dolphin-emu / fifoci Goto Github PK

Might be a little complex to implement as windows doesn't print logs to stdio.

Needs some careful handling: PR tests shouldn't show up in the timeline and should be tested against the appropriate baseline. Then, FifoCI should report all new differences for each new test run.

The diff is irrelevant when one side fails, so we should probably display something else/nothing instead.

In order to avoid PRs being built before their parent.

In practice it's not too much of a problem, one run of manage.py fixup_data fixes that.

For Linux NV OGL, Windows NV D3D and Windows SW D3D tests.

Non-linear history makes everything way too complicated in terms of UI, and it probably makes sense to just support running on master since in-repo branches are pretty much unused nowadays.

The /version/<hash>/json/ endpoint currently 500s when there have actually been changes. When there are no changes, it successfully returns []. For instance, 5.0-13083 had no changes, and its JSON is empty ([]). However, 5.0-13081 had changes, and its JSON simply gives <h1>Server Error (500)</h1>.

This erroring also has broken the automated comments on pull requests where changes occurred. dolphin-emu/dolphin#4715 (comment) was the last comment I could find (on February 18th, 2017), making either one of the changes from #19 or 3f2efa9 (both on February 19th, 2017) the likely change that broke it. I suspect it's something to do with type being a separate table, but I don't know for sure. Note that the actual code for the automated comments still exists in SADM, and requests that /json endpoint; I captured a log from central where the pull_request_fifoci_status occurred and also Failed to pass event to <github.GHFifoCIEditer object at 0x7f87a1d12588> so it's still getting triggered. Unfortunately there is no stacktrace, so I can't be 100% sure that it's the 500ing that's breaking it, but it seems quite likely. There also is no stacktrace for why the /json endpoint is 500ing.

FifoCI seems to be broken since 5.0-12891, or since dolphin-emu/dolphin#9182. FifoCI build 551 (sw, ogl) worked fine, while 552 failed to generate any images, with messages like ls: cannot access '/tmp/dolphin.446HbQhTNL/.dolphin-emu/Dump/Frames/*.png': No such file or directory (sw, ogl). The problem seems to be due to an assumption that the generated AVI file name is framedump0.avi, while now it is e.g. 00000000_2020-10-23_18-12-35_0.avi. The relevant code is this:

It could probably be fixed by changing it to $DUMPDIR/*.avi, but that seems a bit jank. Also, oddly, both sw and ogl create AVI files; the SW renderer path doesn't seem like it is being used.

fifoci-ogl-lin-radeon also appears to be broken, but for some other reason (and from much further back); it can't find ninja. Build 80 succeeded, build 81 failed. Not sure what happened there.