| 11 Jul 2025 |
 raitobezarius | New rule | 18:40:24 |
| raitobezarius | Let's hash the drv name | 18:40:34 |
| raitobezarius | And replace the whole thing with 32 random chars | 18:40:41 |
| raitobezarius | If you do keep failed, you get to rename it to the original supposed name | 18:40:54 |
| raitobezarius | (idea brought to me by horrors' despair) | 18:41:20 |
 puck | In reply to @raitobezarius:matrix.org
Let's hash the drv name use the modulo drv hash tbh | 18:41:53 |
 jade_ | im just thinking ship it | 19:05:56 |
 emily | In reply to @raitobezarius:matrix.org
I mean even if it was 0 byte added, any program can decide to add itself and just go wrong right? but like, that was true before | 19:08:26 |
| emily | it's just that now much more of the limit is being eaten | 19:08:40 |
| emily | In reply to @raitobezarius:matrix.org
If you do keep failed, you get to rename it to the original supposed name I don't like this because it could impede reproducing a failure | 19:09:19 |
| emily | In reply to @emilazy:matrix.org
it's just that now much more of the limit is being eaten like empirically it's the difference between having to patch a bunch of stuff and not | 19:10:03 |
| jade_ | another possible and silly answer is to dump symlinks in /tmp for builds. then you can screw with the builds but that's not "our" problem | 19:14:05 |
| emily | wait how would that fix things using relative build dir directories for sockets to open? | 19:14:55 |
| jade_ | oh, they're doing that? that's unfortunate | 19:15:10 |
| emily | it doesn't seem unfortunate to me. it'll be inevitable once /tmp is closed off | 19:15:33 |
| jade_ | at work we made a symlink at /tmp/something-pc-socket to //.pc/something-socket for our process-compose sockets | 19:15:41 |
| jade_ | because of that problem | 19:15:53 |
| jade_ | * at work we made a symlink at /tmp/something-pc-socket to $sourcedir/.pc/something-socket for our process-compose sockets | 19:16:10 |
| emily | like ultimately that's the scratch space you get to run tests etc. | 19:16:13 |
| jade_ | /nix/var/nix/builds is only 19 chars | 19:18:28 |
| emily | /tmp was only 4 | 19:21:47 |
| emily | I wouldn't be raising it if three separate people didn't notice builds starting to fail because of it | 19:22:41 |
| raitobezarius | Can we summarize this topic in an issue so we can process it in async? I am not against empirical workarounds for this case | 19:22:41 |
| emily | I can probably in ~an hour | 19:23:58 |
| raitobezarius | I just would like to obtain information on the scale of this problem and what does changing this buy us wrt to the ocean of (weird?) software | 19:24:42 |
| raitobezarius | In reply to @emilazy:matrix.org
I wouldn't be raising it if three separate people didn't notice builds starting to fail because of it Don't get us wrong, we consider it to be a serious problem but I just want to ensure we don't pile up workarounds where we could look for a definitive solution | 19:25:33 |
| emily | Randy who regularly rebuilds the world reported failures, another contributor who doesn't rebuild the world but reviews lead package stuff reported failures, and Jujutsu CI broke through several separate codepaths | 19:25:54 |
| emily | last one definitely doesn't use Lix (picked up the issue on the backport), first one definitely does, no idea about the middle. but clearly it's nontrivially breaking | 19:26:32 |
| emily | *leaf package | 19:26:43 |
| emily | In reply to @raitobezarius:matrix.org
Don't get us wrong, we consider it to be a serious problem but I just want to ensure we don't pile up workarounds where we could look for a definitive solution I think there is no definitive solution to a path length limit being low. ultimately some stuff will be tested with /tmp so if we close off /tmp at some point then the best we can do is /nix/b/0 or something which still might break things | 19:28:31 |
