Decompression pipeline

decompress.c reverses the compression pipeline. Entry point is decompress_msz().

Phases

1. Read footer (file.c::read_footer) — the last 88 bytes of the file give stream offsets, the divisions offset, spectrum count, and source data formats.
2. Memory-map — the whole MSZ is mmap'd. Stream blocks are read directly from the mapping; no input copy.
3. Plan — set_decompress_runtime_variables() resolves (algorithm, accession) to decoder function pointers and stores them on data_format_t.
4. Decompress — decompress_parallel() spawns N threads of decompress_routine(). Each thread:
   • Allocates its own ZSTD_DCtx.
   • Iterates blocks in its division.
   • ZSTD decompress → optional base64/zlib decode → algorithm decoder → original-typed array → emit mzML XML wrapping the array.
5. Concatenate — output is written in original spectrum order.

Block caching

block_len_t carries a cache field. During selective extraction, a block that's already been decompressed isn't decompressed a second time:

if (!blk->cache) {
    blk->cache = zstd_decompress(blk->mem, blk->compressed_size, ...);
}

This matters for extraction patterns that touch the same block multiple times (e.g., scan-number filters that span block boundaries). See extract.c.

Entry points

Function	What it does
decompress_msz()	Top-level entry
decompress_parallel()	Spawns and joins workers
decompress_routine()	Per-thread worker
read_footer()	Parses the 88-byte footer (in file.c)