enh(elixir) Improves and fixes many issues with Elixir grammar

[angelikatyborska]

Closes #3210
Closes #3211

Changes

• If defmacro is a function definition, defmacrop should be too
• If defrecord is a class definition, defstruct should be too

Checklist

• Remove incorrect keywords
• Add missing keywords
• Define literals (nil, true, false)
• Fix highlighting 0 as a number
• Added markup tests
• Updated the changelog at CHANGES.md

[joshgoebel]

class is deprecated now. More likely you'd want this to be a newer multi-match rule and then use title.class for the name of the struct... See wren.js for examples. I can also do this myself after though... I already had a few commits patching the other declarations to the newer standard but decided they didn't fit in our regex fix PR.

[angelikatyborska]

Hmm... since the struct's name is not part of the defstruct call but it's defined by the module name in which defstruct is called, I think I just need to make defstruct a keyword instead?

[joshgoebel]

iex> defmodule User do
...>   defstruct name: "John", age: 27
...> end

Ah, yes, just a keyword then for defstruct... defmodule though would probably be a multi-matcher.

[joshgoebel]

I see how these are better examples (that you added), but if the previous were valid we should also leave them to serve as additional regression tests... one can still use a multi-line string for a single line - that is valid grammar, yes?

[angelikatyborska]

that is valid grammar, yes?

No, those were syntax errors in Elixir. The only allowed characters after the opening '''/""" is optional whitespace, and then a newline is required.

My assumption is that it's not a big deal if the syntax highlighting colors something that doesn't compile so I didn't feel the need to change the regexes for multiline strings, but the test examples should be valid Elixir code, right?

[joshgoebel]

No, those were syntax errors in Elixir.

Whoa. I stand corrected. ;-)

My assumption is that it's not a big deal if the syntax highlighting colors something that doesn't compile

Right, it's not our job to detect invalid code.

but the test examples should be valid Elixir code, right?

For the most part, yes. In this particular case this indeed makes sense. I just see people "swap" test cases often when they should be adding new cases instead, [not removing old ones]. Thought this was perhaps yet another case of that. :)

[joshgoebel]

Perhaps we could add aliases for Elixir extensions also.

[angelikatyborska]

I'm assuming the id should be the PR id, not the issue id?

[joshgoebel]

Yep

[joshgoebel]

@angelikatyborska Thanks so much!