Improve ERB dependency detection

[britto]

Following the discussion started on #13116, I thought it might be worth giving a shot at some of the issues we have identified.

The current implementation can't handle some special cases of oddly-formatted Ruby. Now we are able to detect them:

• Multi-line arguments on the render call
• Strings containing quotes, e.g. "something's wrong"
• Multiple kinds of identifiers - instance variables, class variables and globals
• Method chains as arguments for the render call
• "Glued" identifiers, e.g. render@things, render:partial => 'partial', etc.

Also, this fix reduces the rate of "false positives" which showed up when we had calls/access to identifiers containing render, like surrender and rendering.

Related: #13074.

Update:

There is still one edge case for false positives which will be virtually impossible to avoid whithout actually parsing the template: when we have render inside a string, as in:

<%= content_tag :p, 'Then we render something nice.' %>
<%# "somethings/something" %>

For obvious reasons 💣. But I guess that will not be such an issue. What we have got so far is probably about as far as we can go with a human-readable regex.

[rafaelfranca]

🤘

[josevalim]

Oops, we got some failures in the build:

  1) Failure:
TemplateDigestorTest#test_extra_whitespace_in_render_named_partial [/home/travis/build/rails/rails/actionview/test/template/digestor_test.rb:174]:
digest didn't change
  2) Failure:
TemplateDigestorTest#test_extra_whitespace_in_render_record [/home/travis/build/rails/rails/actionview/test/template/digestor_test.rb:180]:
digest didn't change
  3) Failure:
TemplateDigestorTest#test_extra_whitespace_in_render_with_parenthesis [/home/travis/build/rails/rails/actionview/test/template/digestor_test.rb:186]:
digest didn't change

[britto]

@josevalim thanks! unfortunately, I can't take a look into it right now. It seems to be some problem with multiple mathces. I'll get back to it ASAP.

[britto]

Turns out String#scan gets greedy when we enable multiline-mode. I think a split & scan approach may be more adequate to tackle that.

[britto]

Any clue on that Memcached issue? The suite passes locally.

[carlosantoniodasilva]

@britto I've restarted the failing job, lets wait and see.

[britto]

@carlosantoniodasilva great! 🍏

[arthurnn]

❤️