Forum: Ruby Oniguruma lookbehind question

Announcement (2017-05-07): www.ruby-forum.com is now read-only since I unfortunately do not have the time to support and maintain the forum any more. Please see rubyonrails.org/community and ruby-lang.org/en/community for other Rails- und Ruby-related community platforms.
unknown (Guest)
on 2006-01-07 04:58
(Received via mailing list)
Hi --

For some reason, lookbehind and alternation seem not to be playing
together in a little Oniguruma test.  This is based on the string
splitting thread from a little while ago this evening, and uses a CVS
1.9.0 Ruby acquired about 1/2 an hour ago.

   str =  %Q{abc def "ghi jkl" mno}

   # Look for "..." but just get the ... part:
   re1 = /(?<=")[^"]+(?=")/

   # Test that:
   p str.scan(re1)  # => ["ghi jkl"]

   # Now, do the same thing *or* \S+.  This should, I think,
   # pick up the abc, def, and mno substrings too.

   re2 = /((?<=")[^"]+(?="))|(\S+)/

   # But it doesn't; the part before the alternation never
   # matches, even though it did before (as shown by the
   # captures):

   p str.scan(re2)
   # => [[nil, "abc"], [nil, "def"], [nil, "\"ghi"], [nil, "jkl\""],
   #     [nil, "mno"]]

I know that's all a bit cluttered, but the basic thing is that a
sub-pattern using lookbehind doesn't seem to match any more when
there's an alternation.  Instead, only the second alternative ever
matches.

Does anyone know why?


David

--
David A. Black
removed_email_address@domain.invalid

"Ruby for Rails", from Manning Publications, coming April 2006!
http://www.manning.com/books/black
K.Kosako (Guest)
on 2006-01-07 05:28
(Received via mailing list)
removed_email_address@domain.invalid wrote:

>   # Test that:
>
>   p str.scan(re2)
>   # => [[nil, "abc"], [nil, "def"], [nil, "\"ghi"], [nil, "jkl\""],
>   #     [nil, "mno"]]
>
> I know that's all a bit cluttered, but the basic thing is that a
> sub-pattern using lookbehind doesn't seem to match any more when
> there's an alternation.  Instead, only the second alternative ever
> matches.

Is this pattern work for you?

str =  %Q{abc def "ghi jkl" mno}
re3 = /((?<=")[^"]+(?="))|([\S&&[^"]]+)/
p str.scan(re3)     #=> [[nil, "abc"], [nil, "def"], ["ghi jkl", nil],
[nil, "mno"]]
Ross B. (Guest)
on 2006-01-07 05:43
(Received via mailing list)
On Sat, 07 Jan 2006 02:56:50 -0000, <removed_email_address@domain.invalid> 
wrote:

>    re1 = /(?<=")[^"]+(?=")/
>    # matches, even though it did before (as shown by the
>
> Does anyone know why?

I'm not at all sure about this, but this is my take on it. Firstly, is
this the behaviour you expected?

	str =  %Q{abc def "ghi jkl" mno}
	re2 = /(?:(?<=")[^"]+(?="))|\S+/

	p str.scan(re2)
	# => ["abc", "def", "\"ghi", "jkl\"", "mno"]

?

If so, then I believe the problem is something to do with the fact that
lookaround is atomic, so that when used with capturing groups and
alternations you sometimes experience problems because the regex
immediately forgets the (zero-width, remember) lookaround match, so that
by the time it comes to that 'or' it doesn't have the information to
compare.

Generally, there are restrictions with lookaround (esp lookbehind)
matching, and especially when matching regexps. So far my experiments
with
Oniguruma suggest it's fairly sophisticated in this respect, supporting
stuff like varying-width alternations, fixed repetition and optional
groups in lookbehind, but of course still no star and plus.

Anyway, that's what I think. Hope it helps :)

Cheers,
Ross B. (Guest)
on 2006-01-07 05:49
(Received via mailing list)
On Sat, 07 Jan 2006 03:36:34 -0000, Ross B.
<removed_email_address@domain.invalid> wrote:

> Generally, there are restrictions with lookaround (esp lookbehind)
> matching, and especially when matching regexps.

Oops, I mean nested regexps.
unknown (Guest)
on 2006-01-07 05:52
(Received via mailing list)
Hi --

On Sat, 7 Jan 2006, K.Kosako wrote:

>>   re1 = /(?<=")[^"]+(?=")/
>>   # matches, even though it did before (as shown by the
>
> Is this pattern work for you?
>
> str =  %Q{abc def "ghi jkl" mno}
> re3 = /((?<=")[^"]+(?="))|([\S&&[^"]]+)/
> p str.scan(re3)     #=> [[nil, "abc"], [nil, "def"], ["ghi jkl", nil], [nil,
> "mno"]]

No; I get this:

[[nil, "\""], ["ghi jkl", nil], [nil, "\""]]

This:

   /(?<=")[^"]+(?=")|[^\s"]+/

gives me the result you got for your re3.  But it still seems to be
based on the right-hand alternate being checked first.


David

--
David A. Black
removed_email_address@domain.invalid

"Ruby for Rails", from Manning Publications, coming April 2006!
http://www.manning.com/books/black
Xavier N. (Guest)
on 2006-01-07 05:52
(Received via mailing list)
On Jan 7, 2006, at 3:56, removed_email_address@domain.invalid wrote:

>   # Now, do the same thing *or* \S+.  This should, I think,
>   # pick up the abc, def, and mno substrings too.
>
>   re2 = /((?<=")[^"]+(?="))|(\S+)/
>
>   # But it doesn't; the part before the alternation never
>   # matches,

It shouldn't, since pattern-matching goes left-to-right \S will match
the quote before the first half of the regexp gets a chance, since it
wants to match the first character _after_ the quote.

-- fxn
Ross B. (Guest)
on 2006-01-07 06:10
(Received via mailing list)
On Sat, 07 Jan 2006 03:36:34 -0000, Ross B.
<removed_email_address@domain.invalid> wrote:

> If so, then I believe the problem is something to do with the fact that
> lookaround is atomic

Argh, I meant 'if not'. I'm too tired now, I've spent too long
perfecting
this quiz thing...
I was guessing it was dropping the match without the capture and so
always
matching the alternate but that's not right.

Sorry.
unknown (Guest)
on 2006-01-07 06:10
(Received via mailing list)
Hi --

On Sat, 7 Jan 2006, Xavier N. wrote:

> It shouldn't, since pattern-matching goes left-to-right \S will match the
> quote before the first half of the regexp gets a chance, since it wants to
> match the first character _after_ the quote.

OK, I see.  I was somehow discounting the fact that the first "
*itself* doesn't match the left-hand alternate.

Thanks --


David

--
David A. Black
removed_email_address@domain.invalid

"Ruby for Rails", from Manning Publications, coming April 2006!
http://www.manning.com/books/black
unknown (Guest)
on 2006-01-07 06:13
(Received via mailing list)
Hi --

On Sat, 7 Jan 2006, Ross B. wrote:

> On Sat, 07 Jan 2006 03:36:34 -0000, Ross B.
> <removed_email_address@domain.invalid> wrote:
>
>> If so, then I believe the problem is something to do with the fact that
>> lookaround is atomic
>
> Argh, I meant 'if not'. I'm too tired now, I've spent too long perfecting
> this quiz thing...
> I was guessing it was dropping the match without the capture and so always
> matching the alternate but that's not right.

See Xavier's post.  My mistake was, essentially, expecting the first "
to "know" that it was supposed to match a zero-width condition
governing the state one character later.  Instead, of course, it
asserts itself as a character in its own right; fails to match the
first alternate; and does match the second.

So [^\s"]+ is indeed probably the best thing.  (Other than the
appropriate real libraries, of course :-)


David

--
David A. Black
removed_email_address@domain.invalid

"Ruby for Rails", from Manning Publications, coming April 2006!
http://www.manning.com/books/black
unknown (Guest)
on 2006-01-07 06:20
(Received via mailing list)
Hi --

On Sat, 7 Jan 2006, K.Kosako wrote:

> Is this pattern work for you?
>
> str =  %Q{abc def "ghi jkl" mno}
> re3 = /((?<=")[^"]+(?="))|([\S&&[^"]]+)/
> p str.scan(re3)     #=> [[nil, "abc"], [nil, "def"], ["ghi jkl", nil], [nil,
> "mno"]]

Sorry -- I tested that accidentally with an old 1.9.0.  Yes, with
today's I do get the same as you.

(And I also understand why it's choosing the right-hand alternate :-)
See later posts in thread.)


David

--
David A. Black
removed_email_address@domain.invalid

"Ruby for Rails", from Manning Publications, coming April 2006!
http://www.manning.com/books/black
Ross B. (Guest)
on 2006-01-07 06:29
(Received via mailing list)
On Sat, 07 Jan 2006 04:12:53 -0000, <removed_email_address@domain.invalid> 
wrote:

>> Argh, I meant 'if not'. I'm too tired now, I've spent too long
>> perfecting this quiz thing...
>> I was guessing it was dropping the match without the capture and so
>> always matching the alternate but that's not right.
>
> See Xavier's post.  My mistake was, essentially, expecting the first "
> to "know" that it was supposed to match a zero-width condition
> governing the state one character later.  Instead, of course, it
> asserts itself as a character in its own right; fails to match the
> first alternate; and does match the second.
>

Oh Damn it, yeah I see now. Wish I'd held my tongue now :D
This topic is locked and can not be replied to.