search

datalad / git-annex

A non-official clone of git-annex established for DataLad purposes. No PRs will be merged, but could be used to test perspective git-annex patches. Official git-annex repository: https://git.kitenet.net/index.cgi/git-annex.git/
16 stars 3 forks source link

workflow seems to make wrong assumption about git-annex test TMPDIR #78

Closed joeyh closed 3 years ago

joeyh commented 3 years ago

https://github.com/datalad/git-annex/actions/runs/1194189463/workflow#L230

        crippled-tmp)
          # forces all test repos/paths into the VFAT FS
          export TMPDIR="$CRIPPLEDFS_PATH"

At least that comment is wrong. Setting TMPDIR does not make git-annex test run in TMPDIR. git-annex test always runs tests in the CWD. While git-annex will maybe write this or that odd file to TMPDIR, out of the entire test suite there's only one that uses the TMPDIR in a significant way (the import test).

This was actually a useful mistake, because it caught this bug https://git-annex.branchable.com/bugs/2_mac_crippled_FS__58___Unable_to_remove_all_write/. But, if your goal is to run the whole test suite on various interesting filesystems, your approach wrong.

joeyh commented 3 years ago

Setting TMPDIR does make the test suite set HOME to a temporary directory in the specified directory. Perhaps that's what this workflow meant to use it for, since HOME on a crippled FS could test how gpg sockets, etc work in that situation. Still, the comment is wrong. :)

yarikoptic commented 3 years ago

oh, thank you @joeyh on alerting us to this fact -- I would have never assumed that non-TMPDIR location is used by default for testing. I guess I would have discovered that if I ever executed git annex test somewhere outside of a writeable path. I guess we should cd ${TMPDIR:-.} and not cd $HOME (which we do now for some reason, provenance for which stems from https://github.com/datalad/datalad-extensions/commit/14f70b2acb72b346088089364426a87646bcab5d) before running the tests. Submitted #79 for that.

Setting TMPDIR does make the test suite set HOME to a temporary directory in the specified directory. Perhaps that's what this workflow meant to use it for, since HOME on a crippled FS could test how gpg sockets, etc work in that situation. Still, the comment is wrong. :)

We have a dedicated set of crippled-home and nfs-home where we set $HOME to location on those systems, and then cd to $HOME thus somewhat causing the effect of not only having $HOME there but also using them for those temporary locations annex uses... well - at least we were testing such locations ;)

But with this comment it means that we cannot fully "orthogonalize" HOME from TMPDIR location, since you would set HOME there (well -- we do similar in datalad in global tests fixture), and -tmp matrix runs will also have HOME on those systems. And most likely, if TMPDIR is set outside by github actions itself (forgot to check) our -home runs bring no value and should be removed?

joeyh commented 3 years ago

The test suite overrides HOME to a directory it creates in the TMPDIR, in order to not affect the actual HOME. So your -home passes should probably set TMPDIR to where they want the home directory to be contained.

joeyh commented 3 years ago

I would not worry much about testing HOME orthagonal to TMPDIR.

yarikoptic commented 3 years ago

ok, since it was "a useful mistake", I would just close #79 without merging. I have just adjusted the comments in 6cc47cbb537861aa8cf9ff3b57dfdf22d7e06d7a to reflect the meaning of those test scenarios. Absence of functionality changes would allow for easier longitudinal analysis across builds history collected by con/tinuous.