CVS commit: pkgsrc/lang/rust

[Taylor R Campbell <riastradh%NetBSD.org@localhost>]

> Module Name:    pkgsrc
> Committed By:   he
> Date:           Sun Feb 23 08:53:55 UTC 2025
> 
> Log Message:
> lang/rust: update to version 1.83.0 from pkgsrc-wip.
> [...]
>  * On NetBSD, use patchelf in the install phase to fix up RPATHs.
> [...]
> +.if ${OPSYS} == "NetBSD"
> +       for bin in ${PATCH_BINARIES}; do                                \
> +               f=${DESTDIR}${PREFIX}/$${bin};                          \
> +               if [ -f $$f ]; then                                     \
> +cmd="${PREFIX}/bin/patchelf ${FORCE_RPATH} --set-rpath ${PREFIX}/lib $$f"; \
> +                       echo $$cmd;                                     \
> +                       $$cmd;                                          \
> +               fi;                                                     \
> +       done
> +.endif

I believe this change is wrong and should be reverted.

It is not necessary: rust builds fine without it.

Having unnecessary elf hacks is harmful: it raises the maintenance
burden and difficulty of understanding the Rust build, which is
already extraordinarily high.

The commit message from pkgsrc-wip adding the patchelf hacks said
this:

> commit 643b92b0270478ff7e932715e0beb245235a1ca7
> Author: Havard Eidnes <he%NetBSD.org@localhost>
> Date:   Sun Jan 26 12:19:46 2025 +0100
> 
>     rust183: add use of patchelf in the install step.
>     
>     This so that the new rpath checks in pkgsrc pass, and the result
>     works as instended.  I suspect this is because we suppress $ORIGIN
>     rpaths in pkgsrc, as has been discussed on tech-pkg@.

I believe the discussion was
https://mail-index.netbsd.org/tech-pkg/2025/01/01/msg030311.html
(should always put a link to the discussion in the commit message if
you want to cite it so nobody has to wonder!).

However, at the same time this commit went into pkgsrc-wip, the
critical patch replacing $ORIGIN/.. by @PREFIX@ was missing -- and I
suspect that omission may have been the whole motivation for the rpath
discussion in the first place.

You can see this by comparing the patches as they existed at the time
of that commit in pkgsrc-wip between rust182 and rust183:

git show 643b92b0270478ff7e932715e0beb245235a1ca7:rust182/patches/patch-src_bootstrap_src_core_builder.rs
git show 643b92b0270478ff7e932715e0beb245235a1ca7:rust183/patches/patch-src_bootstrap_src_core_builder.rs

This is the relevant change:

-                Some(format!("-Wl,-rpath,$ORIGIN/../{libdir}"))
+                Some(format!("-Wl,-rpath,@PREFIX@/{libdir}"))

It appears in the rust182 patch, but it was lost in the rust183 patch.

A week later, on Feb 4, jperkin restored the patch in rust183, thus
rendering the patchelf unnecessary again:

> commit 575611bd8b85be8fdf4d8edc61d507f6f0507601
> Author: Jonathan Perkin <jonathan%perkin.org.uk@localhost>
> Date:   Tue Feb 4 15:36:35 2025 +0000
> 
>     rust183: Restore ORIGIN rpath fix.

While working on cross-building rusty packages, I had locally restored
the patch, and I have also removed the patchelf business, and it has
worked fine without the patchelf business.

So I think we should remove the patchelf parts which were never
necessary in the first place except because we had lost a patch in the
rust182 -> rust183 update.