new git work at @OpenAI. packfile-uri concurrency fixes now in upstream. 20 more patches in openai/git’s status work (still only our fork, but stay tuned), + better partial-clone maintenance and tracing.
openai-git-upstream.openai.c…
& github.com/openai/git
see you at git merge
contribution graph was a nightmare to maintain bc it needed spinning disks; & invalidating its redis cache on push was a punitive exercise in systems design. socially, it mixed up impact with volume.
but it is helping insurance marketing tho now; so who is to say if good or bad!
happy friday we've continued our work on git here at @openai.
performance, correctness, & testing improvements are flowing from openai/git to upstream. perf improvements in codex app too; you'll see more efficient use of git in new builds.
follow along: openai-git-upstream.openai.c…
the fact git has lasted this long (20+ years) is bonkers. statistically, something you wouldn’t expect to see in your lifetime. the messi of software.
a testament to good engineering practices. an extremely careful set of maintainers
& most of all: the C programming language.
salvatore points the right way, as usual. we are in a transitional era.
in an era like this, things will be weird! half-done things. the twilight of an era. this has happened before, in the early renaissance.
There are projects where I don't look anymore at the code at all, but only control the design ideas and the "product" itself (DS4). There are other stuff where I check every line and modify (mostly saying how and not myself, sometimes myself) every detail, like for Redis Arrays. Let's say that for 80% of the projects to check the code no longer makes sense, but checking the ideas / algorithms / code structure even only by prompting is, I believe, a good idea. While providing the right design decisions is not just a good idea, it is essential to do great work.









