Short circuit eval regexes in finding module definition

[dduugg]

I was profiling gem RBI generation for Tapioca and found that it was spending half its time in Pry::WrappedModule::Candidate#class_regexes. I noticed that method evaluates three regexes on every invocation since this change, and have reverted it to the prior short-circuit implementation. I've verified that tests pass locally and that the output of Tapioca RBI generation is unchanged (while running 8% faster). Tapioca has since removed the dependency on this Pry code, but I thought you might find this patch useful regardless.

Note that i think there could be a few more possible optimizations here:

• The start-of-string anchor (\A) should probably be preferred over the start-of-line anchor (^) in each regex
• .match? should be preferred over =~ once the minimum supported ruby is 2.4
• I think that the various invocations of #name are susceptible to the this performance bug, and should be cached in an ivar. I've reduced the invocations by two, but not taken on a larger refactor of WrappedModule.

[kyrylo]

Great insights! I appreciate the work that was done and the tips on possible optimizations.