braddr / d-tester

Automated testing for github projects.
http://d.puremagic.com/test-results/
11 stars 5 forks source link

finish gdc pull testing #9

Open braddr opened 11 years ago

braddr commented 11 years ago

Might work, needs to be tested then enabled

ibuclaw commented 6 years ago

One thing I usually notice is that PRs that update gcc version usually fail. Not sure how automated that is on your side (There's a maintained gcc.version file)

braddr commented 6 years ago

Pull requests have been doing for gdc since I don't know when.. years.

However there's a problem that I haven't addressed, cross project prioritization. The queue is prioritized without considering project, strictly a descending order of most recent commit date per pull for the set of to be built pulls. With the dmd master branch getting so much activity, it tends to trump the queue all the time. Even a week or so old pull request from other projects won't be built until there's a lull in dmd master merging.

Related to the gcc version updates, those have also been supported for years, since we talked about and set up that file. Until this weekend though there was a problem (which I didn't know about until I dug in based on your comment) in that if the gcc mirror didn't yet have the version being requested, it would save the error html, which would obviously fail to work properly. I put in a quick fix, but it needs more work. Now it at least detects the mirror not having the file correctly and tries another mirror. After 2 it gives up and fails the build. The problem is that nothing knows to re-try it based on time alone. Future changes to the pull request or to the main branch will trigger rebuilds as per usual.

ibuclaw commented 6 years ago

Ah, which mirror do you use? Maybe the failures I see happen because you're a day or two behind...

braddr commented 6 years ago

Until this weekend, just www.netgull.com. I added gcc.skazkaforyou.com as a fall back yesterday. I'm sure it's just a sync rate issue. After all, yesterday's version update was to 1/21 ON 1/21. That's sort of just asking for sync issues to come in to play. But it was a good way to diagnose the issue, so not all bad. I checked a half dozen or so mirrors yesterday to find one with the snapshot and skaz was the first one that had it, so I ran with it.