arm64 build fixes

[dhomeier]

config.guess updates as required, some additional build patches; pushing now what I've come across so far to keep people from duplicating efforts.

[TheSin-]

You put WAY too many packages in one PR which is why they aren't getting accepted, it's too much to review.

both db53 and db60 are too complex, you only need to add one line to db60 and change one line on db53, that's all, they both need to be forced to use --with-mutex=Darwin/_spin_lock_try. This works for x86 and arm64 as it's what should be used for Darwin according to the m4 file.

[TheSin-]

I do approve the nettle7 fix though, just waiting on @nieder to approve it since it's his package.

[dhomeier]

both db53 and db60 are too complex, you only need to add one line to db60 and change one line on db53, that's all, they both need to be forced to use --with-mutex=Darwin/_spin_lock_try. This works for x86 and arm64 as it's what should be used for Darwin according to the m4 file.

Perhaps x86_64/gcc-assembly was used for performance reasons? But I don't have anything to benchmark that on, so the Darwin one appears good enough. Are you planning to submit your own PR?

[dmacks]

both db53 and db60 are too complex, you only need to add one line to db60 and change one line on db53, that's all, they both need to be forced to use --with-mutex=Darwin/_spin_lock_try. This works for x86 and arm64 as it's what should be used for Darwin according to the m4 file.

Perhaps x86_64/gcc-assembly was used for performance reasons? But I don't have anything to benchmark that on, so the Darwin one appears good enough. Are you planning to submit your own PR?

Because our maintainers are thoughtful enough to document their hackery:

mutex is hardcoded to x86_64/gcc-assembly to avoid a detection and build
failure with Xcode 12. Actual failure is unclear but seems linked to
implicit declarations now being errors during ./configure.

[TheSin-]

I had already seen that note, but I still think it was just the first one found, likely by griping for x86_64. I still think Darwin/_spin_lock_try is the correct answer here.

[dmacks]

I agree that having a bunch of unrelated files in a single PR can get confusing, since a bunch of us are working on them or also in other PRs, and no need to wait to push it as a single set.

• apt (see also Failed: phase compiling: apt-0.5.4-1061 failed #965)
• db53-aes (Issue db53 and db60 #954, fixed by 3bcb64a)
• db60-aes (Issue db53 and db60 #954, fixed by 3bcb64a)
• nettle7 arm64/aarch64 build fixes for nettle7 and libid3tag #962
• cairo (needs pkgconf: PR Add pkgconf package as a replacement for pkgconfig #923)
• fftw3 fftw3 - add support for Apple silicon and fix preprocessor #1042
• openblas OpenBLAS update to 0.3.20 and test/benchmark fix for pre-10.12 #908
• libid3tag (Failed: phase compiling: libid3tag-0.15.1b-5 failed #998 fixed by 85bb077)
• libijs1 Failed: phase compiling: libijs1-shlibs-0.35-1 failed #964, Update config.guess for libijs1 build on arm64 #987

[dhomeier]

I did not see what was still in the works at the time; to avoid more duplicate work wanted to push at least the draft here so others can see what is already in development. It's probably best to keep this as a tracker now that packages are being updated in their own PRs.

[TheSin-]

Have you tested cario with older dpkg? I've like to commit that one if it's working with dpkg 1.10

[dhomeier]

Have you tested cario with older dpkg? I've like to commit that one if it's working with dpkg 1.10

You mean cairo? Yes, just built and even uncommented the test suite; there are a couple of failures, as noted in the comments, but config+build passes on 13.2 both archs, 10.14.6 with both dpkg 1.16 and 1.10.
But note that it BDeps on pkgconf from #923.

[TheSin-]

none of the pkg-config changes are required anymore, I fixed pkg-config tonight, selfupdate and pkg-config will start working.

[kristibektashi]

Why is apt the only unfixed package that doesn't have a separate pull request?

[TheSin-]

I just haven’t had time to move it from my exp since dpkg 1.15+ was accepted and it’s low priority since we don’t have a bindist atm