Closed kyleam closed 4 years ago
Thank for trying it locally. I can trigger the issue locally with the latest git-annex (7.20190912), crawler at 0.4.1, and datalad 0.11.x (60542a0da).
I can also trigger it changing datalad to the v7-default (de6b6c308, before your latest force push).
If I downgrade git-annex to 7.20190819, the test passes. Since I can trigger this, I'll try to look into it more.
A couple of notes:
I've been trying to trace the code in various places to see when the .datalad/crawler json files start being tracked by git-annex. The switch happens with the first run_pipeline() call in test_openfmri_pipeline1. During that call, there is a point where the files are in git rather than annexed, but I haven't been able to find what causes the switch.
As a brute-force approach, I tried converting all the custom git calls to use -c annex.largefiles=nothing
and the test still had the same failure.
difference from my environment: in your first output I see cmd:annex=7.20190912 cmd:bundled-git=UNKNOWN
, so it is a system wide git used then I assume, which is 2.23.0, which is the same as mine systemwide...
but that git-annex-standalone should be the same one as travis uses, and seems to be the case:
- cmd:bundled-git: 2.20.1
- cmd:annex: 7.20190912-1~ndall+1
- cmd:git: 2.20.1
so don't know yet where the difference is... but could you try with git-annex-standalone package from neurodebian?
in your first output I see
cmd:annex=7.20190912 cmd:bundled-git=UNKNOWN
, so it is a system wide git used then I assume, which is 2.23.0, which is the same as mine systemwide
Correct. I typically don't have git-annex-standalone installed unless I need it to test something.
but could you try with git-annex-standalone package from neurodebian?
Sure. I see the same failure when I use git-annex-standalone with datalad at either 0.11.x or on the v7-default branch.
I think I reproduced after downgrade of systemwide git to 1:2.20.1-2 .
Something in the back of my head says that we (or @joeyh) had some intermittent git issue which was resolved... and not sure if worth time investment ATM from our side if we know that it is not present with more recent git. What do you think @kyleam ?
may be quick and dirty workaround would be to skip this test whenever systemwide git is 2.20.1, and release datalad-crawler. I will submit a PR for that shortly so we are ready
gy gy -- that test even marked as @known_failure_v6 #FIXME
;) dedicated issue now: https://github.com/datalad/datalad/issues/3706
had some intermittent git issue which was resolved
That doesn't seem consistent with what I'm seeing. I can trigger the issue with the latest git.
git < 2.22 has the smudge memory leak, and so git-annex refuses to upgrade a direct mode repo to v7 when that version of git is in use.
I don't know of anything else specific to git 2.20.
-- see shy jo
Thank you @joeyh
In a test run (from datalad/datalad#3648) with the latest git-annex, which auto-upgrades to v7 repos, test_openfmri.test_openfmri_pipeline1 errors:
https://travis-ci.org/datalad/datalad/jobs/588507087#L1238