Fix version parsing

[haarg]

This fixes the version number parsing to handle underscores properly.

version.pm is too buggy to use for this.

Fixes #337

[haarg]

Ping @dagolden, @ribasushi, @Leont, @karenetheridge, @rjbs in case you have any input on this.

[oalders]

Also ping @shadowcat-mst, @miyagawa and @neilbowers. Anything to add?

[dagolden]

I'm only following this issue lightly, but I don't think version.pm's handling of underscores is well-advised, and historically did particularly perverse things depending on presence/absence of version.pm and the version of perl (see "Version Number Puzzles" for examples). And that's only for modules, not for distributions, where version numbers are irrelevant to PAUSE except for the presence/absence of underscores.

Generally speaking, the use of underscores in version numbers is a historical accident and cause more problems than it solves. Treating them as invisible for all purposes except deciding whether or not to index a distribution should solve more problems than it creates.

The Lyon 2014 "version number summit" agreed on this as the desired direction for handling of version numbers: https://gist.github.com/dagolden/9559280.

[Leont]

I agree with @dagolden and @haarg, it's nothing less than crazy than a method called numify is not returning a valid number (worse yet, something that sort right when treated as one).

[karenetheridge]

I agree; let's stick to the decisions made at the version number summit (which needs a better name - what was the name of that restaurant?)

I'm also hoping that this will fix #284.

[miyagawa]

The current implementation of MetaCPAN's numification leads to issues like miyagawa/cpanminus#377 (comment) on cpanm. +1 on this patch.

[coveralls]

Coverage increased (+0.03%) when pulling cfd63fc on haarg/version-parse-fix into 8fc9ffa on master.

[coveralls]

Coverage increased (+0.03%) when pulling cfd63fc on haarg/version-parse-fix into 8fc9ffa on master.