Improve code coverage when gbq credentials are not provided

[parthea]

By default, the majority of pandas-gbq unit tests are skipped on forks of pandas-gbq because Google BigQuery integration testing requires credentials. This causes code coverage to appear low on pull requests from forks. One potential way to increase the code coverage reported in pull requests from forks is to use mock to stub external requests.
https://docs.python.org/3.6/library/unittest.mock.html

[jreback]

yeah I realized that now :>

we have a very low check on the PR's anyhow

[max-sixty]

@tswast is there any interface that Google provides that would allow testing against BQ without an account / with a limited-google-provided-account? At the moment people can't run tests without a google account (and given #109, without downloading a json file and putting it in a specific path)

[tswast]

A local emulator would be a good way to provide such a use-case. Google has them for some services like Bigtable and Cloud Pub/Sub, but there isn't one for BigQuery.

[max-sixty]

OK thanks.
This might be a long-shot, but do you think Google might be willing to provide some keys we could put in the repository so, for example, any PR pushed to Travis would test?

[tswast]

I believe this project already has some secret keys used to build on merge.

Some CI systems allow accessing secret keys after a PR from a fork has been manually reviewed by adding a GitHub tag/label to the PR. I don't believe Travis has this feature, though.

[max-sixty]

ok thx

[max-sixty]

Unfortunately I don't think there's a BigQuery mocking library like there is with some other Google products. I agree that would radically improve the testing iteration.

But until then, I propose to close this as out-of-scope; unless people think I'm overestimating how difficult it would be to build the mocks

[tswast]

Yeah, no BigQuery emulator. It does make unit testing annoying / infeasible.