| 13 Sep 2025 |
 emily | this does mean that -linkmode=internal will produce broken binaries unless you also disable PIE I guess but it doesn't seem like people are really, doing that | 19:05:28 |
| emily | I'm also not entirely sure how you would explicitly out of PIE here either but not sure if that will be an issue | 19:06:09 |
| emily | we'll see. it's worth trying anyway | 19:06:17 |
 Lun | https://github.com/NixOS/nixpkgs/pull/442510/commits/3f163abc84ddffb2fea646f9fa973b503c13d592 ? | 19:06:55 |
 dish [Fox/It/She] | seems like theres some stuff not in the right namespace so it breaks | 19:07:53 |
| dish [Fox/It/She] | Download arrow-cpp.log | 19:07:54 |
| dish [Fox/It/She] | attached excerpt from the failed build log | 19:08:03 |
| dish [Fox/It/She] | ahh it truncated a bit hold on | 19:08:43 |
| dish [Fox/It/She] | there we go | 19:09:09 |
| dish [Fox/It/She] | Download arrow-cpp.log | 19:09:10 |
| emily | looks good to me. we should test a normal package on platforms, a package with CGO_ENABLED = 0, syncthing, something pkgsStatic, maybe some WASI thing although I have no idea what | 19:09:41 |
| emily | my #1 fear is basically just something doing -linkmode=internal explicitly and creating an FHS binary | 19:09:53 |
| emily | but it doesn't seem like much is doing that | 19:10:13 |
| Lun | I can kick off a bigger rebuild (later) if there's a nice way to select targets that are likely to break based on their attrs like there was for things that explicitly hardeningDisabled pie | 19:11:22 |
| emily | ooh | 19:11:35 |
| emily | Lun: hey sorry I was dumb actually | 19:11:48 |
| emily | check this out https://github.com/golang/go/blob/master/src/internal/platform/supported.go#L224-L253 | 19:11:55 |
| emily | actual patch is DefaultPIE on everything but ppc64 | 19:12:49 |
| emily | and we could even make InternalLinkPIESupported do return false | 19:13:03 |
| emily | to ensure that we never sielently produce busted PIEs | 19:13:13 |
| emily | (well, and I guess keep the Windows -race case) | 19:13:41 |
| dish [Fox/It/She] | good ol 'meant to be compatible but isn't in subtle ways', though disabling the support doesn't seem to impact anything(did pyarrow, utf8proc, and gdal builds, and they all work well, so I'm not anticipating any other issues from this change) | 19:21:52 |
| dish [Fox/It/She] | good ol 'meant to be compatible but isn't in subtle ways', though disabling the support doesn't seem to impact anything(did pyarrow, utf8proc, and gdal builds, and they all work well, so I'm not anticipating any issues from this change) | 19:22:37 |
| Lun | there being overlapping but not exactly the same checks here and in ld/config.go feels like it implies that that changing both might be needed / there's some subtlety.
I'm not at all familiar with this codebase, how sure are you that changing only platform/supported.go is right? | 19:25:52 |
| emily | low, but my understanding is that the thing I originally suggested was for platforms that do not support non-PIE | 19:29:01 |
| emily | and makes -buildmode=exe the same as -buildmode=pie | 19:29:06 |
| emily | whereas this determines -buildmode=default | 19:29:14 |
| emily | so this way we give an opt-out and things that do -buildmode=exe -linkmode=internal will work properly | 19:29:27 |
| emily | and then by adjusting InternalLinkPIESupported we can ensure we never produce busted FHS executables | 19:29:41 |
| emily | but I am far from confident | 19:29:53 |
