@pocmatos see commit 29a0c44c98d5c8d3660e4bd10eb566c7ec6e46e2
R088 pkgs/racket-pkgs/racket-test/tests/racket/mz-tests.rktl pkgs/racket-pkgs/racket-test/tests/racket/core-tests.rktl
It’s renamed
Thanks for looking into this. I rushed out to a doctors appt and had no time to check what was happening.
Maybe it’s something that needs to be patched up.
If you don’t do it in the meantime, I will sort it out when I get home.
Not sure if it’s better to do this or to remove the file altogehter - hopefully @mflatt or @samth have better insight into this.
@pocmatos I think Matthew does mean that configuration of the tests on a per-vm basis should be in Racket code, rather than in the yaml config file.
I think I understood. As in checking if we are running cgc then we don’t run this section. That would make sense for me if this was a long-term solution but for a short term solution (until the tests are fixed) I found it more straightforward to do it in yaml. As I mentioned though, I am happy to make the changes in the racket tests files instead. I was hoping for him to come back to me on the bug. However, if I should just scrape the changes in yaml and do it in Racket, I can proceed like that.
I was generally avoiding touching the Racket files for ci purposes. :slightly_smiling_face:
I think Matthew’s point is that when running the tests in other settings you want the same configuration
Ah!!!! That makes total sense and I didn’t see it from that point of view.
OK - let me rethink the PR then. :slightly_smiling_face:
One thing to note here is that there are two cultures for how to think about CI — one is “CI is where tests get run” and you base artifacts, acceptance decisions, etc on that. The other is “CI is a backup test runner” and the key running of tests is done manually. Racket has for a long time been in the second category (partly because it predates wide use of CI, for example). Over the years I’ve tried to push more into automation but there’s still a long way to go.
@samth I am totally with you here and completely understand, given how old Racket is, that there’s baggage that influences the current situation. Will try to, in any way I can, to push in the same direction you’re pushing.
You’re already doing lots!
From my point of view, Racket has been here for 20 years and will be here for at least 20 more, so we have time to set things straight. :wink:
That’s the idea :)
Interesting - something has gone wrong installing Racket on macos : https://github.com/LinkiTools/racket/runs/323531933
has been installing for over 3hrs.
Also, just noticed Matthew added a new flag to configure. At some point we should discuss which configurations we should test and whatnot.
I moved the test skip as you suggested into will.rktl. It seems to be harder to find exactly which tests failed in sync.rktl due to the sheer amount of tests that look the same.
Currently we only test the standard configuration plus a couple —disable things on Travis
And mostly that’s ok
Catching up on some of the CI-related github issues, and I’m wondering: why is the cgc variant of racket still around?
is it purely backwards compatibility? and why would code be compatible with cgc but not 3m?
is it something to do with embedding racket in other executables?
Hi @notjack, well cgc is needed to build 3m. :slightly_smiling_face: Part of the bootstrapping process. What’s not needed and I have ongoing work trying to remove is boehm (https://github.com/racket/racket/pull/2665) but I got a bit stuck dealing with VS files.
Because the difference between both is GC, some code doesn’t work so well with the CGC or sometimes there’s bugs in one but not the other and since CGC is used mainly for bootstrapping, there’s not been (I think) motivation to actually fix it all.
Interesting. Why can’t 3m build 3m? Not that it matters much tho, now that we are moving toward RacketCS.
Right.
Well, most of the times an existing 3m can build a new 3m, however sometimes if there’s any deeply embedded code change like bytecode change then we need to bootstrap using cgc.
So we can’t really get rid of it.
@notjack I found out today after a stupid mistake that Actions doesn’t yet support yaml anchors. I am disappointed and hoping the github guys are working day and night to bring this to us.
s/today/yesterday/