add arm64 CI #2978
add arm64 CI #2978
Conversation
I already have this set up and working with my own VPS for arm64 testing at https://gitlab.com/shawnl1/zig/-/jobs But that is a fork of zig, because I cannot setup the web hook, so that it will trigger on commits (to any branch) and pull requests and also show up in the PR interface with the other CIs.
|
I had a quick chat with Drew from SourceHut and he said he'd bump up the priority of getting arm64 build images available. Let's give him a week or two and see if that pans out. If not we can go with GitLab for now. |
|
Hmm even if the SourceHut one pans out, I think it's going to be qemu-based builds. So this GitLab Runner idea is probably our best bet, assuming this VPS is native hardware. Thanks for starting this effort @shawnl, I'll try to get this working over the next couple weeks. |
|
SourceHut now supports arm64 for debian/stable, debian/unstable, and debian/testing builds. |
|
@ddevault Is it real hardware? qemu is not worth it, and is more problems that solutions. |
|
No, it's qemu. What have been your problems with qemu in practice? |
|
bugs bugs bugs bugs. Especially with the SIMD. We need to focus on bugs in Zig, not qemu. Also qemu just won't be fast enough, as the test suite is quite long. |
|
I haven't heard of any significant bugs with qemu's arm64 emulation, and I'm willing to give it the benefit of the doubt. |
|
This was also the experience that Mozilla had, and the QA team had to prioritize getting real hardware to avoid their developers spending their time fixing the android emulator (which is almost certainly qemu). |
|
I do not give the android emulator the benefit of the doubt. It's based on a grossly outdated version of qemu. |
|
Let's just try it out. If it's problematic, we'll find out soon enough. |
|
Here's a sr.ht build running. Let's see how long it takes. I've set it up to only run |
|
The build timed out after 3 hours, and it didn't even finish compiling the C++ code, so I think it's going to be native arm64 testing for us. |
|
Fair. I'm curious to see how scaleaway fares. |
| @@ -0,0 +1,10 @@ | |||
| linux-arm64: | |||
| script: | |||
There was a problem hiding this comment.
please add tags of linux and arm64.
There was a problem hiding this comment.
You know the syntax better.
| @@ -0,0 +1,10 @@ | |||
| linux-arm64: | |||
There was a problem hiding this comment.
Do we need multiple jobs for musl vs glibc?
There was a problem hiding this comment.
the azure script didn't have that, so I didn't think of it.
| @@ -0,0 +1,10 @@ | |||
| linux-arm64: | |||
| script: | |||
| - uname -a | |||
There was a problem hiding this comment.
Remove these before merging?
There was a problem hiding this comment.
no, this is because the logs dont have this data, and when you read the logs it is quite nice to have
| - free -h | ||
| - mkdir build | ||
| - cd build | ||
| - PATH=/usr/lib/ccache:$PATH cmake .. -DCMAKE_BUILD_TYPE=Release -GNinja |
There was a problem hiding this comment.
You assume ccache is installed?
There was a problem hiding this comment.
this doesn't assume that, but will use it if it is installed
| - free -h | ||
| - mkdir build | ||
| - cd build | ||
| - PATH=/usr/lib/ccache:$PATH cmake .. -DCMAKE_BUILD_TYPE=Release -GNinja |
There was a problem hiding this comment.
automatically detects number of CPUs
| @@ -0,0 +1,10 @@ | |||
| linux-arm64: | |||
There was a problem hiding this comment.
What base image do you want?
(please include an image: field)
There was a problem hiding this comment.
I was just doing a shell job, that way ccache can be used without a bunch of hacks, and it is just much simpler, as Zig has no need for root.
Feel free to do it your own way.
| @@ -0,0 +1,10 @@ | |||
| linux-arm64: | |||
There was a problem hiding this comment.
Why is this in an "arm" subdirectory? Doesn't the file have to exist in the root of the repository? (or maybe in a .ci directory? I'm not sure)
There was a problem hiding this comment.
Doesn't the file have to exist in the root of the repository? (or maybe in a .ci directory? I'm not sure)
There is a gitlab setting to use a differn't path.
|
GitLab doesn't support arm64 yet either: https://gitlab.com/gitlab-org/gitlab-runner/merge_requests/725#note_198579090 |
|
It works just fine on arm64. I had it working. Then daurminator told me he
had this one.
El jue., 1 ago. 2019 3:24, Andrew Kelley <[email protected]>
escribió:
… Closed #2978 <#2978>.
—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
<#2978?email_source=notifications&email_token=AAD4W4SMSOSOY5XDWPMDHE3QCKMTLA5CNFSM4IH7XGLKYY3PNVWWK3TUL52HS4DFWZEXG43VMVCXMZLOORHG65DJMZUWGYLUNFXW5KTDN5WW2ZLOORPWSZGOS2FRFRY#event-2525696711>,
or mute the thread
<https://github.com/notifications/unsubscribe-auth/AAD4W4WNG4CPID2LLGQ5M4TQCKMTLANCNFSM4IH7XGLA>
.
|
|
How did you get past the linked issue? The official stance of GitLab is that ARM64 does not work on GitLab Runner yet. |
|
I used shell mode, and not docker. (This also allows me to use ccache to get really fast build times) I also turned off the default runners. |
|
I see there is a branch where this is being worked on |
I already have this set up and working with my own VPS for arm64 testing
at https://gitlab.com/shawnl1/zig/-/jobs
But that is a fork of zig, because I cannot setup the web hook, so that
it will trigger on commits (to any branch) and pull requests and also
show up in the PR interface with the other CIs.
So this also requires some manual coordination with Andrew
https://docs.gitlab.com/ee/ci/ci_cd_for_external_repos/github_integration.html
Andrew could also set up an official gitlab mirror, and then just authenticate the arm64 CI runner I've set up. That would probably be the easiest, using those instructions, and then PMing me with the authtication token:
https://docs.gitlab.com/runner/register/index.html.