Use the buffer limit for the safe decompress(ByteBuffer, int, ByteBuffer, int)

[yawkat]

With a safe decompressor, LZ4DecompressorWithLength.decompress(ByteBuffer src, int srcOff, ByteBuffer dest, int destOff) delegated with srcLen = src.remaining() - srcOff. That mixes the buffer's position with an absolute offset. The method uses absolute indexing and is documented to leave positions unchanged, so the result should not depend on src.position(), but it did. Valid data at srcOff = 3 decompressed with src.position() == 0. With src.position() == 3, it failed with LZ4Exception: Malformed input. With larger positions, the computed length could even be negative.

This change uses src.limit() - srcOff instead and corrects the javadoc from src[srcOff:src.remaining()] to src[srcOff:src.limit()]. I checked the sibling methods:

• decompress(ByteBuffer, ByteBuffer) correctly uses src.position() together with src.remaining().
• The array overloads use src.length - srcOff.
• The other absolute-offset overloads take srcLen explicitly.

Found by fuzzing with micronaut-fuzzing.

Behaviour change: callers that set src.position() to something other than 0 get correct results now, where before they got a spurious failure. A call with position 0 behaves exactly as before.

Test: I added OutOfBoundsTest.byteBufferSourceOffsetIgnoresPosition. It decompresses data at srcOff = 3 with source positions 0, 1, 3 and limit, for fast and safe decompressors and for heap and direct buffers. It also checks that positions are unchanged. Without the fix, the safe case fails with Malformed input. With the fix, OutOfBoundsTest and LZ4Test pass.

This builds on #160 (short input, already merged) and merges cleanly with it.

🤖 Generated with Claude Code