The bzip3 thread turns into a dispute over compression benchmark window sizes

Sources: repository, discussion
Summary: A Hacker News thread on the bzip3 compressor moved from the project's headline comparison to the window sizes the comparison was run at.
Comments: An HN commenter reports that matching zstd's window to bzip3's 512MB block with --long=29 produces output more than twice smaller than bzip3. The same comment gives zstd -T8 -16 at 2054.50 seconds user time and the same command with --long=29 at 1083.06 seconds, so the CPU-time halving is measured against zstd's own default window and not against bzip3, which the thread does not time.
Why it matters: It reframes the project's published size advantage as a configuration artifact, which is the usual failure mode of compressor benchmarks.