Hacker News Reader

Comment by Graphon1 | Hacker News Reader

What could anyone possibly do to improve Regular Expressions ?

Perl6 made a lot of improvements https://docs.perl6.org/language/regexes

Technically not regular expressions since they are appropriate for matching much more than regular grammars, but that's true for almost any programming language's "regular expressions".

> Perl6 made a lot of improvements https://docs.perl6.org/language/regexes

IMO the word "improvements" should be in quotes. Pathological regular expressions are an issue because of things like back references and look-behind matching. If Perl hadn't enshrined the idea of "a regular expression for every problem" people would probably be less skeptical of regular expressions in code.

So why not use grammars, if regexps are too complex...? https://docs.perl6.org/language/grammars

(The extensions to regexps in Perl 5 (and presumably now also in PCRE?) have made it possible to write them more like grammars than anything else. But you would probably not approve. :-) )

Lots of things.

Make syntax consistent across RE engines for one.

Improve the syntax.

Introduce clear modes with different tradeoffs, so you don't get e.g. regex based DOS attacks.

Make them easier to compose, not just through copy pasting...

I've steered more to composing grammars by using tools like Regexp::Grammars - https://metacpan.org/pod/Regexp::Grammars

The real sweet spot I've found is parse in Rebol / Red - http://blog.hostilefork.com/why-rebol-red-parse-cool/ | http://www.rebol.com/docs/core23/rebolcore-15.html

https://github.com/GuillaumeBadi/Verbal-Exprejon

A linq interface to visit the resulting parse tree.

Add types of course.

How do these improve regular expressions?

It doesn't improve regexp; it improves your code that uses regexp to populate type-checked data. Instead of writing a lot of "if foo is int then ..." type checking code you can just annotate the regexp with types and it either extracts those types for you or throws an error or something.

In general you have the same problem when parsing CSV, command line flags, URL query parameters, JSON, etc.

I see! Thanks.

Those type-safety problems for parsing tainted data are a royal pain and, often, security issues! Typed regex are a very good idea and very welcome. At the moment, when I use regex and extraction, I have a lot of boilerplate and I could easily have lots of bugs unfuzzed.

(For those having the type-safety problem with JSON in Python, can I recommend my own library obiwan? https://pypi.python.org/pypi/obiwan/ :) )

At this point, why not just use the more powerful, maintainable, and flexible parser combinators[0]? I thought the point of regex was for quick and dirty bash/grep/perl one-liners and search/replace in vi/emacs. If you're going to actually maintain the code, then use something designed for that.

[0] https://en.wikipedia.org/wiki/Parser_combinator

Well, there is the speed argument. Regular expressions (not the pcre kind with backtracking and stuff) are blazing fast. Ragel[1] is a good example of that. In general, though, I actually agree.

[1]: http://www.colm.net/open-source/ragel/

Among parser combinator libraries, attoparsec[0] is no slouch. It gets very close to the speed of the hand-rolled http-parser library while being much shorter, easier to read, and maintain[1].

[0] https://hackage.haskell.org/package/attoparsec

[1] http://www.serpentine.com/blog/2010/03/03/whats-in-a-parser-...

Another good example is RE2: https://github.com/google/re2

It's pretty fast too. I don't know whether you can compare the performance of Ragel to regex libraries like RE2; they look like they solve different problems.

As I mentioned in another comment, there are applicative parser combinator libraries for regular languages, e.g. https://hackage.haskell.org/package/regex-applicative so asymptotic speed is orthogonal to 'combinatorialness'.

If you look at the API of ocaml-re (and tyre), you'll see it's pretty much equivalent to regex-applicative. Regex combinators are nice.

"Parser combinators", for me, denotes actual parsers, for non-contextual or contextual grammars.

Consider the regex may be program input, so you can't just code them away. A search engine is a good example.

Indeed, that is the exact goal.

For the little history, I started this for URLs[1]. I couldn't (and still can't) figure out a proper interface for that, but someone mentioned he would like something like that for regular expressions, so here it is.

We already have something similar for command line flags in OCaml[2] (and several for JSON).

[1]: https://github.com/Drup/furl [2]: http://erratique.ch/software/cmdliner

Reply on news.ycombinator.com