[h-mayorquin]

This is the solution for the failing tests on #4809 @alejoe91.

As mentioned, the reason the tests fail there is that with the recent changes on gin the annexed files are stored locked, as symlinks, and the resolve() on the extractor kwargs follows them. So the Intan path goes from a name like intan/intan_fpc_test_231117_052630/info.rhd to a name like .git/annex/objects/vf/J8/MD5E-s8082--faf8e5811c5e68da6c2fec030f6bb0f9.rhd/MD5E-s8082--faf8e5811c5e68da6c2fec030f6bb0f9.rhd. After unpickling neo no longer sees a file named info.rhd so it doesn't look for the amplifier.dat and time.dat files next to it and the recording comes back with 0 samples (NeuroNexus has the same problem with its _data.xdat file).

The solution is to use absolute() instead of resolve() which is what most of the extractors already do. This PR does that for the 12 places that used resolve().

I think that relative dumps and containers have the same problem but solving them needs more than absolute() but this will have to wait. I will open an issue when I have some time.

[alejoe91]

I think that relative dumps and containers have the same problem but solving them needs more than absolute() but this will have to wait. I will open an issue when I have some time.

I think this is an issue only for the testing suite. Using locked annexed files in containers is not a use case I would take into consideration

[h-mayorquin]

Yes, actually, it is a problem that requires 1) simlinks to another mounts (which I use) but also 2) renaming (which only git-annex does).

I think this fix here is probably enough until we get a user case of it breaking beyond it.