Random question, how is a regression like a 'no method error' possible? Don't the creators of these languages write extremely exhaustive tests? I don't know much about this, just wondering...
Looks to be a corner case not covered by existing tests prior to now.
“This release contains a bug fix about Refinements and Module#prepend. The mixture use of Module#refine and Module#prepend to the same Class could cause unexpected NoMethodError. This is a regression on Ruby 2.3.2 released last week. See [Bug #12920] for details.”
Using formal methods? Yes but as an industry we simply don't know how to scale that to something as large and unwieldy as the Ruby VM.
And even if we did, correct according to what? There is no mathematical model of the Ruby language to prove it equivalent to (there are a few semantic models for Ruby but they are very limited).
And even if we had that, the computational resources needed likely are not available.
There is rubyspec, but it doesn't cover all possible paths, just the ones that the developers deemed important enough to test.
What I am curious to know is _what_ does ruby use besides rubyspec? Do they run tests against the top 100, 1000, etc rubygems before releasing? I know other communities like Rust (and even Perl) do this kind of regression testing. While this won't catch all possible regressions it seems like a reasonable starting point to uncover issues before a release.
There has been quite a controversy about rubyspec afaik: the tldr is that the de-facto ruby spec is basically 'whatever piece of software capable of running rails'. This is because the ruby devs change stuff around very often and don't keep an official spec of the language, to the extent that the creator of rubyspec had given up.
To be honest the issues with rubyspec in the past were sociological more than anything else. It was a right pain making contributions, and certainly not a welcoming community.
Specs have been part of ruby core for quite a while now, thanks to the efforts of other members of the community.
> Do they run tests against the top 100, 1000, etc rubygems before releasing?
I don't believe they do. They only recently started testing against ruby spec.
I work on an alternative implementation, and we are planning to test against literally all the gems in RubyGems ourselves. We test the top 100 or so at the moment.
If I might make a suggestion, I'd recommend doing some kind of tracing when you run the tests so you can have a mapping of what parts of your code are exercised by which gems/tests. Then when you make changes to your code you can run the tests that relate to that part of the system first, thus giving you more confidence that your change didn't cause a regression. This should be much more efficient than simply doing a full gem run (although I'd recommend that too, since it could uncover hidden dependencies).