Re: Adding loong architecture to dpkg

[Guillem Jover <guillem@debian.org>]

Hi!

On Tue, 2022-11-01 at 19:45:11 +0800, 桑猛 wrote:
> > We hope that the arch name can use loongarch64, mainly based on the
> > following considerations:
> > 1. The current upstream compilers gcc, llvm, etc. use loongarch64.

That's part of the GNU triplet name, which is related but has never
needed to match the dpkg architecture.

> > 2. The current upstream kernel uses loongarch64.

Same here.

> > 3. At present, there is no special convention for the length of the
> architecture name. If loong64 is used, it is easier to be confused
> with the previously used loongarch64.

Well, length might be one aspect of this, as it ends up in filenames,
and metadata, etc, but the «arch» part is just redundant and wasteful
and gives no meaningful information. We had a similar situation with
aarch64 which was a terrible name, which we named as arm64 for the
dpkg architecture.

> The following code in https://man7.org/linux/man-pages/man1/dpkg-architecture.1.html:
> 
>  DEB_BUILD_GNU_TYPE ?= $(shell dpkg-architecture -qDEB_BUILD_GNU_TYPE)
>            DEB_HOST_GNU_TYPE ?= $(shell dpkg-architecture -qDEB_HOST_GNU_TYPE)
>            […]
>            ifeq ($(DEB_BUILD_GNU_TYPE), $(DEB_HOST_GNU_TYPE))
>              confflags += --build=$(DEB_HOST_GNU_TYPE)
>            else
>              confflags += --build=$(DEB_BUILD_GNU_TYPE) \
>                           --host=$(DEB_HOST_GNU_TYPE)
>            endif
>            […]
>            ./configure $(confflags)
> 
> If you use loong64, it will be contrary to the loongarch64 logo used by
> gcc, so we still want to use loongarch64 as the arch name。

This is not a very compelling argument TBH, many dpkg architectures,
more so the ones most used, have GNU triplet CPU names not matching at
all the dpkg architecture: aarch64 → arm64, x86_64 → amd64. We also
have stuff like powerpc64le → ppc64el (which changes both CPU name and
endianness annotation).

Given that any such architecture will have to be bootstrapped from
zero for Debian, reusing an existing name is also not a very
compelling argument. If in addition, as Zhang Ning mentions, those
pre-existing architectures are also not even binary compatible, then
that's even a stronger argument against even considering reusing it at
all.

In any case, the loong64 name has already been used by other
distributions (Gentoo, ArchLinux, Slackware, etc.), so this is really
not unusual.

Thanks,
Guillem