Skip to content

Conversation

@wilzbach
Copy link
Contributor

Similar to #287

@dlang-bot
Copy link
Collaborator

Thanks for your pull request, @wilzbach!

@wilzbach
Copy link
Contributor Author

dmd build failures are due to the test suite requiring gdb 8 :/
Not sure whether we can fix the testsuite or it's easier to install a newer gdb instance on all agents.
And probably the druntime timeouts are the same as we see on SemaphoreCI.

@wilzbach wilzbach changed the title Run DMD's + Druntime's testsuite on Buildkite Run DMD's testsuite on Buildkite Aug 24, 2018
@wilzbach
Copy link
Contributor Author

As this seems more complicated I splitted off the druntime test build into a separate PR: #301

@wilzbach
Copy link
Contributor Author

Not sure whether we can fix the testsuite or it's easier to install a newer gdb instance on all agents.

Alternatively, we could skip the gdb tests for now.

@wilzbach
Copy link
Contributor Author

This is still blocked on test_results/runnable/gdb14330_0.gdb requiring gdb 8.

@wilzbach wilzbach force-pushed the dmd-druntime branch 2 times, most recently from 1e272a5 to 6a39968 Compare October 19, 2018 23:07
@wilzbach wilzbach force-pushed the dmd-druntime branch 3 times, most recently from 8e2384a to 49e6798 Compare October 27, 2018 07:35
@wilzbach wilzbach force-pushed the dmd-druntime branch 3 times, most recently from 62fa26e to 5dfe9e8 Compare November 15, 2018 05:30
@wilzbach wilzbach force-pushed the dmd-druntime branch 3 times, most recently from c3e7e6e to 54407ef Compare December 21, 2018 12:46
@MartinNowak
Copy link
Member

MartinNowak commented Dec 23, 2018

I'm not entirely on board with the direction this takes.

  • Integrating into a single CI is nice
  • Moving to a managed CI service is nice.
  • This and other moves significantly increase the compute requirements, if that's the path we're taking, we should get the D Language Foundation onboard for the server costs.
  • Please talk to @braddr, he and some other people are doing an amazing job of running 17 machines with various OSes for us (https://auto-tester.puremagic.com/hosts/). We should get him on-board. Also the auto-tester is still using an O(N²/2) retest scheme for passing PRs, so that's quite a lot of compute capacity. Running a buildkite-agent isn't much different from running the at-client loop, and in fact there would be some benefits (easy OS updates) to run the agents in a VM.
  • Meshing all different tests into a single Buildkite pipeline has the potential to be confusing, It would be better to clearly split different test-suites into sub-pipelines (e.g. test-suite, bootstrapping, dub-projects) (Trigger Step | Buildkite Documentation).
  • Let's sort out our buildkite setup first. Use custom checkouts #358 would simplify the many scripts quite significantly, I've just added an idea there on how to implement this change.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants