Poor error reporting when verlib.suggest_normalized_version returns None

[daira]

<warner> I'm trying to test tahoe (trunk) against a foolscap branch that I'm working on <warner> and I'm thwarted by packaging and setuptools, again [...] <warner> ... "PYTHONPATH=/path/to/my/foolscap/tree ./bin/tahoe" ... currently emits: PackagingError?: could not parse actual version '0.8.0+12.g1fdaa3e.dirty' of foolscap from '/Users/warner/stuff/python/foolscap' due to TypeError?: expected string or buffer <warner> Warning: version number '0.8.0-12.g1fdaa3e.dirty' found for dependency 'foolscap' by pkg_resources could not be parsed. The version found by import was '0.8.0+12.g1fdaa3e.dirty' from '/Users/warner/stuff/python/foolscap'. pkg_resources thought it should be found at '/Users/warner/stuff/python/foolscap'. The exception was PackagingError?: could not parse '0.8.0-12.g1fdaa3e.dirty' due to TypeError?: expected string or buffer <warner> I don't know where that TypeError? is coming from, or what it's getting that isn't a string <warner> I thought it might have to do with some Versioneer changes I made at one point which cause versions to sometimes be unicode, but hardcoding the version doesn't make the tahoe/packaging error go away

warner tries to even figure out what our 'setup.py build' does these days <warner> ah, found it. Tahoe's verlib.suggest_normalized_version() doesn't recognize Versioneer's (setuptools-compatible) "0.8.0+12.g123abc.dirty" as a parseable version, and returns None <warner> then verlib.NormalizedVersion?(that) explodes when the regexp tries to parse the None <warner> delete.. it.. all.. <warner> actually now it's reminding me of the HHGTTG scene where the intergalactic cruise is grounded for centuries whilst waiting the availability of lemon-soaked paper napkins <warner> the version can't be parsed, therefore refuse to run <warner> this version-parsing code is really getting in my way