Bug#1028654: dpkg: add loongarch64 architecture GNU triplet
Hi! On Mon, 2023-02-20 at 21:30:14 +0800, zhangdandan wrote: > We decide to use "loongarch64-linux-gnu" as the value of the Debian loong64 > port's multiarch tuple. > The reasons for using "loongarch64-linux-gnu" are as follows: > > Firstly, we note that many of the major architectures use the -gnu style in > Debian [3]. > > - amd64: x86_64-linux-gnu > - arm64: aarch64-linux-gnu > - riscv64: riscv64-linux-gnu > > Secondly, "loongarch64-linux-gnu" is also consistent with upstream LoongArch > gcc's latest change (replacing "loongarch64-linux-gnuf64" with > "loongarch64-linux-gnu") [5][6] and the current revision of LoongArch > Toolchain Conventions [7]. Ok, so then just to confirm, the support that was already in dpkg's git and that got reverted should be fine again. If so then I'm going to revert commit f9187c8b13478824c9eff73c92a084cc50c34cad to restore the previous code, and then request a pre-approval from the release team in a couple of days. Thanks, Guillem
Bug#1028654: dpkg: add loongarch64 architecture GNU triplet
Hi maintainers, Here are some explanatory notes on the loong64 port's multiarch tuples issue. Previously, in order to be consistent with the GCC upstream and Toolchain Conventions, the multiarch tuple "loongarch64-linux-gnuf64" [1] was used for the LP64D ABI. We then restored the multiarch tuple used by the Debian gcc package from "loongarch64-linux-gnu" to "loongarch64-linux-gnuf64"[2]. Unfortunately, we didn't realize that the Debian multiarch tuples document [3] predated the gcc multiarch tuple definitions [1]. Thanks for the feedback [4] and helmut's help, we have reconsidered the matter surrounding Debian's multiarch tuple definitions. We decide to use "loongarch64-linux-gnu" as the value of the Debian loong64 port's multiarch tuple. The reasons for using "loongarch64-linux-gnu" are as follows: Firstly, we note that many of the major architectures use the -gnu style in Debian [3]. - amd64: x86_64-linux-gnu - arm64: aarch64-linux-gnu - riscv64: riscv64-linux-gnu Secondly, "loongarch64-linux-gnu" is also consistent with upstream LoongArch gcc's latest change (replacing "loongarch64-linux-gnuf64" with "loongarch64-linux-gnu") [5][6] and the current revision of LoongArch Toolchain Conventions [7]. Please reconsider the loong64 patch (with "loongarch64-linux-gnu") to dpkg. Your opinions are welcome. [1]: https://github.com/gcc-mirror/gcc/commit/b44786f64019e7#diff-a9fbd3415f77b73e7a86c365707a79c6d60311ce5b81fad000a15f970b18fedfR4971 [2]: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1027278 [3]: https://wiki.debian.org/Multiarch/Tuples [4]: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1028654 [5]: https://gcc.gnu.org/pipermail/gcc-patches/2023-February/612272.html [6]: https://github.com/gcc-mirror/gcc/commit/017849d9d88f02 [7]: https://loongson.github.io/LoongArch-Documentation/LoongArch-toolchain-conventions-EN.html#_gnu_target_triplets_and_multiarch_specifiers Thanks! Dandan Zhang
Bug#1028654: dpkg: add loongarch64 architecture GNU triplet
Dear maintainers, Happy new year. Thank you for your reply. We are carefully considering the Debian Multiarch Define of Loongarch64. Thanks, Dandan Zhang
Bug#1028654: dpkg: add loongarch64 architecture GNU triplet
On Tue, 2023-01-24 at 11:28:55 +0100, Guillem Jover wrote: > The reasons for changing the ABI at this point in time are still not > very clear to me (Helmut mentioned some reasons though), and the fact > that this was submitted to the gcc package before any discussion seems > all wrong. :/ > > This divergence between dpkg and at least gcc is currently breaking > rebootstrap, (Sorry, this should have said, "it broke rebootstrap (but is now worked around)", thanks to Helmut for spotting that.) > and the uncertainty of this all is currently blocking > a pre-approval request for the pending dpkg release. So what I think > I'm going to do is back out (revert) the loong64 support for now, and > once this is all clarified I'm happy to reintroduce it again. I'm > planning to do at least another release after the upcoming one before > the Debian release, where this could be readded. > > I'll be sending the pre-approval request later today. Thanks, Guillem
Bug#1028654: dpkg: add loongarch64 architecture GNU triplet
Hi! On Sun, 2023-01-15 at 17:45:03 +0100, Guillem Jover wrote: > On Sat, 2023-01-14 at 18:29:12 +0800, 张丹丹 wrote: > > Package: dpkg > > Version: 1.21.17 > > Severity: wishlist > > Tags: patch > > User: debian-de...@lists.debian.org > > Usertags: loongarch64 > > > - I have added loongarch64 architecture GNU triplet to match the change in > > gcc-12. > > This patch can fix problems when cross-building and ensure gcc and dpkg use > > matched combination. > > Please consider the patch attached. (Reference from rabenda...@gmail.com) > > Hmm, why was the triplet changed? I mean it's not been too long since > it got added into dpkg, and supposedly backwards compatibility should > not be much of an issue (?), but why is this really necessary? > > What does the unqualified triplet stand for now, which is supposed to > be considered the baseline? > > > - The change of gcc-12 for loongarch64, please see the bleow BugID request: > > https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1027278 > > > - Are there any missing? Please tell me. > > See above. Also the patch (if it ends up being really necessary) does > not look entirely correct, the entries should be placed before the > more general match. Attached the modified one which now also passes > the test suite («make check»). The reasons for changing the ABI at this point in time are still not very clear to me (Helmut mentioned some reasons though), and the fact that this was submitted to the gcc package before any discussion seems all wrong. :/ This divergence between dpkg and at least gcc is currently breaking rebootstrap, and the uncertainty of this all is currently blocking a pre-approval request for the pending dpkg release. So what I think I'm going to do is back out (revert) the loong64 support for now, and once this is all clarified I'm happy to reintroduce it again. I'm planning to do at least another release after the upcoming one before the Debian release, where this could be readded. I'll be sending the pre-approval request later today. Thanks, Guillem
Bug#1028654: dpkg: add loongarch64 architecture GNU triplet
On Sat, 2023-01-14 at 18:29:12 +0800, 张丹丹 wrote:
> Package: dpkg
> Version: 1.21.17
> Severity: wishlist
> Tags: patch
> User: debian-de...@lists.debian.org
> Usertags: loongarch64
> - I have added loongarch64 architecture GNU triplet to match the change in
> gcc-12.
> This patch can fix problems when cross-building and ensure gcc and dpkg use
> matched combination.
> Please consider the patch attached. (Reference from rabenda...@gmail.com)
Hmm, why was the triplet changed? I mean it's not been too long since
it got added into dpkg, and supposedly backwards compatibility should
not be much of an issue (?), but why is this really necessary?
What does the unqualified triplet stand for now, which is supposed to
be considered the baseline?
> - The change of gcc-12 for loongarch64, please see the bleow BugID request:
> https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1027278
> - Are there any missing? Please tell me.
See above. Also the patch (if it ends up being really necessary) does
not look entirely correct, the entries should be placed before the
more general match. Attached the modified one which now also passes
the test suite («make check»).
Thanks,
Guillem
diff --git i/data/ostable w/data/ostable
index 7fd2bff08..c1f9d7a7c 100644
--- i/data/ostable
+++ w/data/ostable
@@ -24,6 +24,7 @@ eabihf-gnu-linux linux-gnueabihf linux[^-]*-gnueabihf
eabi-gnu-linux linux-gnueabi linux[^-]*-gnueabi
abin32-gnu-linux linux-gnuabin32 linux[^-]*-gnuabin32
abi64-gnu-linux linux-gnuabi64 linux[^-]*-gnuabi64
+f64-gnu-linux linux-gnuf64 linux[^-]*-gnuf64
spe-gnu-linux linux-gnuspe linux[^-]*-gnuspe
x32-gnu-linux linux-gnux32 linux[^-]*-gnux32
ilp32-gnu-linux linux-gnu_ilp32 linux[^-]*-gnu_ilp32
diff --git i/data/tupletable w/data/tupletable
index a7a878f5b..06d2c4e7f 100644
--- i/data/tupletable
+++ w/data/tupletable
@@ -27,6 +27,7 @@ base-musl-linux- musl-linux-
ilp32-gnu-linux-arm64 arm64ilp32
eabihf-gnu-linux-arm armhf
eabi-gnu-linux-arm armel
+f64-gnu-linux-loong64 loong64
abin32-gnu-linux-mips64r6el mipsn32r6el
abin32-gnu-linux-mips64r6 mipsn32r6
abin32-gnu-linux-mips64el mipsn32el
diff --git i/scripts/t/Dpkg_Arch.t w/scripts/t/Dpkg_Arch.t
index 59855dfa4..f9e080f4e 100644
--- i/scripts/t/Dpkg_Arch.t
+++ w/scripts/t/Dpkg_Arch.t
@@ -16,7 +16,7 @@
use strict;
use warnings;
-use Test::More tests => 18900;
+use Test::More tests => 18907;
use_ok('Dpkg::Arch', qw(debarch_to_debtuple debarch_to_multiarch
debarch_eq debarch_is debarch_is_wildcard
Bug#1028654: dpkg: add loongarch64 architecture GNU triplet
Package: dpkg Version: 1.21.17 Severity: wishlist Tags: patch User: debian-de...@lists.debian.org Usertags: loongarch64 Hi maintainers, - I have added loongarch64 architecture GNU triplet to match the change in gcc-12. This patch can fix problems when cross-building and ensure gcc and dpkg use matched combination. Please consider the patch attached. (Reference from rabenda...@gmail.com) - The change of gcc-12 for loongarch64, please see the bleow BugID request: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1027278 - Are there any missing? Please tell me. Thanks! Dandan Zhang dpkg-1.21.17-add-loongarch64-gnu-triplet.patch Description: Binary data
