Bugfix for significant_digits (signed zero); Numpy-friendlyness; Issue #49; Failing tests for #50;

[Bernhard10]

This pullrequests makes several small changes. Fell free to merge only selected commits if desired.
Most commit contain changes to the main source code and tests.

• I add 2 failing tests to document issue Comparing based on hashes does not work with the significant_digits option #50
• I compare self.down.t1 to None instead of using if self.down.ta in auto_generate_child_rel, because numpy arrays rise an error when used as booleans. I hope this changed behaviour does not introduce any bugs with other edge-cases.
• I added a bugfix to __diff_numbers for the significant_digits option. Now -0 and +0 compare equal.
• Finally I use __diff_dict instead of __diff_iterable for the comparison of collections.abc.Mapping instances (before __diff_dict required the container to be a MutableMapping) - see issue Mappings should be compared with __diff_dict instead of __diff_iterable  #49

[victorhahncastell]

Hi @Bernhard10,

thanks for your PR! I'm gonna go ahead and assign this to myself as it concerns stuff I originally introduced. I'm not sure when exactly I'll get the chance to look into this, but I certainly will :)

Best,
Victor

[seperman]

Hey @Bernhard10
Thanks for the reporting the issue and the PR! I made some aesthetic changes to the code and variable names to make it easier to understand. Your PR has some merge conflicts now. Could you please resolve the conflicts so we can review it? The main changes you will notice are ref has become tree and there are more docstrings.

[Bernhard10]

Hey @seperman
I rebased my pull-request onto the newest version of dev.

[seperman]

Hey @Bernhard10 Thanks for rebasing your PR. It seems like 2 of your tests are failing. Can you please check them out?

[Bernhard10]

Hi @seperman
Yeah, they fail on purpose. They are just here to document issue #50.

I guess actually fixing this issue is not so trivial. I guess I could try to work on it, but it would probably take me a few weeks.

So for now, we can either remove the two tests or mark them as @unittest.skip or @unittest.expectedFailure

[seperman]

@Bernhard10 Cool. Yes please mark them with @unittest.expectedFailure so we can merge this PR. Then when you have a chance please look at the failed cases. I'm focusing on some other bugs at the moment. Maybe @victorhahncastell will be available later to look into the failed cases too! Thanks!

[coveralls]

Coverage remained the same at 100.0% when pulling 5418868 on Bernhard10:dev into 8a33302 on seperman:dev.

[coveralls]

Coverage remained the same at 100.0% when pulling 5418868 on Bernhard10:dev into 8a33302 on seperman:dev.

[coveralls]

Coverage remained the same at 100.0% when pulling 5418868 on Bernhard10:dev into 8a33302 on seperman:dev.

[coveralls]

Coverage remained the same at 100.0% when pulling 4c85626 on Bernhard10:dev into 8a33302 on seperman:dev.

[coveralls]

Coverage remained the same at 100.0% when pulling 4c85626 on Bernhard10:dev into 8a33302 on seperman:dev.

[coveralls]

Coverage remained the same at 100.0% when pulling 4c85626 on Bernhard10:dev into 8a33302 on seperman:dev.

[Bernhard10]

Some tests fail on pypy, but this seems to be a bug in travis-ci, which tries to install the wrong version of numpy (pypy has it's own fork).
See travis-ci/travis-ci#3629

Skipping these Tests on pypy does not seem to work either.

These tests are necessary, because a user might want to compare objects containing numpy arrays with deepdiff. Unfortunately, the universal if iterable raises an error, if iterable is an numpy array.

[coveralls]

Coverage remained the same at 100.0% when pulling f5fec5e on Bernhard10:dev into 8a33302 on seperman:dev.

[seperman]

@Bernhard10 Thanks for digging into the issue with Travis-ci. We have some tests that detect if they are run under Pypy and then they behave differently. So I think maybe we can go around the Travis-ci bug for now that way till they fix it.