Add r method to StringContext (for regexes)

[scabug]

Scala currently has a method StringLike#r for compiling strings to regexes.

Now that Scala has string interpolation and macros I think we can do even better with a r method on StringContext. Here is why:

1. Escaping gets simpler (before: """\d+""".r, after: r"\d+")

2. Parameterization gets simpler (before: ("foo"+Pattern.escape(bar)).r, after: r"foo$bar")

3. Regexes can be syntax-checked at compile-time

4. Pattern matching against regexes gets simpler (before: val re = """\d+(.*)\d+""".r; … case re\(x\) => x, after: case r"\d+$x\d+" => x)

I have a proof of concept implementation available at https://github.com/qerub/scala-regex-stringcontext and can make a patch for Scala proper if people like this idea.

[scabug]

Imported From: https://issues.scala-lang.org/browse/SI-7496?orig=1
Reporter: @qerub

[scabug]

@qerub said:
This issue obviously hasn't blocker priority, but the field is stuck in my browser.

[scabug]

@soc said (edited on May 18, 2013 2:27:49 PM UTC):
Fixed the fields.

I think there are already a few implementations of that, including one in which you can name the capture group inline.

Would be interesting to check and compare what those implementations do and figure out what's a nice subset to include.

[scabug]

@qerub said:
Thanks!

I think I found an implementation of what you're mentioning: http://dcsobral.blogspot.com/2012/01/string-interpolation-on-scala-210.html

Yes, I agree. Let's proceed by collecting all relevant ideas.

[scabug]

@qerub said:
I've updated my PoC implementation with support for direct pattern matching and added a fourth item to the issue description.

Code: qerub/scala-regex-stringcontext@4228dcb#L1R41

[scabug]

@qerub said (edited on May 18, 2013 10:27:41 PM UTC):
Open questions for #4: Should the interpolation points match {{.*}}, {{.+}}, {{.*?}} or {{.+?}}? What should happen if the regex include explicit capture groups?

[scabug]

@dcsobral said:
There's an alternative version of interpolation matching where the interpolated text will always match the existing capture groups. By convention, people write things like (\d+)$x (\d+)$y, but that works just like (\d+) (\d+)$x$y. Anyway, such code was recently used by Adriaan Moors on a presentation (iirc), and can also be seen in this blog.

Personally, I think that style is error prone, since the order in which things get assigned is not related to their position, but to the number of capturing groups. Add a capture group by mistake, and the match will start failing.

Of these two, I do prefer my own version (as described in my blog post already mentioned), and I do like .* over any of the alternatives, but that does decrease the usefulness of matching quite a bit, since one often wants whatever is being capture to follow a pattern of its own.

What I'd really rather do is go with macros and use named patterns. Java 7 supports (?<NAME>X) to specify named capture groups. A Scala macro could check the string for named capture groups and assign them (with or without Java 7 support), and translate a string with them into one with normal interpolation, such as the one in the blog I mention. Since the macro will check the regex for named groups, it can ensure that the names are properly assigned, and give compile time errors when matching if there are unnamed capture groups.

[scabug]

@qerub said:
Thanks for your great input, Daniel!

I like your idea with named capture groups but I don't know how to implement it; it requires a code transform that's non-local to unapplySeq. Any hints?

[scabug]

@adriaanm said:
Daniel, I agree named groups would be an improvement. In my presentation I was just trying to explain string interpolation on a slide. In any case, I hadn't considered the problem you point out -- good point!

[scabug]

@xeno-by said:
We could always go for untyped macros ;) #5903

[scabug]

@jroper said:
I don't know if this has been considered, but I think it would be very useful if the regex expression was not compiled on each invocation of unapplySeq (which seems to be the case for all the proposed solutions so far), rather, it should be stored statically and referenced from the generated unapplySeq method. I'm not sure if this is possible with macros, but compiling regexes causes significant performance issues when their compilation can otherwise be cached - for the use cases that I want to use regex string interpolated extractors (for a router in Play Framework), if the regex had to be compiled in each invocation, that would be a performance show stopper.

[scabug]

@SethTisue said:
a SLIP on this would (I think) be welcome

[scabug]

@soc said:
I think the issue is that this needs compiler support for hoisting regex expressions (and compilation) to ~static places. I absolutely want to see something like this, but I don't have the bandwidth (and only minor desire to go through the SLIP process).