[bug#48825] gnu: Simplify the use of --with-long-double-128 on powerpc64le.

Hi,

Currently in Guix, we explicitly specify --with-long-double-128 on
powerpc64le-linux (and potentially other powerpc64-* systems) in two
places.  It turns out that this is not necessary.  This patch simplifies
our code.

I originally authored this in order to try fixing bug 47698, but it
didn't solve that problem, and it turns out the --with-long-double-128
option is probably unrelated to that bug.  However, it's still worth
making this change to simplify our code.

If nobody has any issues with this, then in 2 weeks I will commit it.
One question is: where should I commit it?  I'd like to commit it to
master, but it causes many rebuilds on powerpc64le-linux (it does not
cause rebuilds on any other platforms), so per the guidelines ((guix)
Submitting Patches) I ought to commit it to core-updates.  However,
because the glibc upgrade from 2.31 to 2.32 on core-updates causes many
problems for powerpc64le-linux, core-updates has never once been a
viable branch for powerpc64le-linux in the time since support was first
added on master.  So if I commit this patch to core-updates, it won't
really do anyone any good right now.  With all this in mind, I think
master is the right place to commit this patch.

On Thu, Jun 03, 2021 at 10:08:00PM -0700, Chris Marusich wrote:
> Hi,
> 
> Currently in Guix, we explicitly specify --with-long-double-128 on
> powerpc64le-linux (and potentially other powerpc64-* systems) in two
> places.  It turns out that this is not necessary.  This patch simplifies
> our code.
> 
> I originally authored this in order to try fixing bug 47698, but it
> didn't solve that problem, and it turns out the --with-long-double-128
> option is probably unrelated to that bug.  However, it's still worth
> making this change to simplify our code.
> 
> If nobody has any issues with this, then in 2 weeks I will commit it.
> One question is: where should I commit it?  I'd like to commit it to
> master, but it causes many rebuilds on powerpc64le-linux (it does not
> cause rebuilds on any other platforms), so per the guidelines ((guix)
> Submitting Patches) I ought to commit it to core-updates.  However,
> because the glibc upgrade from 2.31 to 2.32 on core-updates causes many
> problems for powerpc64le-linux, core-updates has never once been a
> viable branch for powerpc64le-linux in the time since support was first
> added on master.  So if I commit this patch to core-updates, it won't
> really do anyone any good right now.  With all this in mind, I think
> master is the right place to commit this patch.
> 
> -- 
> Chris

couple of thoughts:
powerpc64le is in 'technology preview', so IMO it's fine to make big
changes to it as needed.

On master (and probably on core-updates too) we have a patch for glibc
to force ... something (that I don't remember) on powerpc architectures,
which has the side effect of needing '--with-long-double-128' on
powerpc-linux also in commencement.scm. If we could drop that patch then
I don't think we would need it anymore for powerpc.

> From ad89f9f59d22cc10fbf7dd6f738ce15a6e79b640 Mon Sep 17 00:00:00 2001
> From: Chris Marusich <cmmarusich@gmail.com>
> Date: Sat, 10 Apr 2021 18:16:17 -0700
> Subject: [PATCH] gnu: Simplify the use of --with-long-double-128 on
>  powerpc64le.
> 
> In short, this change adds the "--with-long-double-128" configure option in
> one place and removes it from two other (now-redundant) places.  It does not
> cause any rebuilds on systems other than powerpc64le-linux.
> 
> * gnu/packages/gcc.scm (gcc-configure-flags-for-triplet): Add a clause for
> targets starting with "powerpc64le-" which adds the "--with-long-double-128"
> option.  This causes any package using this procedure to be built using this
> new option on powerpc64le systems.  In particular, this affects the gcc
> package and the gcc-final package, in addition to all the other versions of
> GCC defined in (gnu packages gcc).
> * gnu/packages/commencement.scm (gcc-boot0)[#:configure-flags]: Remove the
> code that adds the "--with-long-double-128" configure option for powerpc64le,
> since it is now redundant. The gcc-boot0 package uses (and adds to) the gcc
> package's configure options. This means that the above change in gcc.scm is
> sufficient to ensure that the gcc-boot0 package's configure options will
> include "--with-long-double-128" on powerpc64le systems.
> * gnu/packages/cross-base.scm (cross-gcc-arguments)[#:configure-flags]: Remove
> the code that adds the "--with-long-double-128" configure option for
> powerpc64le, since it is now redundant. The cross-gcc-arguments procedure
> uses (and adds to) the configure options of its xgcc argument (a package).
> This means that regardless of which gcc from gcc.scm is used as the xgcc, the
> above change in gcc.scm is sufficient to ensure that the cross-gcc-arguments
> procedure's configure options will include "--with-long-double-128" on
> powerpc64le systems.
> ---
>  gnu/packages/commencement.scm | 7 -------
>  gnu/packages/cross-base.scm   | 6 ------
>  gnu/packages/gcc.scm          | 3 +++
>  3 files changed, 3 insertions(+), 13 deletions(-)
> 
> diff --git a/gnu/packages/commencement.scm b/gnu/packages/commencement.scm
> index d4511ed914..db564db9c4 100644
> --- a/gnu/packages/commencement.scm
> +++ b/gnu/packages/commencement.scm
> @@ -2819,13 +2819,6 @@ exec " gcc "/bin/" program
>                             "--disable-shared"
>                             "--enable-languages=c,c++"
>  

I've adjusted this on core-updates to also take effect on powerpc-linux.

> -                           ;; boot-triplet inserts "guix" in the triplet.
> -                           ,@(if (equal? "powerpc64le-guix-linux-gnu" (boot-triplet))
> -                                 ;; On POWER9 (little endian) glibc needs the
> -                                 ;; 128-bit long double type.
> -                                 '("--with-long-double-128")
> -                                 '())
> -
>                             ;; libstdc++ cannot be built at this stage
>                             ;; ("Link tests are not allowed after
>                             ;; GCC_NO_EXECUTABLES.").
> diff --git a/gnu/packages/cross-base.scm b/gnu/packages/cross-base.scm
> index 180594509b..c1e5f2eb79 100644
> --- a/gnu/packages/cross-base.scm
> +++ b/gnu/packages/cross-base.scm
> @@ -153,12 +153,6 @@ base compiler and using LIBC (which may be either a libc package or #f.)"
>                                 "--disable-decimal-float" ;would need libc
>                                 "--disable-libcilkrts"
>  
> -                              ,@(if (string-prefix? "powerpc64le-" target)
> -                                   ;; On POWER9 (little endian) glibc needs
> -                                   ;; the 128-bit long double type.
> -                                   '("--with-long-double-128")
> -                                   '())
> -
>                                 ;; When target is any OS other than 'none' these
>                                 ;; libraries will fail if there is no libc
>                                 ;; present. See
> diff --git a/gnu/packages/gcc.scm b/gnu/packages/gcc.scm
> index a412c93c29..22a0f35422 100644
> --- a/gnu/packages/gcc.scm
> +++ b/gnu/packages/gcc.scm
> @@ -79,6 +79,9 @@ where the OS part is overloaded to denote a specific ABI---into GCC
>           ;; Cilk has been removed from GCC 8 anyway.
>           '("--disable-libcilkrts"))
>  

This can be just 'powerpc64le'

> +        ((string-prefix? "powerpc64le-" target)
> +         '("--with-long-double-128"))
> +
>          (else
>           ;; TODO: Add `arm.*-gnueabi', etc.
>           '())))
> -- 
> 2.30.2
>