It seems that the host's wayland-scanner isn't leaking through PATH after all. Adding echo PATH=$PATH to preBuild:

PATH=/nix/store/ll2b6cslly4v30im7qk34f9y21ziiz1i-cmake-3.29.6/bin:/nix/store/qdb895193mlnczgcffcxwpac6pj252ai-aarch64-unknown-linux-gnu-pkg-config-wrapper-0.29.2/bin:/nix/store/a3n2vxsni6084nn5zcwy35lrvkwmz09y-wayland-1.22.0-bin/bin:/nix/store/yq6n8b0mnk0qxzbs3ajsjcp8ziwqylrl-patchelf-0.15.0/bin:/nix/store/gyfwnd1v9w911gw16ps5rx9ha2r9q2yw-aarch64-unknown-linux-gnu-gcc-wrapper-13.3.0/bin:/nix/store/4wx8k9iig0m3if1zcd2bh0q12hgvwnis-aarch64-unknown-linux-gnu-gcc-13.3.0/bin:/nix/store/n4nlm3h87gmp33nfpvzs1jnknj25c5v6-glibc-aarch64-unknown-linux-gnu-2.39-52-bin/bin:/nix/store/dhv5gh89him9a7ddr56cqg87zfkmjihp-coreutils-9.5/bin:/nix/store/0yicrw03llxadl8yhkya0gkhxb1cbwrw-aarch64-unknown-linux-gnu-binutils-wrapper-2.42/bin:/nix/store/k85314anqw1y1lc7rmxbv4miyfdlyzcj-aarch64-unknown-linux-gnu-binutils-2.42/bin:/nix/store/dhv5gh89him9a7ddr56cqg87zfkmjihp-coreutils-9.5/bin:/nix/store/j71whmq49hfcvhpn4wqzrqcnbq5mafic-findutils-4.10.0/bin:/nix/store/8ynk6xvczjnjv81bfghkjwrap0hn6qah-diffutils-3.10/bin:/nix/store/cdhx572ij0pvn9ayd33lk8118137j6m8-gnused-4.9/bin:/nix/store/mdiqq9b2rvv7fajlmg3f7d6r6g78l547-gnugrep-3.11/bin:/nix/store/a5rvjq2ir4d1wnxwdf4a9zf6hfc6ydsx-gawk-5.2.2/bin:/nix/store/0a0khkw34v25q8k6p44ma1rqa479r1za-gnutar-1.35/bin:/nix/store/aw1kk0gvw7j4vmjxhha2fj2jfsvgrd1p-gzip-1.13/bin:/nix/store/wr6x94hzs5ggmpql0qx5j8q2rpfrhm7q-bzip2-1.0.8-bin/bin:/nix/store/m9288777y854d5k0xkxgacc8rslrhb8r-gnumake-4.4.1/bin:/nix/store/m101dg80ngyjdb02g6jwy80sr7kzj26g-bash-5.2p26/bin:/nix/store/32yyy66vqi5ijh2nfas2nc83sw7qbgn1-patch-2.7.6/bin:/nix/store/0rlbqq6p9wwb7wbn5vnpck40c76czid3-xz-5.6.2-bin/bin:/nix/store/mhk0cm9nrjs855aq1vcivra9z9g9hn44-file-5.45/bin

This makes sense because wayland.dev in buildInputs has offsets (0, 1) (can be seen with env.NIX_DEBUG="6") which don't pass the guard at

Host's wayland.dev includes e.g. lib/pkgconfig/wayland-scanner.pc, but I don't think that broke the cross, because this wouldn't have got fixed by shrinking propagatedBuildOutputs.

I now see that host's wayland.bin also includes a lib/pkgconfig/wayland-scanner.pc (👀)...

We did some more digging, and while trying to build the gnome and plasma tests I uncovered a bunch of other binaries failing to build. I started changing some of these dependencies to wayland.bin, but it's getting a bit too much.

It's unclear why "build systems" (CMakes' find_program`) pick up the host executable instead of the build executable.

Also, the interaction between wayland-scanner, wayland.bin etc are still a bit unclear, we don't quite understand why we seem to be shipping our own custom wayland-scanner.pc file in waylands' bin, alias wayland-scanner to wayland.bin etc, and why we cannot just replace most of this with only wayland, without the additional nixpkgs attributes and custom .pc file. (cc @primeos, @K900 )

However, we don't want to keep staging broken while doing this research, so let's revert the wayland change for now.

Revert PR in #329523.

It's unclear why "build systems" (CMakes' find_program`) pick up the host executable instead of the build executable.

This seems related to #318226; cc @paparodeo?

It's unclear why "build systems" (CMakes' find_program`) pick up the host executable instead of the build executable.

This seems related to #318226; cc @paparodeo?

yeah, strict deps is broken for cmake because cmake will pull executable from buiildInputs which is exactly what you don't want. FWIW i stopped working on the PR given all my other stalled PRs i didn't want to spend more time on it just to have it go nowhere.

[edit] if i get a commitment from a committer to review and push it thru i'll resume working on it but am not planning on resuming otherwise.