolean {\n\treturn data.includes(0);\n}\n\n/**\n * Read a blob as text or binary metadata — the edge-compatible replacement\n * for the `Buffer.from(blob)` pattern used throughout diff.ts and history.ts.\n */\nexport function readBlobContent(blob: Uint8Array): {\n\tisBinary: boolean;\n\ttext: string;\n\tbytes: Uint8Array;\n} {\n\tconst isBinary = hasNullByte(blob);\n\treturn {\n\t\tisBinary,\n\t\ttext: isBinary ? \"\" : decodeUtf8(blob),\n\t\tbytes: blob,\n\t};\n}\n","/**\n * Git-server error types carrying an HTTP status and a retryability flag, so\n * transport layers can map internal failures to responses without inspecting\n * messages. Extend {@link GitError} for app-specific cases (storage backends,\n * quota, …) and {@link formatErrorResponse} keeps working for them.\n */\nexport class GitError extends Error {\n\tstatusCode: number;\n\tretryable: boolean;\n\n\tconstructor(message: string, statusCode = 500, retryable = false) {\n\t\tsuper(message);\n\t\tthis.name = this.constructor.name;\n\t\tthis.statusCode = statusCode;\n\t\tthis.retryable = retryable;\n\t\tError.captureStackTrace?.(this, this.constructor);\n\t}\n\n\ttoJSON(): Record<string, unknown> {\n\t\treturn {\n\t\t\terror: this.name,\n\t\t\tmessage: this.message,\n\t\t\tstatusCode: this.statusCode,\n\t\t\tretryable: this.retryable,\n\t\t};\n\t}\n}\n\n/** A file/directory path not found within a tree (404). */\nexport class GitPathNotFoundError extends GitError {\n\tconstructor(message: string) {\n\t\tsuper(message, 404, false);\n\t}\n}\n\n/** A git object not found (404). */\nexport class GitObjectNotFoundError extends GitError {\n\tconstructor(message: string) {\n\t\tsuper(message, 404, false);\n\t}\n}\n\n/** A ref (branch/tag) not found (404). */\nexport class GitRefNotFoundError extends GitError {\n\tconstructor(message: string) {\n\t\tsuper(message, 404, false);\n\t}\n}\n\n/** The repository itself not found (404). */\nexport class GitRepositoryNotFoundError extends GitError {\n\tconstructor(message: string) {\n\t\tsuper(message, 404, false);\n\t}\n}\n\nexport interface MergeConflictDetail {\n\tfile: string;\n\tbaseLines?: string[];\n\tsourceLines?: string[];\n\ttargetLines?: string[];\n}\n\n/** A merge conflict (409), carrying per-file conflict detail. */\nexport class GitConflictError extends GitError {\n\tconflicts: MergeConflictDetail[];\n\n\tconstructor(message: string, conflicts: MergeConflictDetail[] = []) {\n\t\tsuper(message, 409, false);\n\t\tthis.conflicts = conflicts;\n\t}\n\n\toverride toJSON(): Record<string, unknown> {\n\t\treturn { ...super.toJSON(), conflicts: this.conflicts };\n\t}\n}\n\n/** Authentication failed (401). */\nexport class GitAuthenticationError extends GitError {\n\tconstructor(message: string) {\n\t\tsuper(message, 401, false);\n\t}\n}\n\n/** Authorization failed (403). */\nexport class GitAuthorizationError extends GitError {\n\tconstructor(message: string) {\n\t\tsuper(message, 403, false);\n\t}\n}\n\n/** Too many failed attempts (429). */\nexport class GitRateLimitError extends GitError {\n\tconstructor(message: string) {\n\t\tsuper(message, 429, false);\n\t}\n}\n\n/** Malformed request (400). */\nexport class GitInvalidRequestError extends GitError {\n\tconstructor(message: string) {\n\t\tsuper(message, 400, false);\n\t}\n}\n\n/** Git wire-protocol violation (400). */\nexport class GitProtocolError extends GitError {\n\tconstructor(message: string) {\n\t\tsuper(message, 400, false);\n\t}\n}\n\n/**\n * Map any error to an HTTP response shape. 401s carry the WWW-Authenticate\n * header git clients need before they will prompt for credentials. Non-GitError\n * failures are masked as opaque 500s — internal messages don't leak.\n */\nexport function formatErrorResponse(error: unknown): {\n\tstatus: number;\n\tbody: Record<string, unknown>;\n\theaders?: Record<string, string>;\n} {\n\tif (error instanceof GitError) {\n\t\treturn {\n\t\t\tstatus: error.statusCode,\n\t\t\tbody: error.toJSON(),\n\t\t\theaders:\n\t\t\t\terror.statusCode === 401\n\t\t\t\t\t? { \"WWW-Authenticate\": 'Basic realm=\"Git Repository\"' }\n\t\t\t\t\t: undefined,\n\t\t};\n\t}\n\n\t// The object store's circuit breaker is open (retry.ts's CircuitOpenError):\n\t// the backend is down, not this request — tell clients when to retry.\n\tif ((error as { code?: unknown } | null)?.code === \"EUNAVAILABLE\") {\n\t\treturn {\n\t\t\tstatus: 503,\n\t\t\tbody: {\n\t\t\t\terror: \"ServiceUnavailable\",\n\t\t\t\tmessage: \"Storage is temporarily unavailable\",\n\t\t\t\tretryable: true,\n\t\t\t},\n\t\t\theaders: { \"Retry-After\": \"30\" },\n\t\t};\n\t}\n\n\tif (error instanceof Error) {\n\t\treturn {\n\t\t\tstatus: 500,\n\t\t\tbody: {\n\t\t\t\terror: \"InternalServerError\",\n\t\t\t\tmessage: \"An internal error occurred\",\n\t\t\t\tretryable: true,\n\t\t\t},\n\t\t};\n\t}\n\n\treturn {\n\t\tstatus: 500,\n\t\tbody: {\n\t\t\terror: \"UnknownError\",\n\t\t\tmessage: \"An unknown error occurred\",\n\t\t\tretryable: true,\n\t\t},\n\t};\n}\n","/**\n * Typed view of a git `objects/pack` directory listing.\n *\n * Every pack is a set of sibling files sharing one base name\n * (`pack-<hash>.pack`, `.idx`, and optionally `.keep`/`.bitmap`/`.rev`).\n * Parsing names once into {@link PackFileName} lets callers group, filter and\n * rename by kind instead of re-matching filename strings at each call site.\n *\n * Edge-compatible: no node: imports.\n */\n\n/** Every file kind git writes into `objects/pack`. */\nexport const PACK_FILE_KINDS = [\n\t\"pack\",\n\t\"idx\",\n\t\"keep\",\n\t\"bitmap\",\n\t\"rev\",\n] as const;\nexport type PackFileKind = (typeof PACK_FILE_KINDS)[number];\n\n/** Kinds isomorphic-git actually reads; the rest are git-CLI accelerators. */\nexport const READABLE_PACK_FILE_KINDS: ReadonlySet<PackFileKind> = new Set([\n\t\"pack\",\n\t\"idx\",\n]);\n\nexport interface PackFileName {\n\t/** Shared base of a pack's sibling files, e.g. `pack-1a2b…`. */\n\treadonly base: string;\n\treadonly kind: PackFileKind;\n\t/** The file name as listed, e.g. `pack-1a2b….idx`. */\n\treadonly name: string;\n}\n\nconst KNOWN_KINDS: ReadonlySet<string> = new Set(PACK_FILE_KINDS);\n\nfunction isPackFileKind(value: string): value is PackFileKind {\n\treturn KNOWN_KINDS.has(value);\n}\n\n/** Parse a pack-directory entry; `null` for anything git didn't put there. */\nexport function parsePackFileName(name: string): PackFileName | null {\n\tconst dot = name.lastIndexOf(\".\");\n\tif (dot <= 0) return null;\n\tconst kind = name.slice(dot + 1);\n\tif (!isPackFileKind(kind)) return null;\n\treturn { base: name.slice(0, dot), kind, name };\n}\n\n/** File name of the `kind` sibling of the pack with this base name. */\nexport function packFileName(base: string, kind: PackFileKind): string {\n\treturn `${base}.${kind}`;\n}\n\n/** Parse a directory listing, dropping entries that aren't pack files. */\nexport function parsePackDirectory(entries: readonly string[]): PackFileName[] {\n\treturn entries\n\t\t.map(parsePackFileName)\n\t\t.filter((file): file is PackFileName => file !== null);\n}\n","import type { Repo } from \"../ops/types.js\";\n\n/**\n * Transport-agnostic HTTP response the handlers produce.\n *\n * `body` is pinned to `Uint8Array<ArrayBuffer>` (not the bare `Uint8Array`,\n * whose default type argument isn't consistent across TypeScript versions)\n * so it's always assignable to Fetch API `BodyInit` regardless of which\n * TypeScript/lib version a consumer compiles against — every producer here\n * (`concat`, `pktLine`, `TextEncoder.encode`) is ArrayBuffer-backed already.\n */\nexport interface GitHttpResult {\n\tstatus: number;\n\theaders: Record<string, string>;\n\tbody: Uint8Array<ArrayBuffer>;\n}\n\n/** Optional instrumentation hooks shared by the /http handlers. */\nexport interface HttpHooks {\n\t/** Wrap a timed sub-step. Default: run directly. */\n\tstep?: <T>(label: string, fn: () => Promise<T>) => Promise<T>;\n\t/** Non-fatal problem sink (missing objects, failed repacks). */\n\tonWarn?: (message: string, error?: unknown) => void;\n}\n\nexport const runStep = <T>(\n\thooks: HttpHooks | undefined,\n\tlabel: string,\n\tfn: () => Promise<T>,\n): Promise<T> => (hooks?.step ? hooks.step(label, fn) : fn());\n\n/**\n * The raw promise-fs surface some /http paths need beyond isomorphic-git's\n * plumbing (writing an incoming pack file, listing/deleting pack files).\n * Both `node:fs` and this package's `GitFs` satisfy it via `.promises`.\n */\nexport interface RawFsPromises {\n\treadFile(path: string): Promise<Uint8Array | string>;\n\twriteFile(path: string, data: Uint8Array | string): Promise<void>;\n\tunlink(path: string): Promise<void>;\n\treaddir(path: string): Promise<string[]>;\n\tmkdir(path: string, options?: { recursive?: boolean }): Promise<unknown>;\n\tstat(path: string): Promise<unknown>;\n}\n\n/** Duck-type the `.promises` surface off a repo's fs. */\nexport function rawFs(repo: Repo): RawFsPromises {\n\tconst fs = repo.fs as { promises?: RawFsPromises };\n\tif (!fs?.promises) {\n\t\tthrow new TypeError(\n\t\t\t\"This operation needs a promise fs (`fs.promises`) — node:fs and createGitFs both provide one\",\n\t\t);\n\t}\n\treturn fs.promises;\n}\n","/**\n * Git ref-name validation, mirroring isomorphic-git's own internal `isValidRef`\n * character-class rules (the check `git.branch` and top-level `git.writeRef`\n * run before touching disk).\n *\n * Several of isomorphic-git's OTHER ref-touching primitives — `git.commit`,\n * `git.merge`, `git.deleteBranch`, and top-level `git.resolveRef`/\n * `git.deleteRef` — do NOT run this check internally: they resolve straight\n * through `fs.write`/`fs.rm(join(gitdir, ref))` with no jail to the gitdir.\n * On a shared-storage server (many repos under one prefix or base directory),\n * every branch/ref name that originates from request input must be validated\n * against these predicates before it reaches any of those primitives —\n * otherwise a `\"../\"`-laden name lets a caller with write access to any single\n * repo read, corrupt, or delete another repo's ref/object files.\n */\n\nconst BAD_REF_COMPONENT =\n\t// biome-ignore lint/suspicious/noControlCharactersInRegex: control chars are exactly what git's own ref-name rules reject — this needs to match the same range.\n\t/(^|[/.])([/.]|$)|^@$|@\\{|[\\x00-\\x20\\x7f~^:?*[\\\\]|\\.lock(\\/|$)/;\n\nconst FULL_SHA_RE = /^[0-9a-f]{40}$/i;\n\n/** Validates a fully-qualified ref (must start with refs/heads/ or refs/tags/). */\nexport function isSafeFullRefName(ref: string): boolean {\n\tif (!ref.startsWith(\"refs/heads/\") && !ref.startsWith(\"refs/tags/\")) {\n\t\treturn false;\n\t}\n\treturn !BAD_REF_COMPONENT.test(ref);\n}\n\n/**\n * Validates a bare branch name (no refs/ prefix). Rejects anything that looks\n * like a full ref path — a name of `\"refs/heads/x\"` would otherwise sail\n * through unprefixed at call sites that build `refs/heads/${name}` themselves\n * (doubling the prefix into something that still resolves), or be used as-is\n * at call sites that pass a name already containing `\"refs/\"` straight\n * through. Also rejects 40-hex SHA-shaped values so a stored branch name can\n * never be ambiguous with a commit SHA at write time; use\n * {@link isSafeRefName} on read paths that accept both shapes.\n */\nexport function isSafeBranchName(name: string): boolean {\n\tif (!name || name.startsWith(\"refs/\") || name === \"HEAD\") return false;\n\tif (FULL_SHA_RE.test(name)) return false;\n\treturn !BAD_REF_COMPONENT.test(name);\n}\n\n/** True for a full 40-hex-char commit SHA — the shape {@link isSafeBranchName} deliberately rejects. */\nexport function isFullSha(value: string): boolean {\n\treturn FULL_SHA_RE.test(value);\n}\n\n/**\n * Validates a \"ref\" field that may name either a branch or a commit SHA it's\n * pinned to — the shape read-path route params take (permalinks, raw links).\n * Both shapes still go through the traversal check.\n */\nexport function isSafeRefName(value: string): boolean {\n\treturn isSafeBranchName(value) || isFullSha(value);\n}\n\n/**\n * Validates a repo-relative file path from request input: relative, no `..`\n * segments, no `.git/` prefix, no null bytes. Use this anywhere a path\n * segment comes straight off a URL or form field rather than re-deriving the\n * checks ad hoc.\n */\nexport function isSafeRepoPath(p: string): boolean {\n\tif (p.startsWith(\"/\")) return false;\n\tif (p.split(\"/\").some((segment) => segment === \"..\")) return false;\n\tif (/^\\.git(\\/|$)/i.test(p)) return false;\n\tif (p.includes(\"\\0\")) return false;\n\treturn true;\n}\n\n/**\n * Whether `p` can name a file written into a tree: what {@link isSafeRepoPath}\n * checks, plus every segment non-empty and neither `.` nor `..`, and no\n * `.git` segment anywhere (any case) — git itself refuses to check those out,\n * since a tracked `.git/config` in a subdirectory is a code-execution vector\n * on clone. Use it for paths that become tree entries; reads keep the looser\n * {@link isSafeRepoPath} (which also accepts `\"\"`, the root).\n */\nexport function isSafeTreePath(p: string): boolean {\n\tif (!isSafeRepoPath(p) || p.length === 0) return false;\n\treturn p\n\t\t.split(\"/\")\n\t\t.every(\n\t\t\t(segment) =>\n\t\t\t\tsegment.length > 0 &&\n\t\t\t\tsegment !== \".\" &&\n\t\t\t\tsegment !== \"..\" &&\n\t\t\t\tsegment.toLowerCase() !== \".git\",\n\t\t);\n}\n\n/**\n * Qualify a bare branch name to `refs/heads/<name>` before handing it to\n * isomorphic-git. `resolveRef`/`expand` try several candidate paths in\n * sequence for a bare name — `ref`, `refs/ref`, `refs/tags/ref`,\n * `refs/heads/ref`, … — missing (and, against object storage, paying a real\n * round trip for) the first three every time. For a branch-only ref model,\n * skip straight to the winner. Left untouched: already-qualified refs,\n * `\"HEAD\"` (its own first candidate, already optimal), and 40-hex oids\n * (resolved locally by isomorphic-git with no I/O at all).\n */\nexport function qualifyBranchRef(ref: string): string {\n\tif (ref.startsWith(\"refs/\") || ref === \"HEAD\" || FULL_SHA_RE.test(ref)) {\n\t\treturn ref;\n\t}\n\treturn `refs/heads/${ref}`;\n}\n","/**\n * Parsing the entry headers inside a git packfile (`gitformat-pack(5)`).\n *\n * Shared by the two places that read a pack's structure rather than its\n * objects: `http/pack-reuse.ts` (copying stored entries into a new pack) and\n * `object-size.ts` (answering an object's size without inflating it).\n *\n * Edge-compatible: `Uint8Array` only.\n */\n\n/** Git pack object type bits (bits 6-4 of an entry's first byte). */\nexport const OBJ_COMMIT = 1;\nexport const OBJ_TREE = 2;\nexport const OBJ_BLOB = 3;\nexport const OBJ_TAG = 4;\nexport const OBJ_OFS_DELTA = 6;\nexport const OBJ_REF_DELTA = 7;\n\n/** Bytes in a pack object's id, and in a `REF_DELTA`'s base reference. */\nexport const OID_BYTES = 20;\n\n/** The variable-length type/size header that starts every pack entry. */\nexport interface EntryHeader {\n\treadonly type: number;\n\t/**\n\t * For a whole object, its size in bytes. For a delta entry, the size of the\n\t * **delta stream**, not of the object it reconstructs — that is in the delta\n\t * header (see {@link readDeltaHeader}).\n\t */\n\treadonly size: number;\n\t/** Bytes the header occupies. */\n\treadonly length: number;\n}\n\n/**\n * Read the entry header at `start`.\n *\n * A header that runs off the end of `bytes` yields whatever digits were\n * available: this reads from a bounded slice of a pack, and every caller\n * decides for itself whether the result is plausible (`length` beyond what it\n * read means \"unknown\", not \"zero\").\n */\nexport function readEntryHeader(bytes: Uint8Array, start: number): EntryHeader {\n\tlet byte = bytes[start] as number;\n\tconst type = (byte >> 4) & 0b111;\n\tlet size = byte & 0b1111;\n\tlet shift = 4;\n\tlet at = start + 1;\n\twhile (byte & 0x80) {\n\t\tbyte = bytes[at++] as number;\n\t\tsize += (byte & 0x7f) * 2 ** shift;\n\t\tshift += 7;\n\t}\n\treturn { type, size, length: at - start };\n}\n\n/** Serialize an entry header (the inverse of {@link readEntryHeader}). */\nexport function encodeEntryHeader(type: number, size: number): Uint8Array {\n\tconst bytes: number[] = [];\n\tlet rest = Math.floor(size / 16);\n\tbytes.push((rest > 0 ? 0x80 : 0) | (type << 4) | (size % 16));\n\twhile (rest > 0) {\n\t\tconst low = rest % 128;\n\t\trest = Math.floor(rest / 128);\n\t\tbytes.push((rest > 0 ? 0x80 : 0) | low);\n\t}\n\treturn new Uint8Array(bytes);\n}\n\n/** `OFS_DELTA`'s negative base offset: big-endian base-128 with +1 carries. */\nexport function readOfsDeltaOffset(\n\tbytes: Uint8Array,\n\tstart: number,\n): { distance: number; length: number } {\n\tlet at = start;\n\tlet byte = bytes[at++] as number;\n\tlet distance = byte & 0x7f;\n\twhile (byte & 0x80) {\n\t\tbyte = bytes[at++] as number;\n\t\tdistance = (distance + 1) * 128 + (byte & 0x7f);\n\t}\n\treturn { distance, length: at - start };\n}\n\n/** The two varints a delta stream opens with: base size, then result size. */\nexport interface DeltaHeader {\n\t/** Size of the object the delta is written against. */\n\treadonly baseSize: number;\n\t/** Size of the object the delta reconstructs. */\n\treadonly resultSize: number;\n\t/** Bytes the two varints occupy. */\n\treadonly length: number;\n}\n\n/** Read the base-128 little-endian varint at `start`, 7 bits per byte. */\nfunction readVarint(\n\tbytes: Uint8Array,\n\tstart: number,\n): { value: number; length: number } {\n\tlet at = start;\n\tlet byte = bytes[at++] as number;\n\tlet value = byte & 0x7f;\n\tlet shift = 7;\n\twhile (byte & 0x80) {\n\t\tbyte = bytes[at++] as number;\n\t\tvalue += (byte & 0x7f) * 2 ** shift;\n\t\tshift += 7;\n\t}\n\treturn { value, length: at - start };\n}\n\n/**\n * Read the delta header that follows an `OFS_DELTA` / `REF_DELTA` entry's own\n * header (and, for `REF_DELTA`, its 20-byte base id): the size of the object\n * being modified, then the size of the result. The result size is the answer\n * {@link ObjectSize} exists to get — it is the only place a deltified object's\n * real length is recorded.\n */\nexport function readDeltaHeader(bytes: Uint8Array, start: number): DeltaHeader {\n\tconst base = readVarint(bytes, start);\n\tconst result = readVarint(bytes, start + base.length);\n\treturn {\n\t\tbaseSize: base.value,\n\t\tresultSize: result.value,\n\t\tlength: base.length + result.length,\n\t};\n}\n","/**\n * Reading a pack's version-2 `.idx` (`gitformat-pack(5)`) — the oid → offset\n * table git writes beside every packfile.\n *\n * Kept separate from the code that reads pack *entries*: an index lookup is how\n * you find out which bytes to look at, and it is the cheap half of any question\n * about a packed object's contents (`http/pack-reuse.ts` copies entries into a\n * new pack; `object-size.ts` reads one entry's header).\n *\n * Edge-compatible: `Uint8Array` and `DataView` only.\n */\n\nimport { toHex } from \"./edge-utils.js\";\nimport { OID_BYTES } from \"./pack-entry.js\";\n\nconst IDX_V2 = {\n\tmagic: 0xff744f63,\n\tversion: 2,\n\theaderBytes: 8,\n\tfanoutEntries: 256,\n\t/** Offsets with this bit set index the 64-bit large-offset table. */\n\tlargeOffsetFlag: 0x80000000,\n} as const;\n\n/** An object's position in a pack, as listed by its `.idx`. */\nexport interface PackIndexEntry {\n\treadonly oid: string;\n\treadonly offset: number;\n}\n\n/**\n * Entries of a version-2 `.idx`, in the index's own (oid) order; `null` for any\n * other format, or for one whose declared object count doesn't fit the bytes\n * present — an index that fails either check is not something to guess about.\n */\nexport function readPackIndex(idx: Uint8Array): PackIndexEntry[] | null {\n\tconst view = new DataView(idx.buffer, idx.byteOffset, idx.byteLength);\n\tconst namesStart = IDX_V2.headerBytes + IDX_V2.fanoutEntries * 4;\n\tif (\n\t\tidx.byteLength < namesStart ||\n\t\tview.getUint32(0) !== IDX_V2.magic ||\n\t\tview.getUint32(4) !== IDX_V2.version\n\t) {\n\t\treturn null;\n\t}\n\t// The last fanout slot counts every object in the pack.\n\tconst count = view.getUint32(namesStart - 4);\n\tconst crcStart = namesStart + count * OID_BYTES;\n\tconst offsetsStart = crcStart + count * 4;\n\tif (idx.byteLength < offsetsStart + count * 4) return null;\n\n\tconst entries: PackIndexEntry[] = [];\n\tfor (let i = 0; i < count; i++) {\n\t\tconst offset = view.getUint32(offsetsStart + i * 4);\n\t\t// Packs over 2 GiB use the 64-bit offset table instead; callers that\n\t\t// can't handle it treat this as \"not readable\" rather than guessing.\n\t\tif (offset & IDX_V2.largeOffsetFlag) return null;\n\t\tconst at = namesStart + i * OID_BYTES;\n\t\tentries.push({ oid: toHex(idx.subarray(at, at + OID_BYTES)), offset });\n\t}\n\treturn entries;\n}\n\n/** Offset of `oid` in `idx`, or `null` when the index doesn't list it. */\nexport function findPackIndexOffset(\n\tidx: Uint8Array,\n\toid: string,\n): number | null {\n\treturn readPackIndex(idx)?.find((entry) => entry.oid === oid)?.offset ?? null;\n}\n"],"mappings":";AAQA,IAAM,cAAc,IAAI,YAAY;AACpC,IAAM,cAAc,IAAI,YAAY;AAc7B,SAAS,WAAW,MAAuC;AACjE,SAAO,YAAY,OAAO,IAAI;AAC/B;AAGO,SAAS,WAAW,MAA0B;AACpD,SAAO,YAAY,OAAO,IAAI;AAC/B;AAGO,SAAS,YAAY,MAA0B;AACrD,MAAI,IAAI;AACR,WAAS,IAAI,GAAG,IAAI,KAAK,QAAQ;AAChC,SAAK,OAAO,aAAa,KAAK,CAAC,CAAW;AAC3C,SAAO;AACR;AAOO,SAAS,UAAU,OAA8C;AACvE,MAAI,QAAQ;AACZ,aAAW,KAAK,MAAO,UAAS,EAAE;AAClC,QAAM,MAAM,IAAI,WAAW,KAAK;AAChC,MAAI,SAAS;AACb,aAAW,KAAK,OAAO;AACtB,QAAI,IAAI,GAAG,MAAM;AACjB,cAAU,EAAE;AAAA,EACb;AACA,SAAO;AACR;AAOO,SAAS,MAAM,MAA0B;AAC/C,MAAI,MAAM;AACV,WAAS,IAAI,GAAG,IAAI,KAAK,QAAQ;AAChC,WAAQ,KAAK,CAAC,EAAa,SAAS,EAAE,EAAE,SAAS,GAAG,GAAG;AACxD,SAAO;AACR;AAGO,SAAS,SAAS,MAA0B;AAClD,MAAI,SAAS;AACb,WAAS,IAAI,GAAG,IAAI,KAAK,QAAQ;AAChC,cAAU,OAAO,aAAa,KAAK,CAAC,CAAW;AAChD,SAAO,KAAK,MAAM;AACnB;AAGO,SAAS,QAAQ,KAAsC;AAC7D,QAAM,QAAQ,IAAI,WAAW,IAAI,SAAS,CAAC;AAC3C,WAAS,IAAI,GAAG,IAAI,MAAM,QAAQ,KAAK;AACtC,UAAM,CAAC,IAAI,OAAO,SAAS,IAAI,MAAM,IAAI,GAAG,IAAI,IAAI,CAAC,GAAG,EAAE;AAAA,EAC3D;AACA,SAAO;AACR;AAOA,eAAsB,KACrB,MACkB;AAClB,QAAM,QAAQ,OAAO,SAAS,WAAW,WAAW,IAAI,IAAI;AAC5D,QAAM,OAAO,MAAM,WAAW,OAAO,OAAO,OAAO,SAAS,KAAK;AACjE,SAAO,MAAM,IAAI,WAAW,IAAI,CAAC;AAClC;AAOA,eAAsB,QACrB,MACmC;AAGnC,QAAM,SAAU,IAAI,SAAS,IAAI,EAAE,KAAwB;AAAA,IAC1D,IAAI,kBAAkB,SAAS;AAAA,EAChC;AACA,SAAO,IAAI,WAAW,MAAM,IAAI,SAAS,MAAM,EAAE,YAAY,CAAC;AAC/D;AAeA,eAAsB,cACrB,YACA,QAC0C;AAC1C,QAAM,SAAU,IAAI,SAAS,UAAU,EAAE,KAAwB;AAAA,IAChE,IAAI,oBAAoB,SAAS;AAAA,EAClC;AACA,QAAM,SAAS,OAAO,UAAU;AAChC,QAAM,SAAuB,CAAC;AAC9B,MAAI,QAAQ;AACZ,MAAI;AACH,WAAO,QAAQ,QAAQ;AACtB,YAAM,EAAE,MAAM,MAAM,IAAI,MAAM,OAAO,KAAK;AAC1C,UAAI,KAAM;AACV,aAAO,KAAK,KAAK;AACjB,eAAS,MAAM;AAAA,IAChB;AAAA,EACD,QAAQ;AAAA,EAGR,UAAE;AACD,UAAM,OAAO,OAAO,EAAE,MAAM,MAAM;AAAA,IAAC,CAAC;AAAA,EACrC;AACA,MAAI,UAAU,EAAG,QAAO;AACxB,SAAO,OAAO,GAAG,OAAO,IAAI,CAAC,UAAU,IAAI,WAAW,KAAK,CAAC,CAAC;AAC9D;AAGA,eAAsB,QACrB,MACmC;AACnC,QAAM,SAAU,IAAI,SAAS,IAAI,EAAE,KAAwB;AAAA,IAC1D,IAAI,oBAAoB,SAAS;AAAA,EAClC;AACA,SAAO,IAAI,WAAW,MAAM,IAAI,SAAS,MAAM,EAAE,YAAY,CAAC;AAC/D;AAOO,SAAS,YAAY,MAA2B;AACtD,SAAO,KAAK,SAAS,CAAC;AACvB;AAMO,SAAS,gBAAgB,MAI9B;AACD,QAAM,WAAW,YAAY,IAAI;AACjC,SAAO;AAAA,IACN;AAAA,IACA,MAAM,WAAW,KAAK,WAAW,IAAI;AAAA,IACrC,OAAO;AAAA,EACR;AACD;;;ACvLO,IAAM,WAAN,cAAuB,MAAM;AAAA,EACnC;AAAA,EACA;AAAA,EAEA,YAAY,SAAiB,aAAa,KAAK,YAAY,OAAO;AACjE,UAAM,OAAO;AACb,SAAK,OAAO,KAAK,YAAY;AAC7B,SAAK,aAAa;AAClB,SAAK,YAAY;AACjB,UAAM,oBAAoB,MAAM,KAAK,WAAW;AAAA,EACjD;AAAA,EAEA,SAAkC;AACjC,WAAO;AAAA,MACN,OAAO,KAAK;AAAA,MACZ,SAAS,KAAK;AAAA,MACd,YAAY,KAAK;AAAA,MACjB,WAAW,KAAK;AAAA,IACjB;AAAA,EACD;AACD;AAGO,IAAM,uBAAN,cAAmC,SAAS;AAAA,EAClD,YAAY,SAAiB;AAC5B,UAAM,SAAS,KAAK,KAAK;AAAA,EAC1B;AACD;AAGO,IAAM,yBAAN,cAAqC,SAAS;AAAA,EACpD,YAAY,SAAiB;AAC5B,UAAM,SAAS,KAAK,KAAK;AAAA,EAC1B;AACD;AAGO,IAAM,sBAAN,cAAkC,SAAS;AAAA,EACjD,YAAY,SAAiB;AAC5B,UAAM,SAAS,KAAK,KAAK;AAAA,EAC1B;AACD;AAGO,IAAM,6BAAN,cAAyC,SAAS;AAAA,EACxD,YAAY,SAAiB;AAC5B,UAAM,SAAS,KAAK,KAAK;AAAA,EAC1B;AACD;AAUO,IAAM,mBAAN,cAA+B,SAAS;AAAA,EAC9C;AAAA,EAEA,YAAY,SAAiB,YAAmC,CAAC,GAAG;AACnE,UAAM,SAAS,KAAK,KAAK;AACzB,SAAK,YAAY;AAAA,EAClB;AAAA,EAES,SAAkC;AAC1C,WAAO,EAAE,GAAG,MAAM,OAAO,GAAG,WAAW,KAAK,UAAU;AAAA,EACvD;AACD;AAGO,IAAM,yBAAN,cAAqC,SAAS;AAAA,EACpD,YAAY,SAAiB;AAC5B,UAAM,SAAS,KAAK,KAAK;AAAA,EAC1B;AACD;AAGO,IAAM,wBAAN,cAAoC,SAAS;AAAA,EACnD,YAAY,SAAiB;AAC5B,UAAM,SAAS,KAAK,KAAK;AAAA,EAC1B;AACD;AAGO,IAAM,oBAAN,cAAgC,SAAS;AAAA,EAC/C,YAAY,SAAiB;AAC5B,UAAM,SAAS,KAAK,KAAK;AAAA,EAC1B;AACD;AAGO,IAAM,yBAAN,cAAqC,SAAS;AAAA,EACpD,YAAY,SAAiB;AAC5B,UAAM,SAAS,KAAK,KAAK;AAAA,EAC1B;AACD;AAGO,IAAM,mBAAN,cAA+B,SAAS;AAAA,EAC9C,YAAY,SAAiB;AAC5B,UAAM,SAAS,KAAK,KAAK;AAAA,EAC1B;AACD;AAOO,SAAS,oBAAoB,OAIlC;AACD,MAAI,iBAAiB,UAAU;AAC9B,WAAO;AAAA,MACN,QAAQ,MAAM;AAAA,MACd,MAAM,MAAM,OAAO;AAAA,MACnB,SACC,MAAM,eAAe,MAClB,EAAE,oBAAoB,+BAA+B,IACrD;AAAA,IACL;AAAA,EACD;AAIA,MAAK,OAAqC,SAAS,gBAAgB;AAClE,WAAO;AAAA,MACN,QAAQ;AAAA,MACR,MAAM;AAAA,QACL,OAAO;AAAA,QACP,SAAS;AAAA,QACT,WAAW;AAAA,MACZ;AAAA,MACA,SAAS,EAAE,eAAe,KAAK;AAAA,IAChC;AAAA,EACD;AAEA,MAAI,iBAAiB,OAAO;AAC3B,WAAO;AAAA,MACN,QAAQ;AAAA,MACR,MAAM;AAAA,QACL,OAAO;AAAA,QACP,SAAS;AAAA,QACT,WAAW;AAAA,MACZ;AAAA,IACD;AAAA,EACD;AAEA,SAAO;AAAA,IACN,QAAQ;AAAA,IACR,MAAM;AAAA,MACL,OAAO;AAAA,MACP,SAAS;AAAA,MACT,WAAW;AAAA,IACZ;AAAA,EACD;AACD;;;AC1JO,IAAM,kBAAkB;AAAA,EAC9B;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AACD;AAIO,IAAM,2BAAsD,oBAAI,IAAI;AAAA,EAC1E;AAAA,EACA;AACD,CAAC;AAUD,IAAM,cAAmC,IAAI,IAAI,eAAe;AAEhE,SAAS,eAAe,OAAsC;AAC7D,SAAO,YAAY,IAAI,KAAK;AAC7B;AAGO,SAAS,kBAAkB,MAAmC;AACpE,QAAM,MAAM,KAAK,YAAY,GAAG;AAChC,MAAI,OAAO,EAAG,QAAO;AACrB,QAAM,OAAO,KAAK,MAAM,MAAM,CAAC;AAC/B,MAAI,CAAC,eAAe,IAAI,EAAG,QAAO;AAClC,SAAO,EAAE,MAAM,KAAK,MAAM,GAAG,GAAG,GAAG,MAAM,KAAK;AAC/C;AAGO,SAAS,aAAa,MAAc,MAA4B;AACtE,SAAO,GAAG,IAAI,IAAI,IAAI;AACvB;AAGO,SAAS,mBAAmB,SAA4C;AAC9E,SAAO,QACL,IAAI,iBAAiB,EACrB,OAAO,CAAC,SAA+B,SAAS,IAAI;AACvD;;;ACnCO,IAAM,UAAU,CACtB,OACA,OACA,OACiB,OAAO,OAAO,MAAM,KAAK,OAAO,EAAE,IAAI,GAAG;AAiBpD,SAAS,MAAM,MAA2B;AAChD,QAAM,KAAK,KAAK;AAChB,MAAI,CAAC,IAAI,UAAU;AAClB,UAAM,IAAI;AAAA,MACT;AAAA,IACD;AAAA,EACD;AACA,SAAO,GAAG;AACX;;;ACtCA,IAAM;AAAA;AAAA,EAEL;AAAA;AAED,IAAM,cAAc;AAGb,SAAS,kBAAkB,KAAsB;AACvD,MAAI,CAAC,IAAI,WAAW,aAAa,KAAK,CAAC,IAAI,WAAW,YAAY,GAAG;AACpE,WAAO;AAAA,EACR;AACA,SAAO,CAAC,kBAAkB,KAAK,GAAG;AACnC;AAYO,SAAS,iBAAiB,MAAuB;AACvD,MAAI,CAAC,QAAQ,KAAK,WAAW,OAAO,KAAK,SAAS,OAAQ,QAAO;AACjE,MAAI,YAAY,KAAK,IAAI,EAAG,QAAO;AACnC,SAAO,CAAC,kBAAkB,KAAK,IAAI;AACpC;AAGO,SAAS,UAAU,OAAwB;AACjD,SAAO,YAAY,KAAK,KAAK;AAC9B;AAOO,SAAS,cAAc,OAAwB;AACrD,SAAO,iBAAiB,KAAK,KAAK,UAAU,KAAK;AAClD;AAQO,SAAS,eAAe,GAAoB;AAClD,MAAI,EAAE,WAAW,GAAG,EAAG,QAAO;AAC9B,MAAI,EAAE,MAAM,GAAG,EAAE,KAAK,CAAC,YAAY,YAAY,IAAI,EAAG,QAAO;AAC7D,MAAI,gBAAgB,KAAK,CAAC,EAAG,QAAO;AACpC,MAAI,EAAE,SAAS,IAAI,EAAG,QAAO;AAC7B,SAAO;AACR;AAUO,SAAS,eAAe,GAAoB;AAClD,MAAI,CAAC,eAAe,CAAC,KAAK,EAAE,WAAW,EAAG,QAAO;AACjD,SAAO,EACL,MAAM,GAAG,EACT;AAAA,IACA,CAAC,YACA,QAAQ,SAAS,KACjB,YAAY,OACZ,YAAY,QACZ,QAAQ,YAAY,MAAM;AAAA,EAC5B;AACF;AAYO,SAAS,iBAAiB,KAAqB;AACrD,MAAI,IAAI,WAAW,OAAO,KAAK,QAAQ,UAAU,YAAY,KAAK,GAAG,GAAG;AACvE,WAAO;AAAA,EACR;AACA,SAAO,cAAc,GAAG;AACzB;;;AC/FO,IAAM,gBAAgB;AACtB,IAAM,gBAAgB;AAGtB,IAAM,YAAY;AAuBlB,SAAS,gBAAgB,OAAmB,OAA4B;AAC9E,MAAI,OAAO,MAAM,KAAK;AACtB,QAAM,OAAQ,QAAQ,IAAK;AAC3B,MAAI,OAAO,OAAO;AAClB,MAAI,QAAQ;AACZ,MAAI,KAAK,QAAQ;AACjB,SAAO,OAAO,KAAM;AACnB,WAAO,MAAM,IAAI;AACjB,aAAS,OAAO,OAAQ,KAAK;AAC7B,aAAS;AAAA,EACV;AACA,SAAO,EAAE,MAAM,MAAM,QAAQ,KAAK,MAAM;AACzC;AAGO,SAAS,kBAAkB,MAAc,MAA0B;AACzE,QAAM,QAAkB,CAAC;AACzB,MAAI,OAAO,KAAK,MAAM,OAAO,EAAE;AAC/B,QAAM,MAAM,OAAO,IAAI,MAAO,KAAM,QAAQ,IAAM,OAAO,EAAG;AAC5D,SAAO,OAAO,GAAG;AAChB,UAAM,MAAM,OAAO;AACnB,WAAO,KAAK,MAAM,OAAO,GAAG;AAC5B,UAAM,MAAM,OAAO,IAAI,MAAO,KAAK,GAAG;AAAA,EACvC;AACA,SAAO,IAAI,WAAW,KAAK;AAC5B;AAGO,SAAS,mBACf,OACA,OACuC;AACvC,MAAI,KAAK;AACT,MAAI,OAAO,MAAM,IAAI;AACrB,MAAI,WAAW,OAAO;AACtB,SAAO,OAAO,KAAM;AACnB,WAAO,MAAM,IAAI;AACjB,gBAAY,WAAW,KAAK,OAAO,OAAO;AAAA,EAC3C;AACA,SAAO,EAAE,UAAU,QAAQ,KAAK,MAAM;AACvC;AAaA,SAAS,WACR,OACA,OACoC;AACpC,MAAI,KAAK;AACT,MAAI,OAAO,MAAM,IAAI;AACrB,MAAI,QAAQ,OAAO;AACnB,MAAI,QAAQ;AACZ,SAAO,OAAO,KAAM;AACnB,WAAO,MAAM,IAAI;AACjB,cAAU,OAAO,OAAQ,KAAK;AAC9B,aAAS;AAAA,EACV;AACA,SAAO,EAAE,OAAO,QAAQ,KAAK,MAAM;AACpC;AASO,SAAS,gBAAgB,OAAmB,OAA4B;AAC9E,QAAM,OAAO,WAAW,OAAO,KAAK;AACpC,QAAM,SAAS,WAAW,OAAO,QAAQ,KAAK,MAAM;AACpD,SAAO;AAAA,IACN,UAAU,KAAK;AAAA,IACf,YAAY,OAAO;AAAA,IACnB,QAAQ,KAAK,SAAS,OAAO;AAAA,EAC9B;AACD;;;AC/GA,IAAM,SAAS;AAAA,EACd,OAAO;AAAA,EACP,SAAS;AAAA,EACT,aAAa;AAAA,EACb,eAAe;AAAA;AAAA,EAEf,iBAAiB;AAClB;AAaO,SAAS,cAAc,KAA0C;AACvE,QAAM,OAAO,IAAI,SAAS,IAAI,QAAQ,IAAI,YAAY,IAAI,UAAU;AACpE,QAAM,aAAa,OAAO,cAAc,OAAO,gBAAgB;AAC/D,MACC,IAAI,aAAa,cACjB,KAAK,UAAU,CAAC,MAAM,OAAO,SAC7B,KAAK,UAAU,CAAC,MAAM,OAAO,SAC5B;AACD,WAAO;AAAA,EACR;AAEA,QAAM,QAAQ,KAAK,UAAU,aAAa,CAAC;AAC3C,QAAM,WAAW,aAAa,QAAQ;AACtC,QAAM,eAAe,WAAW,QAAQ;AACxC,MAAI,IAAI,aAAa,eAAe,QAAQ,EAAG,QAAO;AAEtD,QAAM,UAA4B,CAAC;AACnC,WAAS,IAAI,GAAG,IAAI,OAAO,KAAK;AAC/B,UAAM,SAAS,KAAK,UAAU,eAAe,IAAI,CAAC;AAGlD,QAAI,SAAS,OAAO,gBAAiB,QAAO;AAC5C,UAAM,KAAK,aAAa,IAAI;AAC5B,YAAQ,KAAK,EAAE,KAAK,MAAM,IAAI,SAAS,IAAI,KAAK,SAAS,CAAC,GAAG,OAAO,CAAC;AAAA,EACtE;AACA,SAAO;AACR;AAGO,SAAS,oBACf,KACA,KACgB;AAChB,SAAO,cAAc,GAAG,GAAG,KAAK,CAAC,UAAU,MAAM,QAAQ,GAAG,GAAG,UAAU;AAC1E;","names":[]}

package/dist/{chunk-HBOJR5D3.js → chunk-QH3HS7VI.js}

@@ -7,9 +7,10 @@ import {
   fromHex,
   isSafeBranchName,
   packFileName,
+  rawFs,
   sha1,
   toHex
-} from "./chunk-PASVUPKR.js";
+} from "./chunk-HO6OPXUS.js";
 
 // src/ops/branch.ts
 import git from "isomorphic-git";
@@ -138,18 +139,6 @@ async function assertBranchExists(repo, name) {
   await resolveLooseRefFast(repo, `refs/heads/${name}`);
 }
 
-// src/http/types.ts
-var runStep = (hooks, label, fn) => hooks?.step ? hooks.step(label, fn) : fn();
-function rawFs(repo) {
-  const fs = repo.fs;
-  if (!fs?.promises) {
-    throw new TypeError(
-      "This operation needs a promise fs (`fs.promises`) \u2014 node:fs and createGitFs both provide one"
-    );
-  }
-  return fs.promises;
-}
-
 // src/pack-write.ts
 import git2 from "isomorphic-git";
 var PACK_OBJECT_TYPE_BITS = {
@@ -214,6 +203,42 @@ async function storePack(repo, pack) {
   return { base, oids };
 }
 
+// src/semaphore.ts
+function createSemaphore(limit) {
+  if (!Number.isInteger(limit) || limit < 1) {
+    throw new Error(
+      `createSemaphore: limit must be an integer >= 1, got ${limit}`
+    );
+  }
+  let active = 0;
+  const waiting = [];
+  function release() {
+    const next = waiting.shift();
+    if (next) next();
+    else active--;
+  }
+  function acquire() {
+    if (active < limit) {
+      active++;
+      return Promise.resolve();
+    }
+    return new Promise((resolve) => {
+      waiting.push(resolve);
+    });
+  }
+  return {
+    run(fn) {
+      return acquire().then(async () => {
+        try {
+          return await fn();
+        } finally {
+          release();
+        }
+      });
+    }
+  };
+}
+
 export {
   assertSafeBranchName,
   listLooseRefNames,
@@ -222,11 +247,10 @@ export {
   createBranchFrom,
   deleteBranchByName,
   assertBranchExists,
-  runStep,
-  rawFs,
   hashGitObject,
   encodePack,
   packBaseName,
-  storePack
+  storePack,
+  createSemaphore
 };
-//# sourceMappingURL=chunk-HBOJR5D3.js.map
+//# sourceMappingURL=chunk-QH3HS7VI.js.map

package/dist/chunk-QH3HS7VI.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/ops/branch.ts","../src/pack-write.ts","../src/semaphore.ts"],"sourcesContent":["import git from \"isomorphic-git\";\nimport { GitConflictError, GitInvalidRequestError } from \"../git-errors.js\";\nimport { isSafeBranchName } from \"../refs.js\";\nimport type { Repo } from \"./types.js\";\n\nexport interface Branch {\n\tname: string;\n\tcommit: string;\n\tisDefault: boolean;\n}\n\n/**\n * Defense in depth for every branch-name argument below: `git.deleteBranch`\n * and raw resolveRef/writeRef reads don't validate ref names internally the\n * way `git.branch` does (see refs.ts) — guard at the point the primitives are\n * actually called, not just at an API boundary far above.\n */\nexport function assertSafeBranchName(name: string): void {\n\tif (!isSafeBranchName(name)) {\n\t\tthrow new GitInvalidRequestError(`Invalid branch name: ${name}`);\n\t}\n}\n\nconst FULL_SHA_RE = /^[0-9a-f]{40}$/i;\n\ntype RecursiveFileLister = {\n\tlistFilesRecursively(dirpath: string): Promise<string[]>;\n};\n\nfunction hasRecursiveFileLister(\n\tfs: Repo[\"fs\"],\n): fs is Repo[\"fs\"] & RecursiveFileLister {\n\treturn (\n\t\ttypeof fs === \"object\" &&\n\t\tfs !== null &&\n\t\t\"listFilesRecursively\" in fs &&\n\t\ttypeof (fs as Partial<RecursiveFileLister>).listFilesRecursively ===\n\t\t\t\"function\"\n\t);\n}\n\ntype PromisesFs = {\n\treadFile(path: string, encoding: \"utf8\"): Promise<string | Uint8Array>;\n};\n\n// Repo.fs is typed as isomorphic-git's own fs union (promise- or\n// callback-based) since that's what git.readTree's signature allows, but\n// every fs this package actually constructs (git-fs.ts) is promise-based —\n// this just keeps the callback-style branch honest instead of crashing on\n// it. Shared by every loose-ref fast path below instead of each repeating\n// the same cast + presence check.\nfunction getPromisesFs(fs: Repo[\"fs\"]): PromisesFs | null {\n\tconst promises = (fs as { promises?: PromisesFs }).promises;\n\treturn promises ?? null;\n}\n\nfunction decodeFileContent(content: string | Uint8Array): string {\n\treturn typeof content === \"string\"\n\t\t? content\n\t\t: new TextDecoder().decode(content);\n}\n\n/**\n * Lists loose branch names without isomorphic-git's recursive readdir/stat\n * traversal. Object stores already return only leaf keys in a recursive LIST,\n * so this is one request regardless of branch-name nesting. Ordinary\n * filesystems retain isomorphic-git's implementation and packed refs remain\n * supported there.\n */\nasync function listLooseBranches(repo: Repo): Promise<string[] | null> {\n\treturn listLooseRefNames(repo, \"refs/heads\");\n}\n\n/**\n * Loose ref names under `dir` (e.g. `refs/tags`) in one recursive listing,\n * or `null` when the fs can't list recursively (use isomorphic-git's walk).\n */\nexport async function listLooseRefNames(\n\trepo: Repo,\n\tdir: \"refs/heads\" | \"refs/tags\",\n): Promise<string[] | null> {\n\tif (!hasRecursiveFileLister(repo.fs)) return null;\n\tconst names = await repo.fs.listFilesRecursively(`${repo.gitdir}/${dir}`);\n\treturn names.filter(isSafeBranchName);\n}\n\n/** Read packed branch refs so the object-store fast path preserves Git semantics. */\nasync function readPackedBranches(repo: Repo): Promise<Map<string, string>> {\n\tif (!hasRecursiveFileLister(repo.fs)) return new Map();\n\tconst promisesFs = getPromisesFs(repo.fs);\n\tif (!promisesFs) return new Map();\n\ttry {\n\t\tconst content = await promisesFs.readFile(\n\t\t\t`${repo.gitdir}/packed-refs`,\n\t\t\t\"utf8\",\n\t\t);\n\t\tconst refs = new Map<string, string>();\n\t\tfor (const line of decodeFileContent(content).split(\"\\n\")) {\n\t\t\tconst match = /^([0-9a-f]{40}) refs\\/heads\\/(.+)$/i.exec(line);\n\t\t\tif (match?.[1] && match[2] && isSafeBranchName(match[2])) {\n\t\t\t\trefs.set(match[2], match[1]);\n\t\t\t}\n\t\t}\n\t\treturn refs;\n\t} catch {\n\t\treturn new Map();\n\t}\n}\n\n/** Read the symbolic HEAD directly when the object-store fs is available. */\nasync function getCurrentLooseBranch(repo: Repo): Promise<string | null> {\n\tif (!hasRecursiveFileLister(repo.fs)) return null;\n\tconst promisesFs = getPromisesFs(repo.fs);\n\tif (!promisesFs) return null;\n\ttry {\n\t\tconst content = await promisesFs.readFile(`${repo.gitdir}/HEAD`, \"utf8\");\n\t\tconst head = decodeFileContent(content).trim();\n\t\tconst match = /^ref: refs\\/heads\\/(.+)$/.exec(head);\n\t\treturn match?.[1] && isSafeBranchName(match[1]) ? match[1] : null;\n\t} catch {\n\t\treturn null;\n\t}\n}\n\n/**\n * Resolve a fully-qualified ref's oid with one read instead of isomorphic-\n * git's own `resolveRef`, which does a stat *then* a read for a loose ref\n * (two round trips against object storage) — wasteful specifically here,\n * where the caller already knows the ref exists (it came from a directory\n * listing moments earlier). Reads the loose file directly and falls back to\n * `git.resolveRef` only when that doesn't pan out (packed-refs, or anything\n * else the plain-loose-file assumption doesn't cover), so correctness for\n * those cases is unchanged — this only removes the redundant round trip on\n * the common path.\n */\nexport async function resolveLooseRefFast(\n\trepo: Repo,\n\tref: string,\n): Promise<string> {\n\tconst promisesFs = getPromisesFs(repo.fs);\n\tif (promisesFs) {\n\t\ttry {\n\t\t\tconst content = await promisesFs.readFile(\n\t\t\t\t`${repo.gitdir}/${ref}`,\n\t\t\t\t\"utf8\",\n\t\t\t);\n\t\t\tconst oid = decodeFileContent(content).trim();\n\t\t\tif (FULL_SHA_RE.test(oid)) return oid;\n\t\t} catch {\n\t\t\t// Not a loose file (packed-refs, or genuinely absent) — fall through.\n\t\t}\n\t}\n\treturn git.resolveRef({ ...repo, ref });\n}\n\n/** All branches with their tip commits; [] for an empty repository. */\nexport async function listBranches(repo: Repo): Promise<Branch[]> {\n\ttry {\n\t\tconst [looseBranches, looseCurrentBranch, packedBranches] =\n\t\t\tawait Promise.all([\n\t\t\t\tlistLooseBranches(repo),\n\t\t\t\tgetCurrentLooseBranch(repo),\n\t\t\t\treadPackedBranches(repo),\n\t\t\t]);\n\t\tconst branches =\n\t\t\tlooseBranches === null\n\t\t\t\t? await git.listBranches(repo)\n\t\t\t\t: [...new Set([...looseBranches, ...packedBranches.keys()])].sort();\n\t\tconst currentBranch =\n\t\t\tlooseBranches === null\n\t\t\t\t? await git\n\t\t\t\t\t\t.currentBranch({ ...repo, fullname: false })\n\t\t\t\t\t\t.catch(() => null)\n\t\t\t\t: looseCurrentBranch;\n\n\t\treturn Promise.all(\n\t\t\tbranches.map(async (branch) => {\n\t\t\t\tconst packedCommit = packedBranches.get(branch);\n\t\t\t\treturn {\n\t\t\t\t\tname: branch,\n\t\t\t\t\tcommit:\n\t\t\t\t\t\tpackedCommit && !looseBranches?.includes(branch)\n\t\t\t\t\t\t\t? packedCommit\n\t\t\t\t\t\t\t: await resolveLooseRefFast(repo, `refs/heads/${branch}`),\n\t\t\t\t\tisDefault: branch === currentBranch,\n\t\t\t\t};\n\t\t\t}),\n\t\t);\n\t} catch (err: unknown) {\n\t\tif ((err as { code?: string }).code === \"NotFoundError\") return [];\n\t\tthrow err;\n\t}\n}\n\n/** Create `name` pointing at the tip of `startPoint` (no checkout). */\nexport async function createBranchFrom(\n\trepo: Repo,\n\tname: string,\n\tstartPoint: string,\n): Promise<void> {\n\tassertSafeBranchName(name);\n\tassertSafeBranchName(startPoint);\n\tconst object = await resolveLooseRefFast(repo, `refs/heads/${startPoint}`);\n\ttry {\n\t\tawait git.branch({ ...repo, ref: name, checkout: false, object });\n\t} catch (err) {\n\t\tif ((err as { code?: string }).code === \"AlreadyExistsError\") {\n\t\t\tthrow new GitConflictError(`Branch ${name} already exists`);\n\t\t}\n\t\tthrow err;\n\t}\n}\n\n/** Delete a branch ref (validated — deleteBranch has no internal ref check). */\nexport async function deleteBranchByName(\n\trepo: Repo,\n\tname: string,\n): Promise<void> {\n\tassertSafeBranchName(name);\n\t// Same rule as receive-pack: deleting the branch HEAD names leaves every\n\t// clone without a default branch.\n\tconst current =\n\t\t(await getCurrentLooseBranch(repo)) ??\n\t\t(await git.currentBranch({ ...repo, fullname: false }).catch(() => null));\n\tif (current === name) {\n\t\tthrow new GitInvalidRequestError(\n\t\t\t`Cannot delete ${name}: it is the default branch`,\n\t\t);\n\t}\n\tawait git.deleteBranch({ ...repo, ref: name });\n}\n\n/** Throws (NotFoundError) unless the branch resolves. */\nexport async function assertBranchExists(\n\trepo: Repo,\n\tname: string,\n): Promise<void> {\n\tassertSafeBranchName(name);\n\tawait resolveLooseRefFast(repo, `refs/heads/${name}`);\n}\n","/**\n * Writing packfiles: serialize whole (never deltified) objects into a pack,\n * and store a pack plus its index in a repository. Shared by repack (the\n * loose-object fallback) and packed writes (ops/packed-writes.ts).\n *\n * Edge-compatible: Web Crypto SHA-1, CompressionStream, Uint8Array only.\n */\nimport git from \"isomorphic-git\";\nimport {\n\tconcat,\n\tdeflate,\n\tencodeUtf8,\n\tfromHex,\n\tsha1,\n\ttoHex,\n} from \"./edge-utils.js\";\nimport { rawFs } from \"./http/types.js\";\nimport type { Repo } from \"./ops/types.js\";\nimport { packFileName } from \"./pack-files.js\";\n\nexport type PackObjectType = \"commit\" | \"tree\" | \"blob\" | \"tag\";\n\n// Git pack object type bits (bits 6-4 of the header's first byte).\nconst PACK_OBJECT_TYPE_BITS: Record<PackObjectType, number> = {\n\tcommit: 0b0010000,\n\ttree: 0b0100000,\n\tblob: 0b0110000,\n\ttag: 0b1000000,\n};\n\n// First byte: continuation bit | 3-bit type | low 4 bits of length; the rest\n// of the length follows 7 bits at a time, little-endian.\nfunction encodePackObjectHeader(\n\ttype: PackObjectType,\n\tlength: number,\n): Uint8Array {\n\tconst bytes: number[] = [];\n\tlet more = length > 0b1111;\n\tbytes.push(\n\t\t(more ? 0b10000000 : 0) | PACK_OBJECT_TYPE_BITS[type] | (length & 0b1111),\n\t);\n\tlet rest = Math.floor(length / 16);\n\twhile (more) {\n\t\tmore = rest > 0b01111111;\n\t\tbytes.push((more ? 0b10000000 : 0) | (rest & 0b01111111));\n\t\trest = Math.floor(rest / 128);\n\t}\n\treturn new Uint8Array(bytes);\n}\n\n/** Git's object id: sha1(\"<type> <byte length>\\0<content>\"). */\nexport function hashGitObject(\n\ttype: string,\n\tcontent: Uint8Array,\n): Promise<string> {\n\treturn sha1(concat(encodeUtf8(`${type} ${content.length}\\0`), content));\n}\n\nexport interface PackObject {\n\ttype: PackObjectType;\n\tcontent: Uint8Array<ArrayBuffer>;\n}\n\n/** A version-2 pack holding `objects` as full entries, in order. */\nexport async function encodePack(\n\tobjects: readonly PackObject[],\n): Promise<Uint8Array<ArrayBuffer>> {\n\tconst header = new Uint8Array(12);\n\tconst view = new DataView(header.buffer);\n\theader.set(encodeUtf8(\"PACK\"), 0);\n\tview.setUint32(4, 2, false);\n\tview.setUint32(8, objects.length, false);\n\n\tconst chunks: Uint8Array[] = [header];\n\tconst BATCH_SIZE = 100;\n\tfor (let i = 0; i < objects.length; i += BATCH_SIZE) {\n\t\tconst encoded = await Promise.all(\n\t\t\tobjects\n\t\t\t\t.slice(i, i + BATCH_SIZE)\n\t\t\t\t.map(async (object) =>\n\t\t\t\t\tconcat(\n\t\t\t\t\t\tencodePackObjectHeader(object.type, object.content.length),\n\t\t\t\t\t\tawait deflate(object.content),\n\t\t\t\t\t),\n\t\t\t\t),\n\t\t);\n\t\tchunks.push(...encoded);\n\t}\n\t// The trailer is the SHA-1 of every preceding byte.\n\tconst body = concat(...chunks);\n\treturn concat(body, fromHex(await sha1(body)));\n}\n\n/**\n * A pack's conventional base name, `pack-<checksum>`: the trailer SHA-1 is a\n * function of the content, so two writers can never pick the same name for\n * different packs, and rewriting the same pack is idempotent.\n */\nexport function packBaseName(pack: Uint8Array): string {\n\treturn `pack-${toHex(pack.subarray(pack.length - 20))}`;\n}\n\n/**\n * Store `pack` in `repo` and index it (isomorphic-git's `indexPack`, which\n * re-hashes every object). The `.idx` is written last: a pack only becomes\n * visible to readers once its index exists. Returns the base name and the\n * oids the index lists.\n */\nexport async function storePack(\n\trepo: Repo,\n\tpack: Uint8Array,\n): Promise<{ base: string; oids: string[] }> {\n\tconst fsp = rawFs(repo);\n\tconst packDir = `${repo.gitdir}/objects/pack`;\n\tconst base = packBaseName(pack);\n\tawait fsp.mkdir(packDir, { recursive: true });\n\tawait fsp.writeFile(`${packDir}/${packFileName(base, \"pack\")}`, pack);\n\tconst { oids } = await git.indexPack({\n\t\t...repo,\n\t\tdir: packDir,\n\t\tfilepath: packFileName(base, \"pack\"),\n\t});\n\treturn { base, oids };\n}\n","/**\n * A FIFO semaphore for bounding in-flight I/O.\n *\n * An unbounded `Promise.all` over a list whose size the caller (or an\n * attacker) controls — a git tree with tens of thousands of blob entries, say,\n * each costing one object-store read — issues that many concurrent requests\n * from a single push or clone. Sockets and memory then scale with the input,\n * and S3-compatible providers rate-limit per account, so one hostile push\n * degrades every other tenant. Bounding the *in-flight* count caps resource\n * use at `limit` no matter how wide the input is, while still letting reads\n * overlap up to that limit instead of running strictly sequentially.\n */\nexport interface Semaphore {\n\trun<T>(fn: () => Promise<T>): Promise<T>;\n}\n\n/**\n * Create a FIFO semaphore admitting at most `limit` concurrent `run`\n * executions. Waiters are queued as promises and released in call order — no\n * polling or timers — so a caller blocked on a full semaphore is woken the\n * instant a permit frees, not on the next tick of a poll loop.\n */\nexport function createSemaphore(limit: number): Semaphore {\n\tif (!Number.isInteger(limit) || limit < 1) {\n\t\tthrow new Error(\n\t\t\t`createSemaphore: limit must be an integer >= 1, got ${limit}`,\n\t\t);\n\t}\n\n\tlet active = 0;\n\tconst waiting: Array<() => void> = [];\n\n\t/**\n\t * Hand the permit to the next waiter instead of dropping it, so `active`\n\t * only decreases when nothing is queued and never exceeds `limit`.\n\t */\n\tfunction release(): void {\n\t\tconst next = waiting.shift();\n\t\tif (next) next();\n\t\telse active--;\n\t}\n\n\tfunction acquire(): Promise<void> {\n\t\tif (active < limit) {\n\t\t\tactive++;\n\t\t\treturn Promise.resolve();\n\t\t}\n\t\treturn new Promise<void>((resolve) => {\n\t\t\twaiting.push(resolve);\n\t\t});\n\t}\n\n\treturn {\n\t\trun<T>(fn: () => Promise<T>): Promise<T> {\n\t\t\treturn acquire().then(async () => {\n\t\t\t\ttry {\n\t\t\t\t\treturn await fn();\n\t\t\t\t} finally {\n\t\t\t\t\trelease();\n\t\t\t\t}\n\t\t\t});\n\t\t},\n\t};\n}\n"],"mappings":";;;;;;;;;;;;;;;AAAA,OAAO,SAAS;AAiBT,SAAS,qBAAqB,MAAoB;AACxD,MAAI,CAAC,iBAAiB,IAAI,GAAG;AAC5B,UAAM,IAAI,uBAAuB,wBAAwB,IAAI,EAAE;AAAA,EAChE;AACD;AAEA,IAAM,cAAc;AAMpB,SAAS,uBACR,IACyC;AACzC,SACC,OAAO,OAAO,YACd,OAAO,QACP,0BAA0B,MAC1B,OAAQ,GAAoC,yBAC3C;AAEH;AAYA,SAAS,cAAc,IAAmC;AACzD,QAAM,WAAY,GAAiC;AACnD,SAAO,YAAY;AACpB;AAEA,SAAS,kBAAkB,SAAsC;AAChE,SAAO,OAAO,YAAY,WACvB,UACA,IAAI,YAAY,EAAE,OAAO,OAAO;AACpC;AASA,eAAe,kBAAkB,MAAsC;AACtE,SAAO,kBAAkB,MAAM,YAAY;AAC5C;AAMA,eAAsB,kBACrB,MACA,KAC2B;AAC3B,MAAI,CAAC,uBAAuB,KAAK,EAAE,EAAG,QAAO;AAC7C,QAAM,QAAQ,MAAM,KAAK,GAAG,qBAAqB,GAAG,KAAK,MAAM,IAAI,GAAG,EAAE;AACxE,SAAO,MAAM,OAAO,gBAAgB;AACrC;AAGA,eAAe,mBAAmB,MAA0C;AAC3E,MAAI,CAAC,uBAAuB,KAAK,EAAE,EAAG,QAAO,oBAAI,IAAI;AACrD,QAAM,aAAa,cAAc,KAAK,EAAE;AACxC,MAAI,CAAC,WAAY,QAAO,oBAAI,IAAI;AAChC,MAAI;AACH,UAAM,UAAU,MAAM,WAAW;AAAA,MAChC,GAAG,KAAK,MAAM;AAAA,MACd;AAAA,IACD;AACA,UAAM,OAAO,oBAAI,IAAoB;AACrC,eAAW,QAAQ,kBAAkB,OAAO,EAAE,MAAM,IAAI,GAAG;AAC1D,YAAM,QAAQ,sCAAsC,KAAK,IAAI;AAC7D,UAAI,QAAQ,CAAC,KAAK,MAAM,CAAC,KAAK,iBAAiB,MAAM,CAAC,CAAC,GAAG;AACzD,aAAK,IAAI,MAAM,CAAC,GAAG,MAAM,CAAC,CAAC;AAAA,MAC5B;AAAA,IACD;AACA,WAAO;AAAA,EACR,QAAQ;AACP,WAAO,oBAAI,IAAI;AAAA,EAChB;AACD;AAGA,eAAe,sBAAsB,MAAoC;AACxE,MAAI,CAAC,uBAAuB,KAAK,EAAE,EAAG,QAAO;AAC7C,QAAM,aAAa,cAAc,KAAK,EAAE;AACxC,MAAI,CAAC,WAAY,QAAO;AACxB,MAAI;AACH,UAAM,UAAU,MAAM,WAAW,SAAS,GAAG,KAAK,MAAM,SAAS,MAAM;AACvE,UAAM,OAAO,kBAAkB,OAAO,EAAE,KAAK;AAC7C,UAAM,QAAQ,2BAA2B,KAAK,IAAI;AAClD,WAAO,QAAQ,CAAC,KAAK,iBAAiB,MAAM,CAAC,CAAC,IAAI,MAAM,CAAC,IAAI;AAAA,EAC9D,QAAQ;AACP,WAAO;AAAA,EACR;AACD;AAaA,eAAsB,oBACrB,MACA,KACkB;AAClB,QAAM,aAAa,cAAc,KAAK,EAAE;AACxC,MAAI,YAAY;AACf,QAAI;AACH,YAAM,UAAU,MAAM,WAAW;AAAA,QAChC,GAAG,KAAK,MAAM,IAAI,GAAG;AAAA,QACrB;AAAA,MACD;AACA,YAAM,MAAM,kBAAkB,OAAO,EAAE,KAAK;AAC5C,UAAI,YAAY,KAAK,GAAG,EAAG,QAAO;AAAA,IACnC,QAAQ;AAAA,IAER;AAAA,EACD;AACA,SAAO,IAAI,WAAW,EAAE,GAAG,MAAM,IAAI,CAAC;AACvC;AAGA,eAAsB,aAAa,MAA+B;AACjE,MAAI;AACH,UAAM,CAAC,eAAe,oBAAoB,cAAc,IACvD,MAAM,QAAQ,IAAI;AAAA,MACjB,kBAAkB,IAAI;AAAA,MACtB,sBAAsB,IAAI;AAAA,MAC1B,mBAAmB,IAAI;AAAA,IACxB,CAAC;AACF,UAAM,WACL,kBAAkB,OACf,MAAM,IAAI,aAAa,IAAI,IAC3B,CAAC,GAAG,oBAAI,IAAI,CAAC,GAAG,eAAe,GAAG,eAAe,KAAK,CAAC,CAAC,CAAC,EAAE,KAAK;AACpE,UAAM,gBACL,kBAAkB,OACf,MAAM,IACL,cAAc,EAAE,GAAG,MAAM,UAAU,MAAM,CAAC,EAC1C,MAAM,MAAM,IAAI,IACjB;AAEJ,WAAO,QAAQ;AAAA,MACd,SAAS,IAAI,OAAO,WAAW;AAC9B,cAAM,eAAe,eAAe,IAAI,MAAM;AAC9C,eAAO;AAAA,UACN,MAAM;AAAA,UACN,QACC,gBAAgB,CAAC,eAAe,SAAS,MAAM,IAC5C,eACA,MAAM,oBAAoB,MAAM,cAAc,MAAM,EAAE;AAAA,UAC1D,WAAW,WAAW;AAAA,QACvB;AAAA,MACD,CAAC;AAAA,IACF;AAAA,EACD,SAAS,KAAc;AACtB,QAAK,IAA0B,SAAS,gBAAiB,QAAO,CAAC;AACjE,UAAM;AAAA,EACP;AACD;AAGA,eAAsB,iBACrB,MACA,MACA,YACgB;AAChB,uBAAqB,IAAI;AACzB,uBAAqB,UAAU;AAC/B,QAAM,SAAS,MAAM,oBAAoB,MAAM,cAAc,UAAU,EAAE;AACzE,MAAI;AACH,UAAM,IAAI,OAAO,EAAE,GAAG,MAAM,KAAK,MAAM,UAAU,OAAO,OAAO,CAAC;AAAA,EACjE,SAAS,KAAK;AACb,QAAK,IAA0B,SAAS,sBAAsB;AAC7D,YAAM,IAAI,iBAAiB,UAAU,IAAI,iBAAiB;AAAA,IAC3D;AACA,UAAM;AAAA,EACP;AACD;AAGA,eAAsB,mBACrB,MACA,MACgB;AAChB,uBAAqB,IAAI;AAGzB,QAAM,UACJ,MAAM,sBAAsB,IAAI,KAChC,MAAM,IAAI,cAAc,EAAE,GAAG,MAAM,UAAU,MAAM,CAAC,EAAE,MAAM,MAAM,IAAI;AACxE,MAAI,YAAY,MAAM;AACrB,UAAM,IAAI;AAAA,MACT,iBAAiB,IAAI;AAAA,IACtB;AAAA,EACD;AACA,QAAM,IAAI,aAAa,EAAE,GAAG,MAAM,KAAK,KAAK,CAAC;AAC9C;AAGA,eAAsB,mBACrB,MACA,MACgB;AAChB,uBAAqB,IAAI;AACzB,QAAM,oBAAoB,MAAM,cAAc,IAAI,EAAE;AACrD;;;ACxOA,OAAOA,UAAS;AAgBhB,IAAM,wBAAwD;AAAA,EAC7D,QAAQ;AAAA,EACR,MAAM;AAAA,EACN,MAAM;AAAA,EACN,KAAK;AACN;AAIA,SAAS,uBACR,MACA,QACa;AACb,QAAM,QAAkB,CAAC;AACzB,MAAI,OAAO,SAAS;AACpB,QAAM;AAAA,KACJ,OAAO,MAAa,KAAK,sBAAsB,IAAI,IAAK,SAAS;AAAA,EACnE;AACA,MAAI,OAAO,KAAK,MAAM,SAAS,EAAE;AACjC,SAAO,MAAM;AACZ,WAAO,OAAO;AACd,UAAM,MAAM,OAAO,MAAa,KAAM,OAAO,GAAW;AACxD,WAAO,KAAK,MAAM,OAAO,GAAG;AAAA,EAC7B;AACA,SAAO,IAAI,WAAW,KAAK;AAC5B;AAGO,SAAS,cACf,MACA,SACkB;AAClB,SAAO,KAAK,OAAO,WAAW,GAAG,IAAI,IAAI,QAAQ,MAAM,IAAI,GAAG,OAAO,CAAC;AACvE;AAQA,eAAsB,WACrB,SACmC;AACnC,QAAM,SAAS,IAAI,WAAW,EAAE;AAChC,QAAM,OAAO,IAAI,SAAS,OAAO,MAAM;AACvC,SAAO,IAAI,WAAW,MAAM,GAAG,CAAC;AAChC,OAAK,UAAU,GAAG,GAAG,KAAK;AAC1B,OAAK,UAAU,GAAG,QAAQ,QAAQ,KAAK;AAEvC,QAAM,SAAuB,CAAC,MAAM;AACpC,QAAM,aAAa;AACnB,WAAS,IAAI,GAAG,IAAI,QAAQ,QAAQ,KAAK,YAAY;AACpD,UAAM,UAAU,MAAM,QAAQ;AAAA,MAC7B,QACE,MAAM,GAAG,IAAI,UAAU,EACvB;AAAA,QAAI,OAAO,WACX;AAAA,UACC,uBAAuB,OAAO,MAAM,OAAO,QAAQ,MAAM;AAAA,UACzD,MAAM,QAAQ,OAAO,OAAO;AAAA,QAC7B;AAAA,MACD;AAAA,IACF;AACA,WAAO,KAAK,GAAG,OAAO;AAAA,EACvB;AAEA,QAAM,OAAO,OAAO,GAAG,MAAM;AAC7B,SAAO,OAAO,MAAM,QAAQ,MAAM,KAAK,IAAI,CAAC,CAAC;AAC9C;AAOO,SAAS,aAAa,MAA0B;AACtD,SAAO,QAAQ,MAAM,KAAK,SAAS,KAAK,SAAS,EAAE,CAAC,CAAC;AACtD;AAQA,eAAsB,UACrB,MACA,MAC4C;AAC5C,QAAM,MAAM,MAAM,IAAI;AACtB,QAAM,UAAU,GAAG,KAAK,MAAM;AAC9B,QAAM,OAAO,aAAa,IAAI;AAC9B,QAAM,IAAI,MAAM,SAAS,EAAE,WAAW,KAAK,CAAC;AAC5C,QAAM,IAAI,UAAU,GAAG,OAAO,IAAI,aAAa,MAAM,MAAM,CAAC,IAAI,IAAI;AACpE,QAAM,EAAE,KAAK,IAAI,MAAMC,KAAI,UAAU;AAAA,IACpC,GAAG;AAAA,IACH,KAAK;AAAA,IACL,UAAU,aAAa,MAAM,MAAM;AAAA,EACpC,CAAC;AACD,SAAO,EAAE,MAAM,KAAK;AACrB;;;ACrGO,SAAS,gBAAgB,OAA0B;AACzD,MAAI,CAAC,OAAO,UAAU,KAAK,KAAK,QAAQ,GAAG;AAC1C,UAAM,IAAI;AAAA,MACT,uDAAuD,KAAK;AAAA,IAC7D;AAAA,EACD;AAEA,MAAI,SAAS;AACb,QAAM,UAA6B,CAAC;AAMpC,WAAS,UAAgB;AACxB,UAAM,OAAO,QAAQ,MAAM;AAC3B,QAAI,KAAM,MAAK;AAAA,QACV;AAAA,EACN;AAEA,WAAS,UAAyB;AACjC,QAAI,SAAS,OAAO;AACnB;AACA,aAAO,QAAQ,QAAQ;AAAA,IACxB;AACA,WAAO,IAAI,QAAc,CAAC,YAAY;AACrC,cAAQ,KAAK,OAAO;AAAA,IACrB,CAAC;AAAA,EACF;AAEA,SAAO;AAAA,IACN,IAAO,IAAkC;AACxC,aAAO,QAAQ,EAAE,KAAK,YAAY;AACjC,YAAI;AACH,iBAAO,MAAM,GAAG;AAAA,QACjB,UAAE;AACD,kBAAQ;AAAA,QACT;AAAA,MACD,CAAC;AAAA,IACF;AAAA,EACD;AACD;","names":["git","git"]}