TST: only catch deprecation warnings for top-level module imports

[jreback]

No description provided.

[jorisvandenbossche]

@jreback What is the rationale here? Whas there a certain problem? The fact that those raise a warning is useful to test, no?

[jreback]

well the problem is we have pandas.json, which if you are in pandas and do an import json it imports the local module (and not from the std library). So eventually when I get rid of pandas.json this will go away.

I put this in here, because you see the warning when the _version is imported.

[jorisvandenbossche]

Sorry, I don't understand the connection between the import problem of json, and the fact that pd.json.dumps raises a FutureWarning ?

I put this in here, because you see the warning when the _version is imported.

Do you mean

In [1]: from pandas import _version

I don't see a warning there.

[jreback]

try creating a sdist, then untarring it. you will see a file pandas/_version.py, versioneer creates this.

IN this file it does a import json. BUT this is IN the pandas/ directory. guess what it picks up pandas.json. (it has always been this ways, and it just happens to work, because json.loads is works like the library one).

It actually loading the WRONG library. The deprecation warning is because we are deprecating pandas.json.

Its an unfortunate consequence of having a stdlib module name that is pretty common.

[jreback]

this ONLY happens on a package version FYI! It does NOT happen with a dev version. I noticed this when on the new build-test job, which actually does a full install then import pandas; pandas.test(), just like an end-user would do.

It shows a deprecation warning immediately.