cv-download is unable to complete

[nicolas-lelouche]

Running the cv-download command returns the following error on macOS.

[rtbs-dev]

Interesting, this should have been solved with v0.2.1 at commit 78459c5.

• What version is being installed?
• What mechanism is being used to install? (pip install cv-py vs. poetry install from a cloned repo? Other?)

Dependency should be set to pyarrow = "^3.0.0" if installing cv-py 0.2.1

[nicolas-lelouche]

hmm yeah. Looking at the error message again in its entirety, something seems off:

The command seems to load another cv-py version, and then goes off into downloading the wrong pyarrow version.

This is especially confusing since I'm able to confirm that version 0.2.1 of cv-py is installed (or at least that's what pip tells me).

The install was done through pip for this specific environment. No version was specified though.

[rtbs-dev]

I think I know what's happening here. cv-py tries to ensure compatibility with the cord19 datapackages by grabbing tags compliant to the __compatible__ variable here.

Unfortunately, the way pre-dvc (#2 ) workflow sets up in the cord19-cdcs repo is to tag releases when new data gets pushed... which hasn't happened for a while due to github file size limits and other schema errors.

In other words, the code for bugfixes in the cord19-cdcs repo isn't getting used because no tagged release has been done with our jenkins build that incorporates those bugfixes. So I need to repair jenkins first before we use cv-download to get the cdcs data.

In the meantime can you verify the cv-download tool is working for other datapackages? use the -r option with one of the strings listed here.

[nicolas-lelouche]

Makes sense.

Unfortunately it seems like there's another issue (which should be a lot simpler to solve fortunately) for other data packages:
Here is the error I'm getting for package en_core_sci_sm, but the same error occurs for every other package:

The error seems to come from this piece of code where the same __compatible__ variable is checked. The problem is that the other data packages are not compliant to that variable either (ie. they're not there).

I'll try to get a quick PR out adding the data packages.