Regex bug with lookahead AND lookbehind

[jeffy]

I'm pretty sure I found a regex related bug here. Could someone please confirm? I'm using 7.0.9.

Code: Select all

text	text	text

Put the cursor somewhere in the middle word and search for '$' down (regex, without the quotes). The cursor goes to the end of the line. Then search for '(?<=\S)\t(?=\S)' down. It wraps around and finds the first tab (a tab between two non-whitespace characters).

Now do it again, but this time search for '(?<=\S)\t(?=\S)' up. It doesn't work (says not found).

However, if you put the cursor at the start of the line (actually, I think anywhere before the tab itself) and then search for the same thing, it works.

Thanks for checking!

[ben_josephs]

You're right. In fact it appears that all look-behind expressions fail when searching backwards. I suspect that fixing this would be a non-trivial job.

Negative look-behind ( (?<!...) ) appears not to work at all. It always matches, so the regex as a whole behaves as if the look-behind assertion wasn't there.

[jeffy]

I'm glad I'm not the only one. Thanks.

I wonder why it's a non-trivial change, although I expect the explanation is non-trivial

[jeffy]

Another example of this problem I just encountered:

Code: Select all

(?<=[ \t])\bIIMeta\b\s*\(\s*\b(\w+)(|(?:<[?\w ]+>)|(?:<[^<]*<[?\w ]+>[^>]*>)|(?:<[^<]*<[^<]*<[?\w ]+>[^>]*>[^>]*>))\s+(\w+)\b(?!>)\s*\)

This finds a Java function signature with exactly 1 parameter (including up to three levels of generics after the type). It should find either of these:

Code: Select all

	public IIMeta(IIMeta ii_toCopy)  {
	public IIMeta(String s_instanceName)  {

specifically selecting only "IIMeta(...)", but it doesn't work.

However, removing the back-reference (despite selecting the initial whitespace character, which is what I'm trying to avoid) does work:

Code: Select all

[ \t]\bIIMeta\b\s*\(\s*\b(\w+)(|(?:<[?\w ]+>)|(?:<[^<]*<[?\w ]+>[^>]*>)|(?:<[^<]*<[^<]*<[?\w ]+>[^>]*>[^>]*>))\s+(\w+)\b(?!>)\s*\)

Also, the backreference DOES work if you first click the very top of the document, and then search down, but it only finds the first instance, and then gets stuck again.

[ben_josephs]

If you prefix your regex with the modifier (?x) you can add white space without changing the meaning, making the regex (perhaps) somewhat easier to read:

Code: Select all

(?x) (?<=[ \t]) \b IIMeta \b \s* \( \s* \b (\w+) ( | (?:<[?\w ]+>) | (?:<[^<]*<[?\w ]+>[^>]*>) | (?:<[^<]*<[^<]*<[?\w ]+>[^>]*>[^>]*>) ) \s+ (\w+) \b (?!>) \s* \)

For the sake of simplicity, remove the Java template stuff, which isn't required in your example:

Code: Select all

(?x) (?<=[ \t]) \b IIMeta \b \s* \( \s* \b (\w+) \s+ (\w+) \b (?!>) \s* \)

It is now apparent that the word boundary anchors (\b) are redundant, and can be removed:

Code: Select all

(?x) (?<=[ \t]) IIMeta \s* \( \s* (\w+) \s+ (\w+) (?!>) \s* \)

As can the (?!>) look-ahead:

Code: Select all

(?x) (?<=[ \t]) IIMeta \s* \( \s* (\w+) \s+ (\w+) \s* \)

And some parentheses:

Code: Select all

(?x) (?<=[ \t]) IIMeta \s* \( \s* \w+ \s+ \w+ \s* \)

It is now clear that your regex matches the function name and its single parenthesised typed parameter.

I don't see anything wrong.

(By the way, the expression (?<=[ \t]) is a look-behind assertion, not a back-reference.)

[bbadmin]

The Boost regular expression engine used by TextPad does not support backwards searches. (It's author says it would require a whole new state machine implementation and the dropping of lots of features.) The workaround that TextPad implements is to iterate backwards, a character at a time, and try to match the search pattern forwards from there. This works in most cases, but not with look behinds.