| 20 Feb 2025 |
|  tobim joined the room. | 09:58:16 |
 p14 | I'm seeing something quite weird with gmpxx whereby it will succeed/fail at the configure stage checking build system compiler gcc... yes based on something bogus in the environment. I'm wondering whether its success/failure is actually conditioned on something arbitrary relating to what's in the environment somehow. I've made a build hook; with makeSetupHook which I'm putting in the stdenv's extraNativeBuildInputs. Loads of other packages build fine but not gmpxx. I can comment out preBuildHooks+=(something) or put it back, and this makes it pass/fail. The compiler detection fails unable to find libc. | 11:03:37 |
| p14 | Interestingly a develop shell reliably fails so I can't figure out why it's working in the working case. | 11:04:49 |
| p14 | I think I'm hitting a variant of https://github.com/NixOS/nixpkgs/issues/265121, that the buildPackages.stdenv.cc doesn't work as expected in a pkgsLLVM environment. | 11:05:35 |
| p14 | (because gmpxx uses this) | 11:05:45 |
| p14 | Right, figured it out. My develop environment was failing because of contamination from a direnv environment causing the wrong targetOffset, which meant I was trying to debug the wrong thing. And the preBuildHook was actually affecting the configure phase but later. | 11:55:47 |
| p14 | * Right, figured it out. My develop environment was failing because of contamination from a direnv environment causing the wrong targetOffset, which meant I was trying to debug the wrong thing. And the preBuildHook was actually affecting the configure phase but later; owing to how it affected the inputs to the derivation. | 11:56:01 |
| 21 Feb 2025 |
 Tristan Ross | I'm kinda wondering if it's possible to do "cross" things without activating the cross logic that causes the prefix/suffix to be added in. If it could be done with toolchain stuff and CPU model stuff, that'd be nice. For toolchains, we'll probably need multiple bootstrap tarballs. A base and then one for each toolchain. | 16:39:58 |
