Nix darwin hashes unreliable from Linux runner

[ndrwstn]

Summary

Since PR #9309 switched nix hash updates to a single Linux runner, Darwin hashes are unreliable. Large bun.lock changes (around v1.1.43+) appear to have broken the cross-computed hashes, and nix-darwin now fails with hash mismatches.

System

• macOS 26.2 (Build 25C56)
• Darwin Kernel 25.2.0, arm64 (Apple Silicon)

Error example

> nix run github:anomalyco/opencode/v1.1.47 --refresh
error: hash mismatch in fixed-output derivation '/nix/store/9a6m0j4bq1pambplyfn75ip291psq4v9-opencode-node_modules-1.1.47-6cd2a68.drv':
         specified: sha256-20RGG2GkUItCzD67gDdoSLfexttM8abS//FKO9bfjoM=
            got:    sha256-yvrNFmYeMYviyCVSahuLFXmyoIMT136WoJ0MEbk+AD8=
error: Cannot build '/nix/store/4idyk603r28knk02k1lbnzc3dlypahgf-opencode-1.1.47-6cd2a68.drv'.
       Reason: 1 dependency failed.
       Output paths:
         /nix/store/k0bazhcbmxbhg5zsbffypn5nm5crh7g6-opencode-1.1.47-6cd2a68

Request

Please restore macOS runners for darwin hash computation (x86_64-darwin, aarch64-darwin) or make the derivation host-independent.

Related to #9309.

[github-actions]

This issue might be a duplicate of existing issues. Please check:

• Bug: Nix hashes for nodes_modules incorrectly generated #8977: Bug: Nix hashes for nodes_modules incorrectly generated - discusses race conditions in CI hash computation
• Nix build fails on aarch64-linux and x86_64-darwin: missing nodeModules #8742: Nix build fails on aarch64-linux and x86_64-darwin: missing nodeModules - related to missing platform-specific hashes
• Nix package: hash mismatch in node_modules fixed-output derivation #7344: Nix package: hash mismatch in node_modules fixed-output derivation - similar hash mismatch errors

Feel free to ignore if none of these address your specific case.

[ndrwstn]

Wanted to tag @gigamonster256 since he was the person working on the nix actions.

[gigamonster256]

running jj bisect run --range "(v1.1.30..dev@upstream) & files(nix/hashes.json)" --command "nix build .#opencode.node_modules" on aarch64-darwin points to 8ce19f8 as the first bad CI update... trying to figure out why that is the case