public inbox for binutils@sourceware.org
 help / color / mirror / Atom feed
From: Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
To: John Baldwin <jhb@FreeBSD.org>
Cc: Jan Beulich <jbeulich@suse.com>,  binutils@sourceware.org
Subject: Re: [PATCH] ld: Add lib32 directories for 32-bit emulation on FreeBSD/amd64
Date: Thu, 21 Dec 2023 13:03:29 +0100	[thread overview]
Message-ID: <yddbkajdapa.fsf@CeBiTec.Uni-Bielefeld.DE> (raw)
In-Reply-To: <6499aee9-3935-41c1-8897-a177f6173e8a@FreeBSD.org> (John Baldwin's message of "Wed, 20 Dec 2023 09:47:22 -0800")

Hi John,

>>> ... it doesn't look implausible that things may have worked on earlier
>>> versions (or else perhaps someone would have noticed long ago), and that
>>> hence your change might break things there.
>> I'm certain they didn't: I originally developed this patch 4 years ago,
>> but either forgot to submit it back then or hoped an active member of
>> the FreeBSD community would.  This must have been in the FreeBSD 11 or
>> 12 timeframe, and obviously nothing has happened/been improved since.
>
> FreeBSD has always used /usr/lib and for the "native" ABI and /usr/lib32 for
> 32-bit ABIs on 64-bit platforms.  This includes both i386 on x86-64 as well
> as 32-bit powerpc on powerpc64 and more recently 32-bit arm (armv7) on
> aarch64.  Given that, I believe the patch to be correct (and it likely applies
> for powerpc64* using "powerpc" emulation and aarch64* using "armv7" emulation).

good to see one platform that is consitent across architectures here ;-)
Given that I don't have access to any of the other FreeBSD platforms,
I'll leave such a patch to those that do.

>> My recent forays into FreeBSD have been less than pleasant,
>> unfortunately: see GCC PR target/112959 (install.tex needs updates on
>> FreeBSD) for an overview of the issues on the GCC side.  It seems the
>> FreeBSD community either cares little about GCC and binutils these days
>> (having moved to lld as /usr/bin/ld and clang/clang++), or doesn't
>> believe in upstream bug reports, let alone submitting patches ;-(  This
>> is not just about GCC/binutils; the same seems to happen on the LLVM
>> side where they completely ripped out the cmake-based build system and
>> replaced it with one of their own (based on BSD make).  Trying to build
>> upstream LLVM on FreeBSD is just as unpleasant as GCC.  E.g. GCC won't
>> work with lld (cf. GCC PR target/112745), so you need GNU ld here...
>
> While LLVM is the primary toolchain for FreeBSD, GCC is not completely
> ignored.  I maintain a set of ports for GCC packages customized to build
> FreeBSD's base system.  I recently added a new port for GCC 13 for this
> purpose and can currently build FreeBSD's development trunk (userland +
> kernel) on x86-64 with both GCC 12 and GCC 13.  (Right now GCC 12 is
> used in a CI job in FreeBSD's CI infrastructure, and the GCC 13 job is
> being added this week or next.)

There are regular posts of FreeBSD testresults to the gcc-testresults
list, so you are not alone here.  This matches what the Oracle Solaris
guys do for both their solaris-userland repo and use of GCC as a shadow
compiler building OS/Net (the core OS including kernel and Solaris
userland).

> I do have a couple of patches to GCC I should post upstream (switching
> i386 to default to -march=i686 in newer versions, and adding a
> __freebsd_kprintf__ format for the in-kernel printf() function), just
> haven't done paperwork yet on the GCC side.

Good: maybe the other issues I've found building GCC trunk on FreeBSD
can also be resolved and the experience for non-FreeBSD developers
become a bit smoother in the future.  I believe both those GCC
developers (testing their patches on more platforms) and FreeBSD (having
less patches that break their builds) would benefit from that.

> In regards to the build system, FreeBSD's entire base system builds with
> a single build infrastructure using make (and always has).  For any
> external tool used in the base, build glue written in make is used.  This
> was true of GCC and binutils when they were part of base as well.

I won't argue with that.  What becomes problematic IMO is when you
cannot build upstream GCC or LLVM without additional patches: having to
investigate build problems before you even get to testing your own
patches takes a patience that few developers will have.

Thanks for your explanations.

	Rainer

-- 
-----------------------------------------------------------------------------
Rainer Orth, Center for Biotechnology, Bielefeld University

      parent reply	other threads:[~2023-12-21 12:03 UTC|newest]

Thread overview: 7+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2023-12-13  9:33 Rainer Orth
2023-12-19 22:23 ` Rainer Orth
2023-12-20  7:35   ` Jan Beulich
2023-12-20  9:12     ` Rainer Orth
2023-12-20 17:47       ` John Baldwin
2023-12-21  7:25         ` Jan Beulich
2023-12-21 12:03         ` Rainer Orth [this message]

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=yddbkajdapa.fsf@CeBiTec.Uni-Bielefeld.DE \
    --to=ro@cebitec.uni-bielefeld.de \
    --cc=binutils@sourceware.org \
    --cc=jbeulich@suse.com \
    --cc=jhb@FreeBSD.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).