This will allow to use reporters from matchers avoiding stupid direct
writes to $stderr.
Also it will allow to remove matching from CLI altogether in a phase of
the runner. Allowing to decouple Mutant::Config from VM environment
allowing to serialize it ;)
Step by step. Takes a while.
* I do not use 1.9.3
* Also keeping them in each file increases mental overhead (true it *can* be autoamted)
* None of the files encodes NON ASCII chars.
* I do not expect it makes any difference, since nobody programmatically
will consume strings generated by mutant under the assumption they are UTF-8 encoded.
* 1.9.3 Users have to deal with the encoding fuckup under ruby anyways.
* Classifiers are matcher subclass and delegate to lazy builded matcher
interfaces
* Solves chicken egg problem between target library load and classifier
matcher instantiation
* Currently broken (sorry) but do have to push to fix some nasty
transitive dependency stuff (do not ask).
* Fix fix it tomorrow (hopefully I find the time)
* Mutator and Generator where merged.
* A single pass over all duplications was made.
* It is clear a specific handles?(node) code for finding mutators
is needed. Like virtus does for attributes, should also cache.
* Does not pass on 1.9 mode currently as blocks are unexpectly parsed
differend when it comes to a series of literal booleans.
* Adjust metrics
* Add initial integration spec on method matching
* Yard and Heckle coverage is at 100% (heckle cov is disputable)
* Rcov does not really make sense as MRI 1.8 cannot reach all code
paths.
