| 19 May 2025 |
 K900 | But that's fine probably? | 13:39:47 |
 emily | isn't that just pkgsTargetTarget | 13:40:39 |
| emily | or pkgsTargetWhatever because the Whatever never matters in that case | 13:40:47 |
| emily | (if the target of your target-platform package matters you are living in sin) | 13:40:55 |
| K900 | It does if you're doing Canadian cross | 13:41:06 |
| emily | no | 13:41:17 |
| emily | at the time it matters it is the host-platform package's target | 13:41:25 |
| emily | the target of a target-platform package would be like… the GCC build wants to bundle a JIT library so that the built GCC can link it into target-platform executables, and that JIT library itself wants to target some other target | 13:42:09 |
| emily | that's when pkgsTargetHost vs. pkgsTargetTarget vs. whatever would matter I think. thankfully those don't exist | 13:42:32 |
| emily | but doesn't pkgsTargetTarget build on the build platform? | 13:42:40 |
| emily | maybe not. wow Nixpkgs sucks. wow splicing sucks | 13:42:48 |
| emily | if you can actually make having only pkgsOnBuild and pkgsOnHost and nothing else work correctly then I will be eternally grateful | 13:43:22 |
| emily | it'll also make GCC uglier and more annoying, which will be a great argument for switching to LLVM | 13:43:48 |
| K900 | No, pkgsTargetTarget is [host = target, target = target] | 13:46:26 |
| K900 | And in our current world host == target for packages that don't have target | 13:46:53 |
| emily | yes but we're talking about build = | 13:47:24 |
| emily | oh | 13:47:30 |
| emily | sorry I misread the package set you were talking about | 13:47:34 |
| emily | then I'm confused, isn't this just pkgsBuildTarget | 13:47:49 |
| emily | which we use a fair bit | 13:48:02 |
| K900 | Yes but it doesn't exist on Canadian cross | 13:48:26 |
| emily | I mean, it can, right? you just have to build another compiler | 13:49:48 |
| K900 | Well yes but that kinda defeats the point of Canadian cross | 13:51:02 |
| emily | is there a point of Canadian cross? | 13:52:30 |
| K900 | Yes, the point of Canadian cross is that you use your slow build machine to build a cross-compiler for a fast host machine | 13:53:19 |
| K900 | And then use that cross-compiler to build packages for target | 13:53:27 |
| K900 | But faster | 13:53:31 |
| emily | you could just build the compiler on the fast host machine. | 13:55:33 |
| emily | to me the sensible use-case for Canadian cross looks like "we can only get x86 CI, not AArch64, but we want to offer AArch64 binaries of our compiler for various targets" | 13:56:23 |
| emily | in which case building an extra compiler is basically fine. | 13:56:32 |
