| RAOF | 🎶 My flaky hardware brings all the race-condition bugs to the yard 🎶 | 05:38 |
|---|---|---|
| * Saviq sent a diff code block: https://libera.ems.host/_matrix/media/r0/download/libera.chat/f28b3d032efb158386997cefd826d1efb8780d6a | 08:06 | |
| Saviq | What. | 08:06 |
| Saviq | If I'm reading this correctly, it's not liking the ShmFile symbols there, but also it wants them back? | 08:09 |
| Saviq | Or is it that #MISSING says the symbol is not there, when expected? | 08:10 |
| Saviq | So it's likely optimized out in LP? | 08:10 |
| alan_g[m] | Yeah, like I said I'm on it. AFAICS there's no reason it should be there, but maybe with some build options it exists as a weak symbol (and that happens in spread) | 08:13 |
| alan_g[m] | I know how to "brute force" a solution. But still trying to root cause | 08:14 |
| Saviq | we have `-lto` in spread/sbuild, but not in LP | 08:15 |
| Saviq | COMMON_CONFIGURE_OPTIONS: -DCMAKE_INSTALL_LIBEXECDIR="lib/x86_64-linux-gnu/mir" -DMIR_BUILD_INTERPROCESS_TESTS=OFF -DMIR_LINK_TIME_OPTIMIZATION=ON | 08:15 |
| Saviq | DEB_BUILD_MAINT_OPTIONS: | 08:15 |
| Saviq | vs. | 08:15 |
| Saviq | COMMON_CONFIGURE_OPTIONS: -DCMAKE_INSTALL_LIBEXECDIR="lib/x86_64-linux-gnu/mir" -DMIR_BUILD_INTERPROCESS_TESTS=OFF -DMIR_LINK_TIME_OPTIMIZATION=ON -DMIR_LINK_TIME_OPTIMIZATION=OFF | 08:15 |
| Saviq | DEB_BUILD_MAINT_OPTIONS: optimize=-lto optimize=-lto | 08:15 |
| alan_g[m] | Huh?! Why would LTO add a symbol? | 08:17 |
| Saviq | Well, _actually_ we have `noopt` in spread/sbuild, but not in LP | 08:17 |
| Saviq | https://github.com/MirServer/mir/blob/0f15cbbffcbcc8e89400c17c137f29453d69905b/spread/build/sbuild/task.yaml#L2 | 08:18 |
| alan_g[m] | Ah, that would make sense | 08:18 |
| -GitHub[m]:#mir-server- **[MirServer/mir]** Saviq opened [pull request #2530](https://github.com/MirServer/mir/pull/2530): spread: optimize in sbuild, too | 08:20 | |
| -GitHub[m]:#mir-server- | 08:20 | |
| -GitHub[m]:#mir-server- > Otherwise symbols may not be as we expect them. | 08:20 | |
| alan_g[m] | My "brute force" would be to ensure the symbols exist regardless | 08:24 |
| Saviq | Feel free to reject my PR, but IMO we should keep sbuild as close to LP as possible | 08:25 |
| alan_g[m] | We should indeed. I first want to see your PR fail in spread without reverting 555c3a06eaf613e4d18e81c0b23c6854509c9975 as proof | 08:27 |
| Saviq | +1 | 08:28 |
| alan_g[m] | But ... it fails for another reason before it gets there :( | 09:02 |
| alan_g[m] | 2022-08-04T08:38:47.5208900Z /usr/bin/ld: final link failed: No space left on device | 09:02 |
| Saviq | Gah. | 09:06 |
| Saviq | I'll find out if we can ask for moar space on there | 09:08 |
| Saviq | But this did fail as expected: | 09:11 |
| Saviq | https://github.com/MirServer/mir/runs/7655531762?check_suite_focus=true#step:5:13523 | 09:11 |
| Saviq | Wait. I'm looking at the wrong thing. | 09:12 |
| Saviq | A couple sbuild runs went fine (jammy and devel), which I suppose was unexpected (but could be due to tooling / toolchain) | 09:15 |
| Saviq | I've pushed a revert of 555c3a06 now to compare | 09:16 |
| -GitHub[m]:#mir-server- **[MirServer/mir]** AlanGriffiths opened [pull request #2531](https://github.com/MirServer/mir/pull/2531): Clean out symbols that don't (or may not) exist from mircore symbols.map | 09:32 | |
| -GitHub[m]:#mir-server- | 09:32 | |
| -GitHub[m]:#mir-server- > Some of these went with recent ABI changes, some never existed, some only exist without optimization | 09:32 | |
| alan_g[m] | Saviq we probably want to combine approaches once proven; but, I've proposed an alternative that should also work | 09:33 |
| -GitHub[m]:#mir-server- **[MirServer/mir]** AlanGriffiths requested a review from Saviq for [pull request #2531](https://github.com/MirServer/mir/pull/2531): Clean out symbols that don't (or may not) exist from mircore symbols.map | 09:50 | |
| alan_g[m] | Saviq can we land #2531? Once builder space is sorted #2530 can follow | 09:58 |
| Saviq | Ofc, l will look | 10:02 |
| -GitHub[m]:#mir-server- **[MirServer/mir]** bors[bot] merged [pull request #2531](https://github.com/MirServer/mir/pull/2531): Clean out symbols that don't (or may not) exist from mircore symbols.map | 10:56 | |
| alan_g[m] | Now wait for the PPA to give up another issue... | 10:57 |
| Saviq | :hamsterdance: | 11:00 |
| Saviq | So now we're left with https://github.com/MirServer/mir/issues/2527 | 12:59 |
| Saviq | I'll get an environment to reporduce | 12:59 |
| alan_g[m] | That looks like the compiler confusing itself | 13:03 |
| Saviq | That may well be, I'll try and bisect to see if it's anything we did (and can undo) or a toolchain upgrade | 13:14 |
| Saviq | Actually let me look at what it's installing between when it worked and stopped | 13:15 |
| Saviq | Right, that's a gcc 11 → 12 migration | 13:18 |
| alan_g[m] | Already reported upstream? Or time for a "minimal reproducer"? | 15:12 |
| Saviq | dunno yet | 15:17 |
| Saviq | Time o/ Good night all | 16:22 |
| alan_g[m] | Until tomorrow then. 😁 | 16:23 |
| alan_g[m] | And that's EOD for me too | 17:00 |
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!