gcc10-bootstrap @10.5.0+universal: is it supposed to have been tested on i386+ppc?

[barracuda156]

I tried to build gcc10-bootstrap +universal on 10A190 x86 with Macports universal_archs set to i386 ppc. The build failed on linking after trying to use ppc with -m64.

Nothing at all should have been 64-bit here: it is on Core Duo, which does not even support 64-bit, and Macports is configured to build only 32-bit archs.

[catap (Kirill A. Korinsky)]

Well, I never tried to build gcc10-bootstrap +universal as i386 ppc on 10.6, but each release is tested on 10.5 which includes i386 ppc ppc64 and it builds.

[barracuda156]

Replying to catap:

Well, I never tried to build gcc10-bootstrap +universal as i386 ppc on 10.6, but each release is tested on 10.5 which includes i386 ppc ppc64 and it builds.

On 10.6 (any version up to 10.6.8) meaningful archs, I guess, will be x86_64, i386 and ppc on 64-bit hardware and i386 and ppc on 32-bit.

[barracuda156]

Replying to catap:

1. This might be a bug in the muniversal 1.0 PG and not specific to the port itself. But not sure, maybe the second thing is breaking it.

2. Why do we need multiarch and multilib both? I seems that we only need multiarch but not multilib: ​https://unix.stackexchange.com/questions/458069/multilib-and-multiarch

Or ideally it should be fine-tuned somehow to a choices which actually work and to save unneeded compilation time.

[kencu (Ken)]

mulitiarch is not useful on darwin/macports and can be removed without affecting anything.

​https://github.com/gcc-mirror/gcc/blob/9693459e030977d6e906ea7eb587ed09ee4fddbd/gcc/configure.ac#L894

This was mentioned when the port was initially pushed, as well… but as it doesn’tvactually *break* anything (it’s just ignored) I dropped the point.

[barracuda156]

Replying to kencu:

mulitiarch is not useful on darwin/macports and can be removed without affecting anything.

​https://github.com/gcc-mirror/gcc/blob/9693459e030977d6e906ea7eb587ed09ee4fddbd/gcc/configure.ac#L894

This was mentioned when the port was initially pushed, as well… but as it doesn’tvactually *break* anything (it’s just ignored) I dropped the point.

Apparently enabling multiarch without multilib did break the build for me now on i386 (well, it is broken either way, but now differently). So my suggestion was either wrong or redundant.

Anyway, I give up for now, gonna return to this later. I can survive without ppc on i386.