{"entries":[{"lastCommit":{"author":{"email":"git@mary.my.id","name":"Mary","when":"2026-05-28T12:30:27.000000+07:00"},"committer":{"email":"git@mary.my.id","name":"Mary","when":"2026-05-28T12:30:27.000000+07:00"},"hash":"beaa3e1eb66408fa5415e0814b38a50b04a95ce3","message":"feat: AES-256 + SHA-256 KDF for password-protected entries\n\ndecodes the 0x06f10701 coder. the password is encoded UTF-16LE,\nhashed against the coder's salt + an 8-byte LE counter for\n2^numCyclesPower SHA-256 rounds, and the digest becomes the AES-256\nkey. the IV (up to 16 bytes) sits alongside the salt in the coder\nproperties block.\n\nCBC decryption is driven by Web Crypto via a small trick: 7z does\nnot PKCS#7-pad ciphertext, so we append one synthetic 16-byte block\nwhose plaintext is 0x10*16 (a valid \"one full block of padding\"\ntrailer). Web Crypto's AES-CBC decrypts the extended stream, strips\nthe synthetic trailer, and hands back exactly the original\nplaintext bytes. the synthetic block is computed with one\nAES-CTR pass (the standard Web Crypto path that exposes ECB\nencryption); no key-and-tables port required.\n\npassword is plumbed through unsevenzip(reader, { password }) all\nthe way down — including the readContainer pass so EncodedHeader\nchains with AES (encryption_header.7z) work. covered by libarchive's\nencryption.7z, encryption_header.7z, encryption_partially.7z (all\nuse \"12345678\").\n","tree":"2b3a323b5edcb3356258cbf7019d415ea36e5b89"},"mode":"0100644","name":"aes.ts","size":8989},{"lastCommit":{"author":{"email":"git@mary.my.id","name":"Mary","when":"2026-05-28T12:49:41.000000+07:00"},"committer":{"email":"git@mary.my.id","name":"Mary","when":"2026-05-28T12:49:41.000000+07:00"},"hash":"f4acdac85d0189313af6daba0209eaf1d68f2356","message":"feat: add IA-64 BCJ filter (and cover ARM-Thumb)\n\nports the Intel IA-64 BCJ filter from lzma-rust2/src/filter/bcj/ia64.rs.\neach 16-byte instruction bundle carries a 5-bit template byte and three\n41-bit instruction slots; the per-template branch table picks which\nslots to rewrite. uses BigInt for the 48-bit field extraction since\nJS bitwise can't cover that span.\n\npy7zr's lzma2_bcj_ia64 and lzma2_bcj_armt fixtures are now covered;\nARM-Thumb's decoder shipped in an earlier commit but had no test\nbecause libarchive's corpus lacks one.\n","tree":"00deddccb4dbd8a698e1fa46a94e87eed3897c83"},"mode":"0100644","name":"bcj.ts","size":17848},{"lastCommit":{"author":{"email":"git@mary.my.id","name":"Mary","when":"2026-05-28T13:31:20.000000+07:00"},"committer":{"email":"git@mary.my.id","name":"Mary","when":"2026-05-28T13:31:20.000000+07:00"},"hash":"ce5ee10e97c4291ee734fd9f8ead4d4159ec846b","message":"refactor: normalize error messages\n\nput each message in `:

` form so the generic part never\ncarries interpolated values, and unify wording across the codebase:\nfinal-value mismatches use `expected N, got M`, partial-progress stalls\nuse `produced N of M bytes`, and NID/byte checks fold the expected\nvalue into the name.\n","tree":"48fe5f64d106c2475e857407d0cf30ff8676cc25"},"mode":"0100644","name":"delta.ts","size":1406},{"lastCommit":{"author":{"email":"git@mary.my.id","name":"Mary","when":"2026-05-28T14:09:55.000000+07:00"},"committer":{"email":"git@mary.my.id","name":"Mary","when":"2026-05-28T14:09:55.000000+07:00"},"hash":"e85356c4febc40ae4a73cd2c8dace0303664e0ff","message":"fix: decode LZMA1 streams safely across chunk refills\n\ntwo interacting bugs corrupted any LZMA1 stream that didn't fit in the\nStreamReader's initial buffer (LZMA2 was unaffected because each chunk is\nfully buffered before decode).\n\n1. `LzmaDecoder.decode` wrapped each decode step in try/catch on a\n RangeUnderflow exception, snapshotting range/code/reps/state to roll\n back. but every `decodeBit` call also mutates the probability tables\n (`probs[idx] = ...`), and the snapshot never restored those. on retry\n the decoder ran with already-shifted probabilities, decoded the wrong\n bit, and produced bogus distances downstream — manifesting as\n `LZMA distance overflow: dist= full=`.\n\n2. `decodeLzmaStream` always passed `exhausted=true` when refilling: it\n checked `input.buffered === 0` *after* draining via take(), which is\n trivially true. that masked the upstream's true end-of-stream signal,\n and after the snapshot rework the decoder treated every refill as the\n last one — reading zero padding past the live buffer and again\n producing corrupt output.\n\nreplaces the snapshot+throw pattern with a pre-buffering watermark:\n`RangeDecoder.canStep` reports whether enough bytes are buffered for any\nsingle decode step (worst-case 48 bytes derived from the longest match\npath). `decode` returns false when below the watermark *and* the upstream\nisn't yet exhausted. on real end-of-stream, `#readByte` returns zero\npadding (matching the LZMA encoder convention) so the final output bits\nstill decode. the caller's refill loop now passes `input.done` so the\nexhaustion signal is honest.\n\n`ByteSource` gains a `done` readonly so both the StreamReader and the\ninter-coder IteratorByteSource expose this signal uniformly. regression\ntest fragments a real LZMA1 fixture into 7-byte chunks to drive the\nmulti-feed path that the prior tests never exercised.\n","tree":"4181022492f05759442426f44989ea362308f436"},"mode":"0100644","name":"folder.ts","size":13252},{"lastCommit":{"author":{"email":"git@mary.my.id","name":"Mary","when":"2026-05-29T08:08:09.000000+07:00"},"committer":{"email":"git@mary.my.id","name":"Mary","when":"2026-05-29T08:08:09.000000+07:00"},"hash":"c3e9b212fba2509472d5bb96644685360c98d253","message":"refactor: fuse isMatch bit and literal decode for the hot path\n\nthe dominant decode step for literal-heavy payloads is \"a plain literal\nfollows a literal\" (state < LIT_STATES). previously this ran as two\nRangeDecoder method calls — decodeBit(isMatch) then decodeLiteralSymbol —\neach loading this.range/this.code at entry and storing them at exit, so\nthe range-coder state round-tripped through fields between them.\n\nadd decodeIsMatchThenLiteralSymbol, which decodes the isMatch bit and (if\nit is a literal) flows straight into the branchless literal loop with\nrange/code held in locals throughout. decode() takes this fused path only\nfor state < LIT_STATES and inlines the stateAfterLiteral transition there;\nthe match-byte literal path (state >= LIT_STATES) and all match/rep\ndecoding are unchanged.\n\nmeasured on the manga .7z corpus: aggregate ~21.1 -> ~22.4 MiB/s (+6%),\nroughly +27% over the pre-optimization baseline. no new hot-path deopts.\n","tree":"035498c6f75c7dff589e76c3437acecb8f7ba264"},"mode":"0100644","name":"lzma.ts","size":35637}],"lastCommit":{"author":{"email":"git@mary.my.id","name":"Mary","when":"2026-05-29T08:08:09.000000+07:00"},"committer":{"email":"git@mary.my.id","name":"Mary","when":"2026-05-29T08:08:09.000000+07:00"},"hash":"c3e9b212fba2509472d5bb96644685360c98d253","message":"refactor: fuse isMatch bit and literal decode for the hot path\n\nthe dominant decode step for literal-heavy payloads is \"a plain literal\nfollows a literal\" (state < LIT_STATES). previously this ran as two\nRangeDecoder method calls — decodeBit(isMatch) then decodeLiteralSymbol —\neach loading this.range/this.code at entry and storing them at exit, so\nthe range-coder state round-tripped through fields between them.\n\nadd decodeIsMatchThenLiteralSymbol, which decodes the isMatch bit and (if\nit is a literal) flows straight into the branchless literal loop with\nrange/code held in locals throughout. decode() takes this fused path only\nfor state < LIT_STATES and inlines the stateAfterLiteral transition there;\nthe match-byte literal path (state >= LIT_STATES) and all match/rep\ndecoding are unchanged.\n\nmeasured on the manga .7z corpus: aggregate ~21.1 -> ~22.4 MiB/s (+6%),\nroughly +27% over the pre-optimization baseline. no new hot-path deopts.\n","tree":"035498c6f75c7dff589e76c3437acecb8f7ba264"}}