Small clarification:
A log date is not just text.
It is a timestamp wearing a format.
Metarc separates the instant from its textual representation, encodes that structure, then lets zstd compress the transformed stream.
This morning, I compressed time.
Not metaphorically.
In logs, dates look like plain text:
2026-05-11 10:36:19
But they are actually : instant + format
Metarc compress this inner structure before zstd even sees the bytes.
Result on Loghub: up to 26.5% smaller than tar+zstd.
Want to try it?
→ Star the repo if you find it interesting
→ Create issues (Bug, Comments, Idea)
→ Test it on your own projects
→ Tell me your results
https://t.co/6TIDb8Ha1B
Looking for real examples of repos that got messy after using AI (Lovable, Claude, Copilot, GPT, etc.).
If you have/know one where things became harder to maintain over time, drop a link 👇
Even small or partial repos are useful.
Metarc tries to compress structure before bytes.
I rewrote a 2016 compression idea in Go, and early results are promising:
much faster on real repos, with sizes often close to tar.gz.
It’s also turning into a playground for meta-compression ideas.
https://t.co/Zi2WIO9Cct
A pod can die OOMKilled without the neat “classic memory leak” curve people expect.
I wrote up a recent Go production case where the real issue was not just memory growth.
Link in reply.
#golang