Refactored/Simplified ish module

[snoack]

As the codebase of the ish module is growing as it increases in popularity and adopters, I found that the code needs a refactoring. Therefore I submitted this PR which does following improvements to the code:

• Turned TRUE_STRINGS and FALSE_STRINGS into a set (this is the correct, and also a more efficient, data structure as we don't mind the order of these strings but simply check whether a given string is in the given set).
• Replaced u'..'.lower() with the respective literals.
• Merged the classes TrueIsh and FalseIsh into a common class BoolIsh (this way the logic is more consistent)
  • That way I could also easily (and without duplication) implement the __repr__ method for a more natural user exprience when using the REPL.
• Moved the logic to normalize and check strings for their true-ish-ness into separate methods, also removing some unnecessary complexity there, and making that logic easier to understand.
• Simplified the module magic, by simply setting sys.modules[__name__] = ish instead subclassing the module type, as suggested in #2.

[judy2k]

This sounds great. I'll take a look when I'm back at my laptop. Thanks for contributing!

[judy2k]

Don't forget to add yourself to CONTRIBUTORS. txt!

[snoack]

Are you still at EuroPython? Maybe we can have a sprint for this project!

[judy2k]

I wish I was! I'm in Bilbao but not at the sprints :(

[snoack]

I created two more PRs that are based on this one. However, it turned out that when you create a PR based on a not yet merged branch, it will be created for your own fork instead for the upstream repo:

snoack/stupid-python-tricks/pull/1 snoack/stupid-python-tricks/pull/2

So I simply added those commits to this PR now. Sorry for the confusion.

[snoack]

Here comes another commit. It fixes a regression introduced by the initial commit in this PR. This was that the ValueError raised when the value isn't recognized, included the normalized instead the original string. I also adapted the tests to protect against that scenario.

[snoack]

I realized that stripping digits might not be a that good idea, as '1' for example would be normalized to '' (empty string), evaluating to False, then. Fixed that, only stripping whitespaces and punctuation now.

Also, I reconsidered the empty string scenario. With #5, I suggested to add '' (empty-string) to FALSE_STRINGS. However, since strings are stripped before looked-up, " " and even "?" (with the change mentioned above) would be recognized as False-ish after normalization. So instead, we should check for empty string before normalization. Or even simpler, only entering the code path with the special handling for strings if other evaluates to True. Did that with the changes above.

[snoack]

I just rebased this PR on top of #3, which I rebased on top of master before.