File
Blob: src/worker/git/pack/indexer/writeIdx.ts
| 1 | /** |
| 2 | * Binary writer for Git pack-index v2. |
| 3 | * |
| 4 | * Produces a standard `.idx` file that can be verified byte-for-byte against |
| 5 | * `git index-pack` output. The format is fully deterministic given the same |
| 6 | * pack contents. |
| 7 | * |
| 8 | * Layout (all multi-byte integers are big-endian): |
| 9 | * [4] magic 0xff744f63 |
| 10 | * [4] version 0x00000002 |
| 11 | * [1024] fanout 256 x uint32 cumulative OID counts |
| 12 | * [N*20] oid table sorted raw 20-byte SHA-1 hashes |
| 13 | * [N*4] crc32 table per-object CRC-32 in OID-sorted order |
| 14 | * [N*4] offset table uint32, MSB set -> index into 64-bit table |
| 15 | * [var] 64-bit table 8-byte entries for offsets >= 0x80000000 |
| 16 | * [20] pack SHA-1 trailing checksum copied from the pack |
| 17 | * [20] idx SHA-1 SHA-1 of everything above |
| 18 | */ |
| 19 | |
| 20 | import type { PackEntryTable } from "./types"; |
| 21 | import { asBufferSource } from "@/worker/common/webtypes"; |
| 22 | |
| 23 | /** Compare two raw 20-byte OIDs located at positions a and b in a flat buffer. */ |
| 24 | function compareOids(oids: Uint8Array, a: number, b: number): number { |
| 25 | for (let i = 0; i < 20; i++) { |
| 26 | const diff = oids[a + i] - oids[b + i]; |
| 27 | if (diff !== 0) return diff; |
| 28 | } |
| 29 | return 0; |
| 30 | } |
| 31 | |
| 32 | export function buildOidSortedEntryIndices( |
| 33 | table: PackEntryTable, |
| 34 | objectCount: number |
| 35 | ): Uint32Array { |
| 36 | const sortedIndices = new Uint32Array(objectCount); |
| 37 | for (let i = 0; i < objectCount; i++) sortedIndices[i] = i; |
| 38 | // Duplicate OIDs are invalid in normal packs but can appear in defensive |
| 39 | // tests and malformed inputs. Tie-break by pack entry index so every derived |
| 40 | // artifact writes duplicate rows in the same deterministic order. |
| 41 | sortedIndices.sort((a, b) => compareOids(table.oids, a * 20, b * 20) || a - b); |
| 42 | return sortedIndices; |
| 43 | } |
| 44 | |
| 45 | export async function writeIdxV2( |
| 46 | table: PackEntryTable, |
| 47 | objectCount: number, |
| 48 | packChecksum: Uint8Array |
| 49 | ): Promise<Uint8Array> { |
| 50 | const N = objectCount; |
| 51 | |
| 52 | // 1. Build sorted index by OID (raw 20-byte comparison). |
| 53 | const sortedIndices = buildOidSortedEntryIndices(table, N); |
| 54 | |
| 55 | // 2. Count large offsets (>= 0x80000000). |
| 56 | let largeCount = 0; |
| 57 | for (let i = 0; i < N; i++) { |
| 58 | if (table.offsets[i] >= 0x80000000) largeCount++; |
| 59 | } |
| 60 | |
| 61 | // 3. Calculate total size. |
| 62 | const totalSize = |
| 63 | 4 + // magic |
| 64 | 4 + // version |
| 65 | 256 * 4 + // fanout |
| 66 | N * 20 + // OID table |
| 67 | N * 4 + // CRC-32 table |
| 68 | N * 4 + // 32-bit offset table |
| 69 | largeCount * 8 + // 64-bit offset table |
| 70 | 20 + // pack checksum |
| 71 | 20; // idx checksum |
| 72 | |
| 73 | const buf = new Uint8Array(totalSize); |
| 74 | const dv = new DataView(buf.buffer); |
| 75 | let pos = 0; |
| 76 | |
| 77 | // 4. Magic + version. |
| 78 | buf[pos++] = 0xff; |
| 79 | buf[pos++] = 0x74; |
| 80 | buf[pos++] = 0x4f; |
| 81 | buf[pos++] = 0x63; |
| 82 | dv.setUint32(pos, 2, false); |
| 83 | pos += 4; |
| 84 | |
| 85 | // 5. Fanout table. |
| 86 | const fanoutStart = pos; |
| 87 | // Fill with zeros first; we'll accumulate below. |
| 88 | pos += 256 * 4; |
| 89 | |
| 90 | // Compute fanout: for each first-byte bucket, count how many sorted OIDs |
| 91 | // have a first byte <= that value. |
| 92 | const buckets = new Uint32Array(256); |
| 93 | for (let i = 0; i < N; i++) { |
| 94 | const idx = sortedIndices[i]; |
| 95 | const firstByte = table.oids[idx * 20]; |
| 96 | buckets[firstByte]++; |
| 97 | } |
| 98 | let cumulative = 0; |
| 99 | for (let b = 0; b < 256; b++) { |
| 100 | cumulative += buckets[b]; |
| 101 | dv.setUint32(fanoutStart + b * 4, cumulative, false); |
| 102 | } |
| 103 | |
| 104 | // 6. OID table (sorted). |
| 105 | for (let i = 0; i < N; i++) { |
| 106 | const idx = sortedIndices[i]; |
| 107 | buf.set(table.oids.subarray(idx * 20, idx * 20 + 20), pos); |
| 108 | pos += 20; |
| 109 | } |
| 110 | |
| 111 | // 7. CRC-32 table (in OID-sorted order). |
| 112 | for (let i = 0; i < N; i++) { |
| 113 | const idx = sortedIndices[i]; |
| 114 | dv.setUint32(pos, table.crc32s[idx], false); |
| 115 | pos += 4; |
| 116 | } |
| 117 | |
| 118 | // 8. 32-bit offset table. |
| 119 | const offsets32Start = pos; |
| 120 | pos += N * 4; |
| 121 | |
| 122 | // 9. 64-bit offset table (only for offsets >= 0x80000000). |
| 123 | const offsets64Start = pos; |
| 124 | let largeIdx = 0; |
| 125 | for (let i = 0; i < N; i++) { |
| 126 | const idx = sortedIndices[i]; |
| 127 | const off = table.offsets[idx]; |
| 128 | if (off >= 0x80000000) { |
| 129 | // Mark MSB in the 32-bit table entry to point into the 64-bit table. |
| 130 | dv.setUint32(offsets32Start + i * 4, 0x80000000 | largeIdx, false); |
| 131 | // Write 64-bit offset as two uint32. |
| 132 | dv.setUint32(offsets64Start + largeIdx * 8, 0, false); // high 32 bits (always 0 for < 4GB) |
| 133 | dv.setUint32(offsets64Start + largeIdx * 8 + 4, off, false); |
| 134 | largeIdx++; |
| 135 | } else { |
| 136 | dv.setUint32(offsets32Start + i * 4, off, false); |
| 137 | } |
| 138 | } |
| 139 | pos = offsets64Start + largeCount * 8; |
| 140 | |
| 141 | // 10. Pack checksum. |
| 142 | buf.set(packChecksum, pos); |
| 143 | pos += 20; |
| 144 | |
| 145 | // 11. Idx checksum: SHA-1 of everything before this field. |
| 146 | const idxHash = new Uint8Array( |
| 147 | await crypto.subtle.digest("SHA-1", asBufferSource(buf.subarray(0, pos))) |
| 148 | ); |
| 149 | buf.set(idxHash, pos); |
| 150 | |
| 151 | return buf; |
| 152 | } |