June 26, 2026
Compressed files, expanded drama
Hellishly Slow Level 13 Deflate Compression
A tiny file-size win sparked big feelings over whether painfully slow saving is genius or madness
TLDR: A new file-compression mode makes files only a hair smaller, but can take over 50 times longer to do it. Commenters were torn between admiring the beautiful absurdity and asking the practical question: when does saving a tiny amount of space justify that much extra waiting?
A programmer just dropped a wildly slow new way to squeeze files a tiny bit smaller, and the internet instantly turned it into a debate about genius, obsession, and absolutely cursed trade-offs. The basic idea is simple: this new setting spends dramatically more time hunting for the “best” way to pack data, but the reward is microscopic — about 0.134% smaller files than the already-strong previous setting. In one benchmark, that meant saving roughly 87,000 bytes while taking 56 times longer. Yes, people absolutely noticed.
The comments split into two camps: the “I love this kind of nerd madness” crowd and the “please explain why anyone would ever do this” crowd. One fan cheered that computing is all about trade-offs and begged for the money math: when does an 8,000% increase in wait time actually pay off in storage or bandwidth bills? Another immediately raised the stakes by name-dropping Zopfli, ECT, Brotli, and LZMA — basically saying, if we’re being ridiculous, let’s be properly ridiculous.
Then came the side quests. People wanted to know about memory use, not just speed. One commenter marveled that after all these years there’s still room to improve this old file format at all, before swerving into a eyebrow-raising cultural hot take. And perhaps the funniest vibe came from the compression hobbyists who treated this whole saga like catnip: if you’re already testing bizarre file-packing combos and even neural compressors that take minutes, this isn’t madness — it’s content.
Key Points
- •DEFLATE level 13 is presented as a standard-compatible libdeflate encoder mode that trades much more compression time for slightly smaller output.
- •The implementation keeps libdeflate’s near-optimal parser but expands search work with a full 32 KiB window, up to 15 optimization passes, and static Huffman optimization on blocks up to 50,000 bytes.
- •For text-like data, the encoder may raise the soft block size from 300,000 to 1,000,000 bytes after sampling up to 64 KiB and checking for no NULL byte and at most 97 distinct byte values.
- •Development followed a zero compression regression policy on the Silesia corpus, retaining only changes that reduced at least one file without enlarging any other compressed file.
- •On the Silesia benchmark, level 13 reduced output by 86,990 bytes or 0.134% versus level 12, while taking 56.4 times longer to compress overall.