The shell substitutions look like the value would come from buildPlatform, not hostPlatform (the default) like the other substitutions. Does it really exist? hello.drvPath doesn't have it. Otherwise I wouldn't be surprised if these packages were broken when cross-compiled.

The shell substitutions look like the value would come from buildPlatform, not hostPlatform (the default) like the other substitutions. Does it really exist? hello.drvPath doesn't have it. Otherwise I wouldn't be surprised if these packages were broken when cross-compiled.

Right you are. I did the following on master:

nix-build -A pkgsCross.aarch64-multiplatform-musl.deterministic-uname

Then looked at the shebang at the top of the result (which is where @shell@ expands):

/nix/store/a1s263pmsci9zykm5xcdf7x9rv26w6d5-bash-5.2p26/bin/bash

Then used file:

$ file /nix/store/a1s263pmsci9zykm5xcdf7x9rv26w6d5-bash-5.2p26/bin/bash
/nix/store/a1s263pmsci9zykm5xcdf7x9rv26w6d5-bash-5.2p26/bin/bash: ELF 64-bit LSB executable, x86-64, version 1 (SYSV), dynamically linked, interpreter /nix/store/ddwyrxif62r8n6xclvskjyy6szdhvj60-glibc-2.39-5/lib/ld-linux-x86-64.so.2, BuildID[sha1]=1098c4d83aff81fd6ddd5bc06fb9442388f4f51d, for GNU/Linux 3.10.0, not stripped

Definitely wrong.

It looks like the correct thing to do is use runtimeShell.

while substituteAll implementation may be terrible, the interface is much nicer than using substitute. couldn't substitueAll be re implemented so that the key / value pairs just passed to substitute using the --subst-var-by flag? like:

# some nix code 
substituteAllish {
 src = ./the/file;
 substitutions = {
   inherit coreutils;
   stuff = someComputation;
   etc.
 };
};

# substituteAllish mock implementation
{ lib, src, substitutions):
let
subs =  lib.flatten ( map (pair:  [ "--subst-var-by" pair.name pair.value ]) (lib.attrsets.attrsToList substitutions ));
in
substitute { inherit src; substitutions = subs; etc. }

?

I've been cooking exactly that patch @a-n-n-a-l-e-e. My hope is that I'll get through enough of these substituteAll refactorings to get a sense of how often there's a defect embedded in them to recommend a secondary approach. So far, there's basically a defect per patch.

Edit: made a PR with the cooked patch: #301350

Let's collaborate on what better would look like on #237216.