ENH: appveyor -- use newer git-annex + try to re-enable codecov submission #5125
ok, not too bad -- 2 failures
in addition to one known one
pushed workaround (well -- just not expecting expected behavior, no change in "windows testiness" since were not testing before either due to older annex version) for the first failure. The 2nd failure -- looks legit issue with annex that it does add a commit to git-annex branch even if was ran with always.commit set to False, doesn't replicate locally with crippled fs, so might be indeed windows specific, and regression (if was ran/passed before, didn't double check). So needs a skip (to get going) or someone to look into it/report. your desires? (possibly relevant "recentish" PR #4405 which relates to that test/code)
oh, and that known one for test_get_flexible_source_candidates_for_submodule is #4932 and apparently it failed not on windows before but on conda linux while testing standalone build.
edit: standalone - so might be git version related??
there fail was
and here (only in the 2nd run on appveyor)
so seems to be the same and the
Apparently git installation already comes with magic.mgc database on windows. Would you like to replace the existing file: Path: C:\\Program Files\Git\usr\share\misc\magic.mgc Size: 5615456 bytes (5484 KiB) Modified: 2019-12-10 18:24:42 with the file from archive: Path: usr\share\misc\magic.mgc Modified: 2020-11-03 15:02:02 ? (Y)es / (N)o / (A)lways / (S)kip all / A(u)to rename all / (Q)uit? Is it a "real" one or somehow created by some initial call of something linked against libmagic -- i do not know. I guess we need to test how "interactive" installation of git-annex would go -- would it alert about overrides or not, and either official git installation comes with magic.mgc
…nt annex Symlinks support on windows is still something not everyone would have. We had been skipping this test on appveyor since we had no recent annex there. So this does not change status quo but allows test to pass on appveyor
It cannot handle it properly. Fixes dataladgh-4333
I think we should not invest in broad compatibility right now, but focus on a narrow slice, and get that working properly. I don't see any reason to not go with the latest, which should be most similar to real Win10 boxes of real users.