unfortunately CoqIDE is the only IDE that gets automatic testing in the Coq Platform and so on. Would be useful to add for VsCode, but probably takes quite a bit of CI engineering
As a lower-tech fix, anything that needs coqide changes needs vscoq changes
This rule could be checked in a checklist or automatically (insert usual SwEng tradeoffs).
Of course, that assumes that the protocol changes broke coqide and that ci notices; neither is obvious. Does CI even test using coqide? That's not easy I guess...
yeah, I think at a high level of abstraction the XML protocol is considered legacy/deprecated
we would like to have proper language server, but this going to be WIP for the foreseeable future
okay, but doesn't that round that up to "state of the art and fully supported"? I hope it's better supported than the coqtop protocol...
Karl Palmskog said:
unfortunately CoqIDE is the only IDE that gets automatic testing in the Coq Platform and so on. Would be useful to add for VsCode, but probably takes quite a bit of CI engineering
Not true. Coqtail (which also uses the XML protocol) is also in Coq's CI. And VsCoq has an unreleased branch (the document manager) in it as well.
OK, so some testing in Coq's CI, but not in the Platform CI I think?
I don't think one can run VSCode on headless CI servers, so how should VsCoq be tested? Also testing UI software is in general a quite involved business, unless you just want to test if it starts at all.
One could look at what CoqIDE and Coqtail testing look like and try to be on par with them (even if this probably does not mean actual UI testing).
of course there’s more, but a head for headless servers is an apt-get install
away with xvfb (untested link: http://elementalselenium.com/tips/38-headless)
I am definitely open for PRs on this, but I don't have the bandwidth for this right now.
Until somebody does, maybe just ban changes to the protocol until testing is in place.
Including the "backward-compatible" changes that have been attempted — the last one I saw had to be reverted.
cognitive bias warning: I'd probably miss any successful changes!
:eyes: hello, I made a few changes to VsCoq but I probably can't make more changes unless there are tests. I really need tests to detect issues quickly. I plan to add some headless browser tests. but could I ask... I want to run my tests on CI, would it be ok if I switch the CI to using GitHub Actions?
Sounds reasonable to me, I don't think we can trust Travis much anyway.
I don't think @Maxime Dénès has anything against swapping out Travis for VsCoq CI either?
Adding tests is actually something we do want for VsCoq 2, so adding the necessary infrastructure would be very welcome! Ideally, if we could setup something reasonably standard for testing vscode extensions, it would help reusing it. And of course, I'm in favor of switching from Travis to GitHub Actions.
from another similar discussion: https://github.com/redhat-developer/vscode-extension-tester
Also the existing unit tests are broken right? I can't run them. Normally unit tests are very easy to run.
Bumping. I just want to make sure my perception is correct. You can just answer "yes" if this is true.
@Huỳnh Trần Khanh yes, please go ahead and switch to GitHub Actions CI for VsCoq. And adding more automated tests is good.
I think the bump was about unit tests being broken.
OK, then we'll have to ping @Maxime Dénès to check on the "VsCoq unit tests are currently broken" point
It could well be. At this point, as I wrote, our energy is focused on VsCoq 2, and resolving the blockers on the Coq side. So I haven't checked for a while the status of unit tests in the current branch.
Last updated: Jun 04 2023 at 23:30 UTC