Just suppress the spurious IndexError when it happens. Rely on the environments where the behavior doesn't fail until the upstream issue can be resolved. Fixes #201.

Merge pkg_resources.Environment.__getitem__() code cleanup (pull request #47 and subsequent changes).

Correct sort order by using list.sort with support for 'key' and 'reverse'. This method is less efficient, but allows the tests to pass.

Restore install._install with a comment to capture its requirement on earlier NumPy versions. Fixes #199.

--HG--
extra : amend_source : a76d060c98e0048506e5aadddd675151db9d273c

--HG--
extra : amend_source : 4efa6b87d3acaefebdfcc953e78a452ffc1f160d

--HG--
extra : amend_source : 20ab7547c8478eb084767fe701e627bdd462ba16

Since paths are case-insensitive on Windows, zipped egg modules may be loaded
using different but equivalent paths. Importing each such different path causes
a new zipimporter to be instantiated. Removing cached zipimporter instances must
then not forget about removing those created for differently spelled paths to
the same replaced egg.

Other missed zipimporter instances are those used to access zipped eggs stored
inside zipped eggs. When clearing zipimporter instances got a given path, we
need to clear all the instances related to any of its subpaths as well.

--HG--
extra : rebase_source : 86aeadd1e639fbc83d27a0c551fdc2b8a68a6f85

--HG--
extra : rebase_source : 79778a670897cb92c17307f2535fcac6447e16b4

Better commented the code, especially the result caching & lazy distribution
list sorting strategies implemented there. Added several TODO comments
indicating things to look into there in the future.

General code cleanup:
 - stopped reusing the project_name variable for different purposes
 - added an assertion to detect corrupt distribution list caches

--HG--
extra : source : 86ef8f59ef4e1fc253c155b9ca0856497455372d