Re: 01/01: build-system/meson: Really skip the 'fix-runpath' phase on armhf.

[Marius Bakke <mbakke@fastmail.com>]

[-- Attachment #1: Type: text/plain, Size: 2240 bytes --]

ludo@gnu.org (Ludovic Courtès) writes:

> Hello,
>
> Mark H Weaver <mhw@netris.org> skribis:
>
>> Marius Bakke <mbakke@fastmail.com> writes:
>
> [...]
>
>>> I'm sorry, I forgot to address your actual concerns.  The (buggy)
>>> workaround was put in place and discussed in
>>> <https://bugs.gnu.org/30761>.  The meat of it can be found in (guix
>>> build-system meson):
>>>
>>>     ;; XXX PatchELF fails to build on armhf, so we skip
>>>     ;; the 'fix-runpath' phase there for now.  It is used
>>>     ;; to avoid superfluous entries in RUNPATH as described
>>>     ;; in <https://bugs.gnu.org/28444#46>, so armhf may now
>>>     ;; have different runtime dependencies from other arches.
>>
>> Thanks for this, but I'd still like to know the answer to my questions:
>> "What does the [fix-runpath] phase accomplish, and how will armhf users
>> be disadvantaged by the removal of that phase?"
>
> As discussed in <https://bugs.gnu.org/31970> and
> <https://bugs.gnu.org/31974>, Meson does not (or did not) adjust
> RUNPATHs upon installation (contrary to what Libtool does, for
> instance.)
>
> Consequently, the RUNPATH is left with /tmp/guix-build-… entries, which
> is not great but okay, but more importantly if usually lacks OUTPUT/lib
> as well.

I haven't seen /tmp in RUNPATH during my testing, which would be a
*huge* security problem.  The only consequence I've noticed from
dropping 'fix-runpath' is that it sometimes contain entries that are not
in NEEDED (but often required for a "neighbour" library, so no big deal).

> However, the commit Marius referred to¹ as well as what you reported for
> Epiphany in #31974 suggest that things are improving in Meson proper,
> and that we might be able to remove that ‘fix-runpath’ phase altogether
> soon.
>
> I think we should simply try building things without ‘fix-runpath’ and
> see if ‘validate-runpath’ reports anything.
>
> Thoughts?

I'm in favor of removing it on all architectures and see how it fares.
I suspect the main reason for adding it was that <out>/lib was often
lacking, which is mitigated by 09a45ffb146fda75b87f89c729c31d1da5bf93da.

I'll prepare patches for this for the next staging round.

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 487 bytes --]