The .NET regular expression engine's capturing behavior is not the same as the ECMAScript standard.

[otac0n]

For regular expressions such as this: ((a+)?(b+)?c+)*

There are 3 capturing groups (one for each left-parenthesis).

If this is matched against a string like the following: bbbccaac

The .NET implementation will list the following capture groups: ((a+)?(b+)?c) = "aac" (a+) = "aa" (b+) = "bbb"

Whereas the ECMAScript spec specifies the following capturing behavior: ((a+)?(b+)?c) = "aac" (a+) = "aa" (b+) = undefined

The .NET implementation gives no indication that the (b+) capturing group did not participate in its most recent match attempt.

[hakanson]

Does using RegexOptions.ECMAScript help?

http://msdn.microsoft.com/en-us/library/system.text.regularexpressions.regexoptions(v=VS.100).aspx

[otac0n]

@hakanson: We are already using the ECMAScript option, which works well for the most part. It is just this little piece that is different.

[fholm]

I think this is something we'll have to live with for now, doing a custom regular expression implementation for this small detail is too much for too little gain currently. I'll leave the ticket open, and we'll look into it eventually.

[hakanson]

-1 for me for not looking in the code in Core.fs

    let options = (options ||| RegexOptions.ECMAScript) &&& ~~~RegexOptions.Compiled
    let key = (options, pattern)
    this.RegExp <- env.RegExpCache.Lookup key (fun () -> new Regex(pattern, options ||| RegexOptions.Compiled))

I'm new to F#; does this mean you are implementing your own compiled RegExp cache? I ask because there is a Regex.CacheSize Property that controls an internal cache of compiled regular expressions. I assume it gave you more control to have your own cache, but thought I would add for completeness (as the risk of looking uninformed a second time on the same issue).

http://msdn.microsoft.com/en-us/library/system.text.regularexpressions.regex.cachesize.aspx

[fholm]

Yes we do maintain our own regexp cache, we found it to be faster actually.

[otac0n]

We found that in a loop like this...

while (true)
{
    var r = new RegExp("...");
}

...that .NET's regex cache was not helping.

When we implemented the regexp cache shown above, we saw a 50% reduction in the time on the SunSpider regexp test.

[ChaosPandion]

@otac0n - From the looks of it the BCL only caches for static methods on the Regex object so the increase in performance makes sense.