CI: Port Windows build to Azure Pipelines #7757
Conversation
|
@hcorion Have a look please. |
|
Travis still wins by speed (more correct cache inheritance), so there is no need to remove it. |
|
Well there's something fishy afoot. There must be something wrong with the way the Azure is configured on RPCS3's end. On mine, the linux caching is excellent and beats out travis. You can see it here. Windows builds are at least faster than appveyor, so we can yeet that into the sun. |
JohnHolmesII
force-pushed
the
azure
branch
2 times, most recently
from
5515b33
to
b91a90d
Compare
JohnHolmesII
changed the title
|
Well, I think due to the issues with RPCS3's azure pipeline being temperamental, the other changes I had planned are best reserved for another time. In addition, there is a dearth of options available for build caching on Windows. I'm looking into something rather hacky, but even if it works, it's best reserved for another PR. Azure is at least as fast as appveyor was, and is usually faster, so I think the PR is mergeable as is at this point. |
|
Nice work! |
|
If you remove Appveyor I need to update the website as well so don't merge before that Edit: No update was needed, I fixed what I was misremembering some time ago |
JohnHolmesII
changed the title
|
@hcorion Wow, thanks for reminding me, I completely forgot. Deployment is set up and working. @Nekotekina In order for deployments to work, Azure will need more permissions. In azure pipelines, select the rpcs3 pipeline and there should be a prompt that the pipeline needs "updated permissions" in order to run. |
JohnHolmesII
changed the title
|
Need to add cacert.pem download to the windows artifact. |
|
@JohnHolmesII Can you just use the same github connection as llvm-mirror? |
|
Well, there seems to be a bug in the way Azure handles variables set from scripts. Even simple strings get corrupted with a mysterious roaming apostrophe, which is nondeterministic. I will likely need to manually scrub vars that are passed around. Do not merge until this issue is solved, or else we could corrupt various stages of deployment. |
|
Do does it technically work now? |
|
@Nekotekina The pr works sometimes. Any environment variables exported can sometimes be corrupted by a single quote. It is random. Look at the Artifact for the windows build, a single quote is in the filename. SOmetimes the single quote shows up in the "BRANCH" variable, which breaks compiliation, since it is expanded by macros in main_window.cpp. I think I'll either have to report and wait for a response from Azure, or do a workaround by making that code read in a file or something. I already put in a workaround for the GitHub release description, but there is work needed for others. |
|
Alright, well I've worked around all but one of the side effects. I think the changes to main_window are even for the best: rpcs3 version checking should be unified imo. @AniLeo The "all but one" just happens to be something related to the release artifacts. The title of the release will have an apostrophe at the end, as seen here. The description part (that has the hash and filesize) is properly protected against the bug, so it should be easier to parse. I think it should be easy to scrape the title tho and simply discard the final apostrophe, but I don't know the details of that on the website's end. Other than that, I think is should be squared away. |
|
As the website is right now it will work fine since it uses the tag name in URL which is correct, but you may want to fix that release name either way |
JohnHolmesII
force-pushed
the
azure
branch
2 times, most recently
from
4dcc915
to
24f1b62
Compare
|
@Nekotekina It doesn't appear that the llvm libs have sha256 attatched for windows. Can you add that so I can make the Windows build automatically redownload based on checksum? Same thing with glslang. |
The STRINGIZE macro was a little awkward, and difficult to control at configure time. Since other version information is already included, the full branch name is now added as a function. It's runtime instead of compile-time checking, but it seems worth it.
Previously, there was no way of forcing a re-download of cached dependencies when they were replaced by new ones. In addition, there was really no verification of downloads or cache. Now, changing a few lines at the top of the file will automagically force a cache update.
|
Well, in the mean time I simply inlined the hash for now. The script can always be changed later I suppose. In addition, I made changes to how caching dependencies is done. The previous setup was rather poor, but now it is quite robust. |
This is the beginning work of a port from Appveyor to Azure. It does not change compilers for windows.
It will build and package an artifact, however there are several things that need to be done before this can be merged:
Unify the yaml itself so Windows is inside the matrixFix the caching system. This is desperately needed.Overall it looks promising. The actual build steps for clang, gcc, and msvc are all the same. Hopefully when all the caching is set up, we can get 5 minute CI.
Optional steps:
Remove travis