v0.7.12

add hash capacity

v0.7.11

• Fixed FASTQ tail trimming so quality lines retain valid @ characters, preventing accidental base masking.
• Enhanced interleaved FASTQ detection to recognize both /1 /2 and Illumina 1:N (line 0, column 1) style headers.

v0.7.10

This release fixes critical bugs in the FastaReader and BufferFastaReader that caused incorrect sequence parsing and data loss. Both readers now use a new, robust parsing engine that correctly handles multi-line sequences, comments, and whitespace. FASTA file reading is now stable, accurate, and reliable.

v0.7.8

• This release introduces a major database construction workflow improvement with the new add_library command, allowing users to safely add local FASTA files while automatically preventing duplicates.

• The database build commands have been refactored into build (for the full process) and build_db (for building hashes only) to make their intent clearer. Additionally, build_db commands now support a new -c flag, allowing advanced users to specify an exact hash table capacity to skip the time-consuming estimation step.

v0.7.7

fix(build): Resolve critical data loss and UB in parallel data processing

This commit addresses two critical, independent bugs that together caused
non-deterministic data loss and corrupted data reads during the
database build process.

1. Fixed Critical Data Loss in set_page_cell (The Logic Bug)

• Problem: The set_page_cell function had a fatal logic error in its
  loop exit condition (if result.is_ok() || idx == first_idx).
• Consequence: When par_iter caused a key conflict (a non-matching
  key in an occupied slot), the result was Err but idx == first_idx
  was true. The loop would immediately break, silently discarding
  the data instead of performing linear probing. This led to massive,
  random data loss depending on thread race conditions.
• Fix: The loop logic was rewritten to correctly match on the
  fetch_update result. It now only breaks on Ok (success) or after
  a full circular probe, and correctly increments idx on Err to
  perform linear probing as intended.

2. Fixed Data Corruption (UB) in read_first_block_from_file (The I/O Bug)

• Problem: The function previously allocated a Vec<u8> (1-byte
  alignment) and unsafely cast its pointer to *const u32 (4-byte
  alignment), causing Undefined Behavior (UB).
• Consequence: This alignment mismatch resulted in non-deterministic,
  corrupted data being read from the file, which was the root cause of
  seeing different values (e.g., 3995202615 vs 2133185527) at the
  same index on different runs.
• Fix: Replaced the implementation to first allocate a Vec<u32>
  (guaranteeing alignment) and then used bytemuck to safely cast to
  &mut [u8]. This enables high-performance, zero-copy reads directly
  into the buffer while being memory-safe.

These two fixes together ensure the database build is now correct,
deterministic, and robust against race conditions, fully resolving
the data loss and corruption issues.

v0.7.6

Add support for auto-detecting interleaved FASTQ

v0.7.5

multiple gzip members

v0.7.4

• input files

v0.7.3

• input files

v0.7.2

• jemalloc