Standard .bz2 files, bounded memory, and caller-owned allocators and I/O — nothing hidden, nothing linked.
const compressed = try bzip2.compress(gpa, data, .{ .level = 9 });
const restored = try bzip2.decompress(gpa, compressed, .{});Both slices belong to the caller. Verification is on by default, so restored is either exactly data or the call fails.
BZip2 trades speed for ratio. Text, source code, logs, database dumps and uncompressed containers all compress considerably better with it than with a faster format, at the cost of taking longer. If throughput matters more than output size, a different format is the better tool.
Use the level to trade between the two. The level is the block size in units of 100 kB, so level 1 codes 100 kB blocks and level 9 codes 900 kB ones.
| Level 1 | Level 9 | |
|---|---|---|
| Block size | 100 kB | 900 kB |
| Compression | faster | slower |
| Ratio | slightly worse | slightly better |
Intermediate levels are often the right answer, because the block-sorting stage dominates the cost and the difference between the extremes is usually small.
Streams are validated structurally before use, and both the block and stream CRCs are verified before a result is returned. For anything fetched from outside the process, also cap the output:
const restored = try bzip2.decompress(gpa, downloaded, .{
.max_output_size = 64 * 1024 * 1024,
});A short stream can expand into a much larger one, and the cap turns that into error.OutputTooLarge instead of an allocation whose size came from the sender.
This is a from-scratch, native Zig implementation of the BZip2 format. The format specification and the reference implementation used for this project is libarchive/bzip2.