Update gcc/gfortran to 12.x + Apple Silicon support, update openblas to 0.3.21, gsl to 2.7.1

[mkoeppe]

We would use 12.2.0 plus the feature patch for Apple Silicon support that the homebrew folks use:
https://github.com/Homebrew/homebrew-core/blob/HEAD/Formula/gcc.rb#L12

Depends on #34266
Depends on #34228

CC: @culler @jhpalmieri @kliem

Component: packages: standard

Author: Matthias Koeppe

Branch/Commit: e89ba55

Reviewer: John Palmieri

Issue created by migration from https://trac.sagemath.org/ticket/33816

[mkoeppe]

comment:1

Looks like homebrew is able to use upstream gcc 12 on M1 - https://github.com/Homebrew/homebrew-core/blob/HEAD/Formula/gcc@12.rb

[mkoeppe]

comment:2

... with one big patch: https://github.com/Homebrew/homebrew-core/blob/HEAD/Formula/gcc@12.rb#L43

[mkoeppe]

Description changed:

--- 
+++ 
@@ -1 +1,3 @@
-just released
+We would use 12.2.0 plus the feature patch for Apple Silicon support that the homebrew folks use:
+https://github.com/Homebrew/homebrew-core/blob/HEAD/Formula/gcc.rb#L12
+

[mkoeppe]

Branch: u/mkoeppe/update_gcc_gfortran_to_12_x___apple_silicon_support

[mkoeppe]

Commit: 9ae559b

[mkoeppe]

New commits:

1dfeb14	build/pkgs/gcc: Update to 12.2.0
9ae559b	build/pkgs/gcc/patches/gcc-12.2.0-arm.patch: Import from https://raw.githubusercontent.com/Homebrew/formula-patches/1d184289/gcc/gcc-12.2.0-arm.diff

[mkoeppe]

Author: Matthias Koeppe

[jhpalmieri]

comment:7

I don't know if this is ready yet but I tried it anyway on my new M2 laptop. It failed; the end of the log file reads

0  0x1010cc1a0  __assert_rtn + 140
1  0x100f53a8c  mach_o::relocatable::Parser<arm64>::parse(mach_o::relocatable::ParserOptions const&) + 4536
2  0x100f25d38  mach_o::relocatable::Parser<arm64>::parse(unsigned char const*, unsigned long long, char const*, long, ld::File::Ordinal, mach_o::relocatable::ParserOptions const&) + 148
3  0x100f8e4ac  ld::tool::InputFiles::makeFile(Options::FileInfo const&, bool) + 1468
4  0x100f91360  ___ZN2ld4tool10InputFilesC2ER7Options_block_invoke + 56
5  0x188a041f4  _dispatch_client_callout2 + 20
6  0x188a17954  _dispatch_apply_invoke + 224
7  0x188a041b4  _dispatch_client_callout + 20
8  0x188a15a04  _dispatch_root_queue_drain + 680
9  0x188a16104  _dispatch_worker_thread2 + 164
10  0x188bc4324  _pthread_wqthread + 228
A linker snapshot was created at:
	/tmp/libstdc++.6.dylib-2022-09-20-133305.ld-snapshot
ld: Assertion failed: (_file->_atomsArrayCount == computedAtomCount && "more atoms allocated than expected"), function parse, file macho_relocatable_file.cpp, line 2061.

I can post the full log file if you want.

[mkoeppe]

comment:8

Thanks for testing. Also on Intel I ran into a build error. This will need more work.

[jhpalmieri]

comment:9

I'm happy to test. It's just a MacBook Air, but it builds Sage faster than my iMac Pro, and I like giving it tasks.

[sagetrac-git]

Branch pushed to git repo; I updated commit sha1. New commits:

49c407b

build/pkgs/gcc/spkg-build.in: Remove workarounds for ancient OS X

[sagetrac-git]

Changed commit from 9ae559b to 49c407b

[sagetrac-git]

Branch pushed to git repo; I updated commit sha1. New commits:

ba6de83

build/pkgs/gcc/build-gcc: Unset AS, LD

[sagetrac-git]

Changed commit from 49c407b to ba6de83

[mkoeppe]

comment:12

On Intel, this still fails tox -e local-macos-nohomebrew -- gfortran, but it passes with the full bootstrapped build tox -e local-macos-nohomebrew -- gcc.

The failure when trying to build gfortran using Apple's gcc (clang) is introduced by the patch - passing the argument -nodefaultrpaths to the compiler.