preg: subject length, backref, quirks and speed

johanafm

Are there any known issues with preg, or more specifically preg_match_all?

I had a regular expression that would find all tags of certain formats. The problem is that it worked as it should sometimes and that it would fail at other times. Additionally, when it failed, the subject string was longer and by looking at a substring of no more than 600 characters it would work. Coincidence?

I resorted to not finding all various tagnames in a general manner using a back reference and instead went through the subject once per tagname I needed to find.

Patterns to find:
A) [tagname]inner text[/tagname]
😎 [tagname tagtext]inner text[/tagname]

Do note that the two following patterns differ on capturing subpatterns as I realized I needed the text between the tags captured for later use without the tags. Other than that, I had expected the two versions to do the same job.

Pattern (1):
General pattern that did not work as I expected:

preg_match_all(
	 "/(\[([^]]*)|(?:([^\x20]*)[^]]*)\])" . ".*?" . "\[\/\\2\]/suD",
	 $str, $matches);

Pattern (2):
Tagname specific pattern that works

	$findThis = array("topheadline", "headline", "tag", "ingress",
				   "byline", "articlelink" );
	for ($i = 0; $i < count($findThis); ++$i) {
		preg_match_all(
			"/\[" . $findThis[$i] . "(?:[ ][^]]*)?\](.*?)\[\/". $findThis[$i] . "\]/sDu",
			$str, $matches[$findThis[$i]]);
	}

Pattern (1) also never failed if I replaced $str with substr($str, 0, 600). Does this mean that there are limitations on subject string length under certain circumstances? Or is that just pure coincidence?

And finally, Pattern (2) performs much faster than Pattern(1) did. Any idea why?

Thankful for any input.
regards
johan

NogDog

I don't have the energy to analyze regexes right now, but if your goal is to implement a bbcode sort of thing, I'd strongly suggest looking at the [man]bbcode[/man] PECL extension or the PEAR HTML_BBCodeParser package, as (1) they use a more efficient algorithm than the regex approach, and (b) they are thoroughly tested.

johanafm

Thanks for the reply, but actually I'm not looking to implement a BB-code replacement, but I'll have a look at the bbcode extension as well since I suppose there ought to be a whole lot of similarities to the matter at hand. I've been asked to build XML files from articles in a database. Those articles have stuff like "[headline]The Headline[/headline]" and [articlelink id="someID" otherstuff="someotherstuff"] etc embedded in the article texts, appearing in random order and sometimes one embedded inside the other.
Once I've found and removed the photo tags, headline and some other stuff, I just need to strip the remaining tags from the text, whatever their content or placement.

As I'm uncertain of wether I've seen all the existing tagnames I would have preferred a general solution over using specific tagnames. But as I consider it unlikely that other tagnames actually exist, I mainly asked because I dislike not knowing why something fails.