File
Blob: src/worker/git/object-store/candidates.ts
| 1 | import type { IdxView } from "./types"; |
| 2 | |
| 3 | import { bytesToHex, hexToBytes } from "@/worker/common/hex"; |
| 4 | import { findOidIndexFromBytes, getNextOffsetByIndex, getOidHexAt } from "./idxView"; |
| 5 | |
| 6 | export type IndexedPackSource = { |
| 7 | packKey: string; |
| 8 | packBytes: number; |
| 9 | idx: IdxView; |
| 10 | }; |
| 11 | |
| 12 | export type PackedObjectCandidate = { |
| 13 | source: IndexedPackSource; |
| 14 | packSlot: number; |
| 15 | objectIndex: number; |
| 16 | offset: number; |
| 17 | nextOffset: number; |
| 18 | oid: string; |
| 19 | }; |
| 20 | |
| 21 | export type OidRun = { |
| 22 | /** Inclusive first `.idx` row whose OID matches the lookup OID. */ |
| 23 | startIndex: number; |
| 24 | /** Inclusive final `.idx` row whose OID matches the lookup OID. */ |
| 25 | endIndex: number; |
| 26 | }; |
| 27 | |
| 28 | export type CandidateCollectionOptions = { |
| 29 | excludePackKey?: string; |
| 30 | }; |
| 31 | |
| 32 | type NormalizedOid = { |
| 33 | oidHex: string; |
| 34 | oidBytes: Uint8Array; |
| 35 | }; |
| 36 | |
| 37 | function normalizeOid(oid: string | Uint8Array): NormalizedOid | undefined { |
| 38 | if (typeof oid === "string") { |
| 39 | const oidHex = oid.toLowerCase(); |
| 40 | if (oidHex.length !== 40) return undefined; |
| 41 | return { oidHex, oidBytes: hexToBytes(oidHex) }; |
| 42 | } |
| 43 | |
| 44 | if (oid.byteLength !== 20) return undefined; |
| 45 | return { oidHex: bytesToHex(oid), oidBytes: oid }; |
| 46 | } |
| 47 | |
| 48 | function oidAtIndexMatches(rawNames: Uint8Array, index: number, oidBytes: Uint8Array): boolean { |
| 49 | const rawStart = index * 20; |
| 50 | for (let offset = 0; offset < 20; offset++) { |
| 51 | if (rawNames[rawStart + offset] !== oidBytes[offset]) return false; |
| 52 | } |
| 53 | return true; |
| 54 | } |
| 55 | |
| 56 | function candidateFromIndex( |
| 57 | source: IndexedPackSource, |
| 58 | packSlot: number, |
| 59 | objectIndex: number, |
| 60 | oidHex: string |
| 61 | ): PackedObjectCandidate | undefined { |
| 62 | const nextOffset = getNextOffsetByIndex(source.idx, objectIndex); |
| 63 | if (nextOffset === undefined) return undefined; |
| 64 | |
| 65 | return { |
| 66 | source, |
| 67 | packSlot, |
| 68 | objectIndex, |
| 69 | offset: source.idx.offsets[objectIndex], |
| 70 | nextOffset, |
| 71 | oid: oidHex, |
| 72 | }; |
| 73 | } |
| 74 | |
| 75 | /** |
| 76 | * Locate the full contiguous duplicate-OID run in an idx view. |
| 77 | * |
| 78 | * Git idx files are sorted by object ID, so duplicate rows for the same OID |
| 79 | * are adjacent. The binary-search hit can land anywhere inside that run; this |
| 80 | * helper expands both directions so callers do not accidentally inspect only |
| 81 | * one arbitrary duplicate. |
| 82 | */ |
| 83 | export function findOidRunInIdx(idx: IdxView, oid: string | Uint8Array): OidRun | undefined { |
| 84 | const normalized = normalizeOid(oid); |
| 85 | if (!normalized) return undefined; |
| 86 | |
| 87 | const hitIndex = findOidIndexFromBytes(idx, normalized.oidBytes); |
| 88 | if (hitIndex < 0) return undefined; |
| 89 | |
| 90 | let startIndex = hitIndex; |
| 91 | while (startIndex > 0 && oidAtIndexMatches(idx.rawNames, startIndex - 1, normalized.oidBytes)) { |
| 92 | startIndex--; |
| 93 | } |
| 94 | |
| 95 | let endIndex = hitIndex; |
| 96 | while ( |
| 97 | endIndex + 1 < idx.count && |
| 98 | oidAtIndexMatches(idx.rawNames, endIndex + 1, normalized.oidBytes) |
| 99 | ) { |
| 100 | endIndex++; |
| 101 | } |
| 102 | |
| 103 | return { startIndex, endIndex }; |
| 104 | } |
| 105 | |
| 106 | /** |
| 107 | * Enumerate every packed-object candidate for an OID in snapshot order. |
| 108 | * |
| 109 | * Within a pack this returns the entire duplicate run in idx row order. The |
| 110 | * caller owns higher-level policy such as whether to accept the first material |
| 111 | * object, prefer full objects, or skip a target pack during refs-only backfill. |
| 112 | */ |
| 113 | export function collectPackedObjectCandidates( |
| 114 | sources: readonly IndexedPackSource[], |
| 115 | oid: string | Uint8Array, |
| 116 | options: CandidateCollectionOptions = {} |
| 117 | ): PackedObjectCandidate[] { |
| 118 | const normalized = normalizeOid(oid); |
| 119 | if (!normalized) return []; |
| 120 | |
| 121 | const candidates: PackedObjectCandidate[] = []; |
| 122 | for (let packSlot = 0; packSlot < sources.length; packSlot++) { |
| 123 | const source = sources[packSlot]!; |
| 124 | if (source.packKey === options.excludePackKey) continue; |
| 125 | |
| 126 | const run = findOidRunInIdx(source.idx, normalized.oidBytes); |
| 127 | if (!run) continue; |
| 128 | |
| 129 | for (let objectIndex = run.startIndex; objectIndex <= run.endIndex; objectIndex++) { |
| 130 | const candidate = candidateFromIndex(source, packSlot, objectIndex, normalized.oidHex); |
| 131 | if (candidate) candidates.push(candidate); |
| 132 | } |
| 133 | } |
| 134 | |
| 135 | return candidates; |
| 136 | } |
| 137 | |
| 138 | /** |
| 139 | * Find the first snapshot pack that contains an OID, preserving the existing |
| 140 | * binary-search hit semantics inside that pack. |
| 141 | */ |
| 142 | export function findFirstPackedObjectCandidate( |
| 143 | sources: readonly IndexedPackSource[], |
| 144 | oid: string | Uint8Array, |
| 145 | options: CandidateCollectionOptions = {} |
| 146 | ): PackedObjectCandidate | undefined { |
| 147 | const normalized = normalizeOid(oid); |
| 148 | if (!normalized) return undefined; |
| 149 | |
| 150 | for (let packSlot = 0; packSlot < sources.length; packSlot++) { |
| 151 | const source = sources[packSlot]!; |
| 152 | if (source.packKey === options.excludePackKey) continue; |
| 153 | |
| 154 | const objectIndex = findOidIndexFromBytes(source.idx, normalized.oidBytes); |
| 155 | if (objectIndex < 0) continue; |
| 156 | |
| 157 | return candidateFromIndex(source, packSlot, objectIndex, getOidHexAt(source.idx, objectIndex)); |
| 158 | } |
| 159 | |
| 160 | return undefined; |
| 161 | } |