Fix build pipeline

[ejose19]

No description provided.

[ejose19]

@Morganamilo can you trigger a pipeline to confirm if this fixes the issue? So #259 can cherry-pick the 2nd commit and merge.

[Morganamilo]

You should be able to test this yourself by just pushing a tag starting with v.

[Morganamilo]

I see a lot of extra stuff in this PR stuff that we were exploring in #259. Is that stuff actually needed for this to work or are you just building on that?

[ejose19]

Was waiting for pipeline (https://github.com/ejose19/paru/actions/runs/594038571) to finish to confirm you, but I managed to get it to work. Main issue is actions/runner-images#2658, which first resolution was to use actions/runner-images#2698 (comment), however that didn't work as seem gh is using an older host, so the other resolution was to use a patched version of glibc that works for linux4 (based on PR lxqt/lxqt-panel#1562) which solves the issue for x86_64 (arm build doesn't seems to be affected).

Regarding the pacman.conf changes, yes they're needed. Seems github screwed another thing so these two must be kept until solved upstream.

EDIT: Pipeline completed successfully.

[Morganamilo]

I very much hate how awful the hacks are here but it works.

[ejose19]

I agree with you, also I'm not very comfortable using an unsigned package in the base image of paru build. So either:

• Wait for an official package of glibc-linux4
• Wait for github to solve this (this started over 2 weeks, so who knows how long it will take)
• Migrate building to a more mature solution (travis, circleci)

I will close this PR for now.

[ejose19]

@Morganamilo I've continue to investigate about the issue, seems that buildx is not fully compatible with the patch of runc, when I removed buildx the 1st workaround worked without doing anything extra, so I filed moby/buildkit#1995 to monitor the progress of that particular issue.

Regarding the other issues (ro mounts), it's buildx again with moby/buildkit#1267, and applying the workaround specified in comments allows to continue with the simplest change in Dockerfile.

So either we keep waiting for a proper solution, or we split the build.yml so aarch64 is built with buildx (aarch64 is not affected by glibc bug) and x86_64 is built using docker build directly (on which the latest runc works and there won't be a need to edit pacman.conf)

[Morganamilo]

The build worked for without the last commit, what does the last commit actually fix?

[Morganamilo]

My hope is this will be fixed by the next release. At least some of the issuess. I used this PR to build 1.3.0, hoping the next release will be a while off.

[ejose19]

This is still a work in progress, as I said I'm waiting for a reply on moby/buildkit#1995 to finish this PR. The previous "commit" included the 2nd workaround but it was only a proof of concept to confirm the issue solved, I'm pretty sure it's not the solution we want and didn't thought you would publish that one.

So I'll mark this as a draft instead but wanted to share my advances on this with you. And I doubt this will be completely fixed by next release as buildx clearly stated they won't fix the readonly mounts, so at least that fix must be included in this PR.

[ejose19]

The folks at moby already solved moby/buildkit#1995, so this PR is complete and you can merge it (there's no need to use any patch now). Here's the pipeline to confirm it.

I would recommend releasing another version and deprecate/hide current latest one at that one was built with an unsigned package, and I'm sure paru users will appreciate using only trusted builds.

[ejose19]

They already updated the default moby/buildkit:buildx-stable-1 that comes with setup-buildx-action, so I removed the reference to latest, leaving this PR with only the "essential" changes.

@Morganamilo Do you need anything else to be done here before merging?