LINQ-like regular expressions

[DartBot]

This issue was originally filed by [email protected]

---

Regular expressions are cryptic. Everybody has problems trying to remember how to construct a regular expression and a day later they forget what the strange thing means. There is no argument about this. The best approach to solve regular expressions that I can think of is:

var regExp = startsWith 3 letters
             andThen 6 numbers or "blah"
             andThen atLeast 2 "-";

Easily readable.

[floitschG]

Putting into 'Area-Language' for now (due to the LINQ reference). However the particular RegExp example is probably more library related than language related.

---

Removed Type-Defect label.
Added Type-Enhancement, Area-Language, Triaged labels.

[DartBot]

This comment was originally written by [email protected]

---

Everybody has problems trying to remember how to construct a regular expression and a day later they forget what the strange thing means. There is no argument about this.

I have to disagree. Common regexp syntax is indeed very terse, but that is actually a good thing. Regexp is not cryptic at all, it is in fact precisely the string that you want to find (!!) with some special syntax for expressing that more variants are possible on certain places. A regular expression for matching string "abc" is exactly this, "abc". A regular expression for matching one digit is "[0-9]", for one or more digits "[0-9]+" -- and I really wouldn't like to write "atLeast 1 number" instead. (And what exactly does "number" mean here? An integer or decimal number? A digit only? Every Unicode codepoint that denotes a digit in some language or 0-9 only?)

A lot of people have troubles constructing and understanding regexps, sure, but a lot of people do not. And from my experience, people get used to the regexp syntax pretty quickly. Some alternative means of constructing regexps might be good, but maybe allowing comments in regexps like Perl allows would be good enough (but that would allow regexp literals, I think).

[DartBot]

This comment was originally written by [email protected]

---

"but a lot of people do not"

http://stackoverflow.com/questions/tagged/regex

It is in fact one of the top tags on stackoverflow.

"And from my experience, people get used to the regexp syntax pretty quickly"

Yes, but that only applies if you are using regex every single day, otherwise you will forget the vast array of special codes it uses.

[DartBot]

This comment was originally written by [email protected]

---

This should be a library (I imagine it using chaining), not a language construct. I highly disagree your example is "easily readable". You're introducing new operators and precedence rules; your formatting suggests you read it one way, but I read it completely differently.

var regExp = (startsWith 3 letters
             andThen 6 numbers) or ("blah"
             andThen atLeast 2 "-");

[DartBot]

This comment was originally written by [email protected]

---

These are minor complaints that are easy to fix. I meant "digits" not numbers, which clears up the first problem mentioned. As for precendence, some kind of convention where it is understood that each line starts a new section would work. The operators are English words that are already understood and easier to remember. Regex is an entire alphabet of strange codes that nobody understands, so you can hardly say Regex isn't worse with respect to this issue.

Your high disagreement does not correspond to the issues you then bought up, which are minor issues about precendence. I am pretty sure you understood the operators even if they are "new".

Assuming precedence was cleared up, and the digits issue was cleared up, any programmer could determine if a string met the criteria in the expression. If you convert that to regex and ask people who don't use regex every day, they will not be able to tell you what it means.

[gbracha]

Set owner to @gbracha.
Added Accepted label.

[DartBot]

This comment was originally written by [email protected]

---

I have now created a specification of how the language could look if anyone is interested.

All of the character classes and so on in traditional regular expressions can obviously just be represented as sets. There may be an object containing all predefined expressions such as:

class Pre {
  Set<String> lowerLetters;
  Set<String> upperLetters;
  Set<String> digits;
  Set<String> whitespace;
}

or something like that. Here are some examples.

var regex = start "$"
            next min 1 digits
        next “.”
            next 2 digits
            end;

Quantifiers can be expressed with the following syntax:

Given the example "googledartgoogledartgoogledart", we could have:

Greedy quantifier that matches the entire string:

var regex = start min 1 letters
            until last "dart" inclusive;

Reluctant quantifier that matches the first "googledart":

var regex = start min 1 letters
            until first "dart" inclusive;

Posessive quantifier:

var googleLetters = Set.from(["g", "o", "l", "e"]);
var regex = start min 1 max 6 googleLetters
            until not googleLetters;

There are examples of lookaheads even though they say before. Dart and javascript do not have lookbehinds.

var regex = "google" before "dart";

var regex = "google" before not "dart";

Some other things:

var regex = 1 not digits;

var regex = start 1 (digits - ["0", "1", "2", "3"])
            end 3 letters;

var regex = 3 letters
            next 2 digits next 3 "-"
            or
            next 10 digits next 1 "-"
            next (3 upperLetters) as discard;

"or" applies to one line above and below if on a line by itself.
    
Anyway, that was a fun exercise, back to more productive things :)

[DartBot]

This comment was originally written by [email protected]

---

Actually that or idea is terrible :) Also, could alternatively use

var regex = start min 3 letters
            until first "dart"
            next "dart"
            next "google"
            next 1 not letters;

to remove need for "inclusive" as a keyword.

[anders-sandholm]

May be suitable for a library - not a core language feature.

---

Added WontFix label.