| 2 Aug 2025 |
 Charles | not yet, sorry, i'll try to get to it today | 15:24:30 |
 raitobezarius | In reply to @emilazy:matrix.org
the bootstrap was successful and it seems to eliminate the majority of build directory references in built binaries. haven't yet got to testing Rust and there are things like Python and other parts of LLVM that hardcode random build directory paths regardless that will need fixing up. they are already non-reproducible in the presence of concurrency anyway though, I do not think further Nixpkgs fixes on that front need to block things So you confirm this doesn't suddenly explode nixpkgs though | 15:24:54 |
| raitobezarius | which is great to hear | 15:24:56 |
 emily | especially since I don't know if they will get backported to 25.05, so just accepting that it will expose more reproducibility issues in Nixpkgs is part of the tradeoff regardless | 15:24:57 |
| emily | yes, I'm running the stack and building lots of stuff | 15:25:13 |
| emily | including https://gerrit.lix.systems/c/lix/+/3852/5 | 15:25:28 |
| raitobezarius | sandbox = true in your case right? | 15:25:36 |
| emily | I was a little nervous the ENOTSUP may not cover everything in practice but so far it seems to | 15:25:40 |
| emily | yeah | 15:25:44 |
| raitobezarius | very cool | 15:25:52 |
| raitobezarius | thank you so much for the chain! | 15:26:11 |
| emily | I think the fixes will not substantially change the shape of the chain, so it's probably reviewable as-is, but I believe there are still one or two review comments I did not address anyway | 15:27:32 |
| raitobezarius | I rebased it and reran CI | 15:27:52 |
| emily | I think a last-minute syntax error or something snuck in so CI results might not be that helpful past a certain point in the chain right now. I'll verify that it passes tests on Linux and Darwin and works on my machine before pushing out revisions though | 15:28:53 |
| emily | I dropped it to fix CI once it became clear I had broken that | 15:29:12 |
| raitobezarius | Okie dokie | 15:29:18 |
| raitobezarius | emily did a pass and +2ed everything I could reasonably do | 15:45:51 |
| raitobezarius | I will leave the base32 helpers to other folks | 15:45:56 |
| raitobezarius | I think I will check your chain and look if I can easily add the symlink feature | 15:46:06 |
| raitobezarius | (and also test it on Linux) | 15:46:15 |
| raitobezarius | (I will fix the compile errors locally) | 15:46:55 |
| emily | I have been testing Linux too, fwiw | 15:50:08 |
| emily | though not as extensively. there are test failures there I need to deal with | 15:50:17 |
| emily | I am increasingly of the opinion that it is unwarranted complexity. we already print out a note naming the build directory on failure | 15:50:36 |
| emily | it is just easy to miss and doesn't identify the failing derivation | 15:50:45 |
| emily | IMO making it attached to the error of the derivation failing (and therefore more prominent) would solve the UX case with fewer moving parts | 15:51:09 |
| emily | right now you get e.g.
note: keeping build directory '/nix/var/nix/b/3qxda0xkh4wllkxh182kvrza5n'
error: builder for '/nix/store/r0729886rq9nvzsngj2dnwwdl5q5dcq0-test.drv' failed with exit code 1;
last 1 log lines:
> test
For full logs, run 'nix log /nix/store/r0729886rq9nvzsngj2dnwwdl5q5dcq0-test.drv'.
| 15:51:50 |
| emily | we should simply list the retained build directory in the "For full logs" portion of the message | 15:52:00 |
| raitobezarius | what happened to my nix | 15:52:02 |
| raitobezarius | In reply to @emilazy:matrix.org
we should simply list the retained build directory in the "For full logs" portion of the message yeah you're right | 15:52:14 |
