Expand string interpolation expressions into dedicated AST

[flying-sheep]

Hi, as seen in #3831, string interpolation like "A #{a} B #{b}" is lexed as if it was 'A '+a+' B '+b.

as mentioned in juliankrispel/decaf#2, there are use cases for having the semantic information survive lexing.

Therefore it would be awesome if we could introduce a new StringInterpolation AST node which is later converted to a concatenation expression.

[michaelficarra]

:+1:. This is how I did it in the CoffeeScriptRedux compiler.

[jashkenas]

IIRC, that would also help out with some other tricky issues. Like the old implicit call rewriter kludge.

[lydell]

@jashkenas I think that particular issue was fixed in commit 76c076db555c9ac7c325c3b285cd74644a9bf0d2 ;) But I do agree that information is lost in many places in the parser.

string interpolation like "A #{a} B #{b}" is lexed as if it was 'A '+a+' B '+b

Just a FYI: this is not true. But it is true that the AST looks the same. grammar.coffee:150.

[juliankrispel]

awesomesauce! This would be super helpful

[lydell]

@juliankrispel I'm working on this (and have come very far, all tests are currently passing): https://github.com/lydell/coffee-script/tree/node-types

[juliankrispel]

@lydell omg this is amazing. Tbh I'm not sure what I'm looking at but it's amazing nonetheless :D

[flying-sheep]

i hope you’re looking at lydell/coffee-script@5e54fc478f95a8a133c7d88f20600b7caf2fdee0, which is mostly new, more specific Node classes.

one of which being StringWithInterpolations, i.e. what we need.

changes like this tell me that doing the other stuff also is a very good idea!

[vendethiel]

It should also fix some other long-standing bugs like this