Message ID | 20210316062805.2890883-1-cmmarusich@gmail.com |
---|---|
Headers | show |
Series | Add support for powerpc64le-linux (wip-ppc64le-for-master) | expand |
As discussed on IRC, it turns out no world rebuild happens: $ ./pre-inst-env guix build --system=x86_64-linux -d --no-grafts texlive-bin texlive-latex-base libelf bdb sed hello gcc-toolchain /gnu/store/9f3vsmg2c9qa91r1dlffbnpr0kbzpvxy-gcc-toolchain-10.2.0.drv /gnu/store/sihk9hiiqhqkckjs4dzl2vdk5dfv2923-hello-2.10.drv /gnu/store/kmpw9qv3pdc31q3b3mnsr8ypqkcd4q95-sed-4.8.drv /gnu/store/2vq0yvhrn6mk2lr40mabhnx6r0pdrccn-bdb-6.2.32.drv /gnu/store/mj1q7a8x4hv2c5n1xqnakxhsf5fny609-libelf-0.8.13.drv /gnu/store/qhnf8nqblgz22id4fzn2bb94mbdzghpk-texlive-latex-base- 51265.drv /gnu/store/pwhz1k213dd8bsc9sq85vds1cqjwc1by-texlive-bin-20190410.drv With commit 16cc8f8851ee38dfaa74ef976b0609a837ba682d (current wip- ppc64le-for-master) which includes those patches AIUI.
Léo Le Bouter <lle-bout@zaclys.net> writes:
> As discussed on IRC, it turns out no world rebuild happens:
Yes, and Christopher Baines explained that the outputs matter, not the
derivations themselves. The outputs of hello and gcc-toolchain are the
same on master as they are on the tip of wip-ppc64le-for-master, so I
think I was just confused: we are good to merge to master, after all.
Chris Marusich <cmmarusich@gmail.com> writes: > Léo Le Bouter <lle-bout@zaclys.net> writes: > >> As discussed on IRC, it turns out no world rebuild happens: > > Yes, and Christopher Baines explained that the outputs matter, not the > derivations themselves. The outputs of hello and gcc-toolchain are the > same on master as they are on the tip of wip-ppc64le-for-master, so I > think I was just confused: we are good to merge to master, after all. Thanks for sending these patches, that means they've appeared in Patchwork [1] and that the Guix Data Service has processed it as a patch series [2]. 1: https://patches.guix-patches.cbaines.net/project/guix-patches/list/?series=7335 2: https://data.guix-patches.cbaines.net/compare?base_commit=341dfe7eda4972af0a027357015ea595314438b0&target_commit=4757434caeac0077f67583701653a7b89a335e61 Looking at that comparison, it seems good to merge! The new powerpc64le-linux derivations could be computed, and there's a relatively small number of x86_64-linux derivations affected [3], which is good, as this can be merged to master then. 3: https://data.guix-patches.cbaines.net/compare/package-derivations?base_commit=341dfe7eda4972af0a027357015ea595314438b0&target_commit=4757434caeac0077f67583701653a7b89a335e61&system=x86_64-linux&target=none&build_change=&after_name=&limit_results=&all_results=on Chris
Christopher Baines <mail@cbaines.net> writes: > Chris Marusich <cmmarusich@gmail.com> writes: > >> Léo Le Bouter <lle-bout@zaclys.net> writes: >> >>> As discussed on IRC, it turns out no world rebuild happens: >> >> Yes, and Christopher Baines explained that the outputs matter, not the >> derivations themselves. The outputs of hello and gcc-toolchain are the >> same on master as they are on the tip of wip-ppc64le-for-master, so I >> think I was just confused: we are good to merge to master, after all. > > Thanks for sending these patches, that means they've appeared in > Patchwork [1] and that the Guix Data Service has processed it as a patch > series [2]. > > 1: https://patches.guix-patches.cbaines.net/project/guix-patches/list/?series=7335 > 2: > https://data.guix-patches.cbaines.net/compare?base_commit=341dfe7eda4972af0a027357015ea595314438b0&target_commit=4757434caeac0077f67583701653a7b89a335e61 > > Looking at that comparison, it seems good to merge! The new > powerpc64le-linux derivations could be computed, and there's a > relatively small number of x86_64-linux derivations affected [3], which > is good, as this can be merged to master then. > > 3: > https://data.guix-patches.cbaines.net/compare/package-derivations?base_commit=341dfe7eda4972af0a027357015ea595314438b0&target_commit=4757434caeac0077f67583701653a7b89a335e61&system=x86_64-linux&target=none&build_change=&after_name=&limit_results=&all_results=on > > Chris Awesome! Thank you for this additional information. It's very helpful. It looks like a few hundred packages, according to that page. I've looked at the patches and I'm not sure exactly why they are rebuilt. One guess is that perhaps changes to build-side code (e.g., guix/build/syscalls.scm) are getting included in the output of some derivations (maybe module-import-compiled derivations?) whose knock-on effects down the line are not obvious. In any case, it's "only" a few hundred packages, judging by the page Chris linked. I would argue that it is better to merge this for the release and just rebuild those few hundred packages, rather than to postpone a change that will make it significantly easier for curious people to hop in and try out / help out with the powerpc64le-linux port. I think we've done a pretty good job of limiting the "rebuild" blast radius, so I think we should go ahead and merge it.
Hi, I have applied this patch series to the master branch and pushed it to Savannah. The final commit in the series is 65c46e79e0495fe4d32f6f2725d7233fff10fd70. Before pushing, I verified again that it still builds both on a Debian ppc64le machine, as well as on a Fedora amd64 machine.