| 7 Nov 2025 |
 maralorn | me | 16:20:45 |
| maralorn | I discussed this with ben. The problem is, that it increases binary size significantly. So we probably can’t enable it by default. And its also not possible to strip it in binaries which don’t want it. | 16:21:59 |
| maralorn | So the best available option would be to replace our profiling builds with an info-table-map build. | 16:22:30 |
| maralorn | * So the best available option would be to replace our profiling builds with an info-table-map build by default. | 16:22:43 |
| maralorn | But I actually really wanna do it. | 16:27:15 |
| maralorn | So maybe we can try enabling it and see if anyone complains. 😄 | 16:27:32 |
 teo (they/he) | Also infotable map completely disabled dead code elimination since the table keeps every section alive | 17:34:54 |
| teo (they/he) | * | 17:35:00 |
| teo (they/he) | What would be great is being able to rebuild base with ipe info easily without rebuilding ghc | 17:42:16 |
 sterni | then that's a hard no… | 19:20:35 |
 chreekat | Didn't I hear from mpickering in some talk recently that profiling mode is almost never what you want nowadays? You already get a bunch of great stuff "for free", and then when you want more you still don't usually want -fprof. I could be remembering this all wrong though | 19:21:25 |
| maralorn | No, that is completely correct. | 19:24:09 |
| maralorn | That’s why we should probably do something about it in nixpkgs. | 19:24:52 |
| maralorn | The problem is, that profiling builds had another ending. I don’t think simply building libraries twice with different flags would just work. | 19:25:37 |
| chreekat | Ahh that's a shame | 19:31:37 |
| maralorn | But I haven’t looked into it much. It might be quite feasible. | 19:39:42 |
| maralorn | But it would also be kinda a breaking change. Who knows whe relies on profiling builds. | 19:40:07 |
| maralorn | * But it would also be kinda a breaking change. Who knows who relies on profiling builds. | 19:40:11 |
 John Ericson | srk ⚡️: I think the PR is finally working! | 19:43:35 |
| John Ericson | (the hnix-store one) | 19:44:10 |
 alexfmpe | In reply to @b:chreekat.net
Didn't I hear from mpickering in some talk recently that profiling mode is almost never what you want nowadays? You already get a bunch of great stuff "for free", and then when you want more you still don't usually want -fprof. I could be remembering this all wrong though I'm told that nowadays you can also get flamegraphs without profiling builds | 20:02:09 |
| John Ericson | alexfmpe: you don't have perms on that library do you? :) | 20:04:21 |
| alexfmpe | Huh? Which lib | 20:08:28 |
| maralorn | Yes, but they are often relatively useless when not all your dependencies are compiled with -finfo-table-map. | 20:09:20 |
| John Ericson | hnix-store | 20:45:10 |
|  @emma:rory.gay left the room. | 22:39:32 |
| 9 Nov 2025 |
|  giladwo joined the room. | 11:41:19 |
 MangoIV | Yeah I know. :/
A third output maybe? Build multiple ones? Compression also improves it. Did you try it out with compression?
| 11:55:29 |
| maralorn | I didn’t. | 12:20:04 |
| maralorn | I think the way to go would be to first figure out a way to provide a third output and then at some point default the profiling output to off and the info table output to on. | 12:21:05 |
