Improve testing coverage

Pandas coverage is currently at 78%, not bad but could be better.
Open issue for first-time contributors: find some ill-tested corner
and bring it into the light.

[KTAtkinson]

Do you have a document that tracks testing coverage? If not would producing such a document/wiki be helpful?

[jreback]

their is some code for using the coverage package to measure this

would appreciate a section / how to on the wiki though

it's possible need some more / better exhorting to make this more user friendly

pls experiment and let us know

[KTAtkinson]

So you mainly look at code coverage? Within your testing is there any concept of test flows or test cases? Tests that are larger than unit tests, which is all I've really encountered.

[jorisvandenbossche]

Would there be interest to use coveralls to track testing coverage? That would be a handy way to see how coverage is for the different modules, and to see where more work is needed.

Example: https://coveralls.io/files/196422109
(based on this commit: jorisvandenbossche@237d784)

You have also the possibility to let coveralls comment on PRs, but I find that sometimes a bit too much (eg see PRs of scipy https://github.com/scipy/scipy/pulls).

[hayd]

@jorisvandenbossche IMO coveralls would be great! Does this slow down the build much?

[jreback]

should be pretty straightforward to add this as @jorisvandenbossche PR almost there.

any takers?

I can turn on the github service so we get notifications of this.

[gliptak]

@jreback Was coveralls enabled? I see codecov but not coveralls in .travis.yml

There also doesn't seem to be https://coveralls.io/github/pydata/pandas?branch=master

[jreback]

no why would we need more than one?

[gliptak]

They offer a bit different organization of the same coverage data, so we do not need to have both ...