| 11 Nov 2024 |
 emily | In reply to @reckenrode:matrix.org
Why would ld64 need patching? There’s nothing wrong with it. well, since it would mean fewer things changing depending on stdenv.targetPlatform. | 22:13:49 |
| emily | but it might not be the right place. probably the wrapper is a better idea. (but then adjusting the target platform would still result in rebuilds, right?) | 22:14:05 |
| emily | (really need targetPlatform = null…) | 22:14:31 |
 Randy Eckenrode | I have two main concerns:
- Darwin is not forced to be like other platforms for the sake of convenience. We’re finally undoing a lot of that after the refactor; and
- Darwin tooling is not hacked up to work around number 1.
| 22:15:53 |
| emily | I think it's more about a general principle that targetPlatform differences should not cause rebuilds of things that don't care about targetPlatform | 22:16:49 |
| emily | whereas currently targetPlatform being Darwin or not rebuilds the world | 22:16:59 |
| emily | in an ideal world the unwrapped LLVM derivation wouldn't even be able to look at targetPlatform, e.g. | 22:17:28 |
| emily | currently, rg 'targetPlatform\.is' pkgs/stdenv is two Darwins and two ghcjses | 22:18:36 |
| emily | (one of the Darwins is harmless, just making makeStaticBinaries throw) | 22:18:57 |
| emily | (and the ghcjs mentions also look harmless) | 22:19:12 |
| Randy Eckenrode | So if we can move this to the wrapper, that seems fine. | 22:21:14 |
| Randy Eckenrode | In reply to @emilazy:matrix.org
whereas currently targetPlatform being Darwin or not rebuilds the world That seems to be a Nix problem where any change in puts cause rebuilds even when they don’t matter. | 22:21:34 |
| Randy Eckenrode | In reply to @emilazy:matrix.org
whereas currently targetPlatform being Darwin or not rebuilds the world * | 22:22:05 |
| Randy Eckenrode | But anyway, to solve the problem at hand, if we can move it somewhere else to avoid the rebuilds, that’s fine. | 22:22:19 |
| emily | it seems like the wrapper would still cause rebuilds because ld-wrapper is used to build clang-unwrapped, etc., but maybe I don't fully understand the rebuild issue | 22:23:28 |
| emily | In reply to @reckenrode:matrix.org
That seems to be a Nix problem where any change in inputs cause rebuilds even when they don’t matter. a problem that the solution to breaks Darwin in other ways… | 22:23:49 |
| Randy Eckenrode | In reply to @emilazy:matrix.org
it seems like the wrapper would still cause rebuilds because ld-wrapper is used to build clang-unwrapped, etc., but maybe I don't fully understand the rebuild issue If you’re using a different linker, I’d expect clang-unwrapped to be rebuilt. There’s no avoiding that. | 22:24:29 |
| emily | well, the linker wouldn't matter, since it's only for the target platform | 22:25:52 |
| emily | but hm | 22:25:54 |
| emily | is ld-wrapper different for the target linker vs. the host one? | 22:26:12 |
| emily | I guess it must be? | 22:26:14 |
| emily | in which case yeah I guess we can move it in there. | 22:26:20 |
| Randy Eckenrode | Of course it is. It references the absolute path to the host linker. | 22:27:08 |
| Randy Eckenrode | * | 22:28:46 |
| emily | do you just reference stdenv.hostPlatform.isDarwin in the ld-wrapper derivation and it means stdenv.targetPlatform.isDarwin when you put it in pkgsBuildTarget? | 22:29:29 |
| Randy Eckenrode | Hmm. Would it be disgustingly hacky to set NIX_DONT_SET_RPATH_aarch64_apple_darwin=1 and NIX_DONT_SET_RPATH_x86_64_apple_darwin=1unconditionally?
| 23:08:53 |
| Randy Eckenrode | * Hmm. Would it be disgustingly hacky to set NIX_DONT_SET_RPATH_aarch64_apple_darwin=1 and NIX_DONT_SET_RPATH_x86_64_apple_darwin=1 unconditionally?
| 23:08:59 |
| Randy Eckenrode | (I’m guessing yes.)
| 23:09:14 |
| Randy Eckenrode | I guess just revert the commit that set NIX_DONT_SET_RPATH_FOR_TARGET for now.
| 23:11:10 |
| Randy Eckenrode | I wonder if NIX_DONT_SET_RPATH should be set in a hook that properly does role mangling.
| 23:11:35 |
