 | Nixpkgs Stdenv | 255 Members |
| 79 Servers |
| Nixpkgs Stdenv | 255 Members |
| 79 Servers |
| Sender | Message | Time |
|---|---|---|
| 10 Aug 2023 | ||
 p14 | dropping into a develop shell makes the problem go away | 13:46:40 |
 trofi | Is it an exact command? THe command coplains to me as error: flake 'flake:nixpkgs/73f18ca8b33ad1281958193e7e35f63b6ef47e52' does not provide attribute | 13:46:45 |
| p14 | Uh, should be, is it possible I have something in my flake registry messing things up? | 13:47:06 |
| p14 | How about ix build --rebuild github:nixos/nixpkgs/73f18ca8b33ad1281958193e7e35f63b6ef47e52#pkgsLLVM.llvmPackages_15.compiler-rt-no-libc.stdenv.cc.bintools -L; stat -c%s result/nix-support/setup-hook ? | 13:47:25 |
| p14 | * How about nix build --rebuild github:nixos/nixpkgs/73f18ca8b33ad1281958193e7e35f63b6ef47e52#pkgsLLVM.llvmPackages_15.compiler-rt-no-libc.stdenv.cc.bintools -L; stat -c%s result/nix-support/setup-hook ? | 13:47:31 |
 Artturin | nix build --rebuild nixpkgs/73f18ca8b33ad1281958193e7e35f63b6ef47e52#pkgsLLVM.llvmPackages_15.compiler-rt-no-libc.stdenv.cc.bintools works fine here | 13:48:03 |
| Artturin | fine as in no error: flake... | 13:49:22 |
| trofi | getting 4781 here. is it a nixos? | 13:49:25 |
| p14 | Yes, it's a nixos. | 13:49:33 |
| p14 | And I get 4781 on my good machine as well | 13:49:42 |
| p14 | But 0 on the bad (obv). | 13:49:52 |
| trofi | And nothing different in build log either compared to the "good" one? Let's see what populates that setup-hook. | 13:50:42 |
| p14 | Build logs appear the same. | 13:51:18 |
| p14 | I'm wondering how I can get my hands on more detailed diagnostics without perturbing the problem as NIX_DEBUG and set -x seem to. One idea is to diff the build closures of the good machine and bad machine somehow | 13:51:54 |
| p14 | Though I kinda suspect they'll be the same | 13:52:02 |
| p14 | Another idea is to use 'perf trace record' or similar and try and get a trace of what the builder is doing. | 13:52:29 |
| p14 | But I'm not sure I'll be able to dig through it, I'm not really sure what I'm looking for. I suppose I could start by trying to identify the writes of the setup hook. | 13:52:54 |
| p14 | Is there an easy way I can attach strace to the builder 🤔 | 13:53:17 |
| trofi | I usually strace -f nix-daemon itself and see what it forks out :) | 13:54:41 |
| p14 | So to make nix use that I'd start by killing the existing daemon and then running that as root, I guess. | 13:55:08 |
| trofi | AFAIU setup-hook is the file that gets passed to the derivation as an input as is: Then pkgs/stdenv/generic/setup.sh merges them into a single file in | 13:56:34 |
| p14 | Download setup-hook-trace.log | 14:01:08 |
| p14 | A fragment of the strace where setup-hook.sh is opened for reading and /nix-support/setup-hook is opened for writing. There is a load of file descriptor dup'ing going on which complicates understanding what is happening | 14:02:02 |
| p14 | I'm struggling a bit to make sense of it. The writes to fd 1 (beginning '# Binutils Wrapper hygiene') are the content which is missing, I think | 14:02:33 |
| trofi | You can also try to add ... .overrideDerivation(oa: { preFixup = "set -x"; }) into your expression. Will probably require slightly changing your build command to nix build --expr with import builtinsFetch.git { ... }; <your thing>`. | 14:02:54 |
| trofi | * You can also try to add ... .overrideDerivation(oa: { preFixup = "set -x"; }) into your expression. Will probably require slightly changing your build command to nix build --expr 'with import builtinsFetch.git { ... }; \<your thing>\. | 14:03:08 |
| p14 | When I do set -x like that, the problem goes away | 14:03:11 |
| p14 | But with the given strace, the problem is present (assuming --rebuild doesn't do something funny whereby the build functions correctly but I'd get a bad file under result/nix-support/setup-hook) | 14:03:57 |
| p14 | In the trace, /nix/store/9kaazhysw3pmzlrslpb1nsgy97hq8hlm-x86_64-unknown-linux-gnu-llvm-binutils-wrapper-15.0.7/nix-support/setup-hook in file descriptor 3 is the empty file, and the writes to file descriptor 1 are what it should contain (but it is empty). | 14:04:35 |
| trofi | you can also add -y to strace to get something more readable. | 14:04:38 |