v1.4.3

CHDSharp v1.4.3

Consolidates all changes since v1.4.2. Closes the partial-tail raw-encode corruption and the last FLAC byte-parity divergence. Battle harness green: 3174/3174 synthetic vs MAME 0.289.

Partial-tail raw-encode corruption fix

EncodeRaw/createraw on inputs of ≥ 513 hunks whose length is not a hunk multiple could produce a self-consistent-but-wrong CHD: stale compression ring-buffer slots from the previous fill cycle leaked into partial final hunks. The fix landed alongside guardrail tests so the class of bug cannot silently return:

• a synthetic partial-tail probe that pins the exact stale-slot bytes,
• partial-last-hunk unit tests per codec,
• a resized long-tail case in the battle harness (raw-encode long-tail x flac/zlib/zstd).

FLAC encoder byte parity

The vendored FLAC encoder now matches chdman byte-for-byte on the remaining divergence: the fixed-predictor analysis window and the tukey window's cosf rounding (native cosf semantics replicated). Verified byte-identical on:

• the 7.8 GB Perfect Dark Zero XGD2 image (1,912,816 hunks, chdman default codec list),
• a 100 MB partial-tail slice of the same source,
• three single-hunk repro cases.

Byte-parity fixes earlier in the cycle

• Stale work-buffer tail + LZMA match-finder insert parity (raw encode).

Diagnostics and output

• Detailed decompression-failure messages: hunk index, codec, CRC/truncation reasons.
• chdman-style progress + speed-meter output.

Verification status

• CHDSharpTest: 12,648/12,648 (436/436 per TFM on net8.0/net9.0/net10.0 → 1,308/1,308 aggregate via the console runners).
• CHDSharpEncoderTest: 1,308/1,308.
• CHDSharpBattleTest: 3174/3174, including raw-encode long-tail x * for all three codecs.

NuGet Package

dotnet add package CHDSharp --version 1.4.3

Targets net8.0, net9.0, net10.0. Pure C# - zero native dependencies.

Full Changelog: v1.4.2...v1.4.3