Idea: Node Pattern sample-based generator

[pirj]

Again, not a feature request, more like to start the discussion. Can't think of a better place for this discussion again.

As a prolific cop author, I personally find Node Pattern hard to write.

Please speak up if you use some tools, that visually indicate that a node pattern you're trying to craft matches a given set of source excerpts. I can think of this as:

(send {nil? (const :Greeter)} :hello str)

§ hello 'Dear'
§ Greeter.hello ''
§ hello 'world'
§ hello 1
§ hello

Think visual regexp matcher we got used to already.

I recall there are Regexp generators that build a minimal Regexp to match a given set of strings (and always returns .* haha).
Wondering if building a similar tool to automatically build node pattern given some code samples it should match, and some that it should not.

Unfortunately, I'm not aware of the principles of building such a tool. Do you have an idea of where to start?

[marcandre]

Yes, I want to have an easy way to check a node pattern. I also was thinking of showing the compiled code (maybe side by side with the node pattern) and have it show in color for executed / not executed, to make it easier to understand non matches. I was even thinking of putting an executable in the gem (maybe rast, or else rubocop-ast)

[jonatas]

I have worked in something like this on fast: https://jonatas.github.io/fast/similarity_tutorial/

I made expression_from part of the gem and you can also use fast -s to search by similarity but it's not generating all the inner possibilities only anonymizing all identifiers from the code sample.

Maybe instead of generating a very generic expression we can generate several about the same code.

I also implemented one very smart shortcut to find "any reference" about some wording:

# Search all references about some word
Fast.shortcut(:ref) do
  Kernel.class_eval do
    def matches_args? identifier
      search = ARGV.last
      regex = Regexp.new(search, Regexp::IGNORECASE)
      case identifier
      when Symbol, String
        regex.match?(identifier) || identifier.to_s.include?(search)
      when Parser::AST::Node
        regex.match?(identifier.to_sexp)
      end
    end
  end
  pattern = <<~FAST
    {
      ({class def sym str} #matches_args? ...)
      ({const send} {nil? _} #matches_args? ...)
    }
  FAST
  Fast::Cli.run!([pattern, '.', '--parallel'])
end

[pirj]

That's already something to start with.
Would you like to hack on it @jonatas ? I guess writing specs for some pretty basic cases would really spark the fire.

[jonatas]

That's already something to start with.

@pirj you mean the example of the shortcut I shared "# Search all references about some word" or from the expression_from?

I'd love to contribute to the project, especially to introduce a basic CLI application to let more people know about the node pattern facilities :)

[pirj]

I meant RSpec examples of how a sample-based node pattern generator would work:

describe 'NP::G' do
  it 'generates' do
    expect(
      generate('hello.hello(1) { }', 'foo.foo(2) { }')
    ).to eq '(block (send (send nil? _) int) _ nil?)'
  end
end

or how a visual matcher will highlight matched and unmatched branches of the NP given a code sample:

describe 'NP::V' do
  it 'highlights' do
    expect(
      highlight_unmatched('(block (send (send nil? _) int) nil?)', 'hello.hello(1) { 2 }')
    ).to eq '(block (send (send nil? _) int) _ *nil?*)'
  end
end

what you like most. The latter is a slightly different feature, it just happened to have born in this ticket.

[jonatas]

I love the idea of highlight_unmatched especially when we need to merge node patterns and add extra edge cases.

[marcandre]

I love the idea of highlight_unmatched especially when we need to merge node patterns and add extra edge cases.

green = match, red = no match, yellow = not reached, light blue = not something that can be matched.

[pirj]

It looks amazing!

[jonatas]

Wow! that is very cool @marcandre! Looking ahead to test it locally :)

I was thinking about we have like a web interface like rubular.com that we could test the patterns against some code. I know that @baweaver started something on https://github.com/baweaver/ast_explorer maybe we could join the ideas to create a good playground for it :)

[marcandre]

@jonatas Opened #105 , check it out! Yes, I'd like something like rubular! Not sure about UI... TTY? Tk? web (can parser run in Opal?)?

[jonatas]

@marcandre that is a good point I was watching a live coding about opal right now on Ruby Kaigi with @youchan. Looks like it transpile to create the code. I think if opal compiles over opal it's possible :)´

The presentation was great and I learned a lot about how to create an AST rewriter in opal.

I think we can start with the server version and see if we can evolve to use Opal in the FE as a second version due to the challenge of transpile all the dependencies to the FE. I'm not sure what is the state of opal actually.

[marcandre]

I put an initial version of a live debugger at https://nodepattern.herokuapp.com/

[jonatas]

@marcandre is a coding machine 😮 👏 ❤️

I love to use it! very cool!

[baweaver]

Great job on the UI there. I'd always considered doing it through Lambda or just a Rails server because ruby > AST isn't technically executing code

[marcandre]

Great job on the UI there.

Thanks!

I'd always considered doing it through Lambda or just a Rails server because ruby > AST isn't technically executing code

Right, but my app does match a Ruby AST against a node pattern, and NodePatterns are very permissive... #eval("insert ruby code") is a valid pattern! I disabled that on the server though...

My plan is to have a local version that can also do local searches.

[baweaver]

If you ever want to collab on any of that feel free to shoot me a message later. Interesting note on eval there.

[baweaver]

Over time I'd love to evolve it into something like Regexr but that's a beast of a task:

https://regexr.com/

[marcandre]

Thanks for the sponsorship @baweaver 🎉

Over time I'd love to evolve it into something like Regexr but that's a beast of a task:

https://regexr.com/

Which features in particular do you think are particularly worthwhile?
I created an issue with what I'm thinking about.

[pirj]

The idea may be interesting, but probably not worth the effort. I'll close for now.