Positive lookahead doesn't look beyond the current line?

[hediyi]

I want to make sure a character is followed by something that not necessarily on the current line. When that something is on next line, using (?=) doesn't work, as in

/(\/)(?=[^\/]*\/[ix]*)/

If /ix doesn't appear in the same line with the previous /, the regexp doesn't match.

If that's really the case, is there a way to make lookahead work across multiple lines?

[kkos]

Oniguruma has variable syntax/behavior function.
I don't know which syntax mode you are using.
If you are using default syntax(== Ruby), then [^\/] can match a newline.
ONIG_SYN_NOT_NEWLINE_IN_NEGATIVE_CC flag is disabled in default syntax.

if /(/)(?=[^\/]/[ix])/ =~ "/\n/ix"
puts "MATCH"
else
puts "NOT MATCH"
end

==> MATCH
ruby 1.9.3p484 (2013-11-22 revision 43786) [x86_64-linux]

You should go to ruby community if you ask about behavior of ruby regexp.

[hediyi]

Thank you very much for the reply! 😇

I'm trying to fix some bug of the Ruby language package of Atom, which I believe is using oniguruma as their regex engine. Actually, the bug is the package can't recognize a multi-line regex. The problem lies somewhere in

(?<![\\w)])((/))(?![?*+])
(?=
  (\\\\/|[^/])*
  /[eimnosux]*\\s*
)

which is basically a more verbose version of the previous example. It's really tricky to differentiate the division operator / and regex delimiter / without parsing the Ruby code. 😂

[hediyi]

I even tried ((/))(?=[^/]*/), it matches a single-line regex, but still not multi-line regex, so I suspected that [^/] in the lookahead doesn't match a LF.

Now that I think about it, maybe I've looked into the wrong spot.

[kkos]

I have never used Atom. I don't know about it.
But if atom/language-ruby uses oniguruma as regexp engine, why the package.json dosn't include dependencies to oniguruma module.
Is Atom running on JavaScript on Browser?

[hediyi]

Pretty much, it's written in CoffeeScript, to be exact, and running on Node.js. The UI is basically parsed HTML and CSS, I guess you can call it a desktop editor created with Web technologies 😄

why the package.json dosn't include dependencies to oniguruma module

It has, in the core: https://github.com/atom/atom/blob/9ea68024acccd7dc7494f50d03496c16b193c0c4/package.json#L46
I think oniguruma backs up all its language packages, and also the find-and-replace feature.

[kkos]

I have installed Atom 1.8.0 in Windows.
I tried to run your last regexp in it.
(Very difficult for me to run coffee-script in Atom. I hacked character-count package.)
And it match with multi-line string.
m[0] = '/'.

file: .atom/.apm/character-count/lib/character-count.coffee
module.exports =
count: ->
m = "/\n\n\n/".match(/((/))(?=[^\/]*/)/)
alert(m[0])

activate: (state) ->
atom.commands.add 'atom-workspace', 'character-count:count', => @count()

[hediyi]

👍 @kkos.

I also tried /[^/]*/ in the find-and-replace (to find slashes in ) with regex, it worked. But in the grammar file of a language package, no luck; even /\n*/ doesn't work, no idea why.

To minimize the possibilities why this doesn't work, I wrote a grammar file from scratch, this is the only rule in the grammar:

When the slashes are in the same line, it matches (see the scopes at cursor):

But if put newlines in between, it doesn't (see the scopes at cursor and the change of color):

It looks like a pattern in the grammar file doesn't search across lines—if it can't find a match in a line, it just starts fresh in the next line.

[hediyi]

Found the answer in TextMate's manual which Atom's grammars are based on:

Note that the regular expressions are matched against only a single line of the document at a time. That means it is not possible to use a pattern that matches multiple lines. The reason for this is technical: being able to restart the parser at an arbitrary line and having to re-parse only the minimal number of lines affected by an edit. In most situations, it is possible to use the begin/end model to overcome this limitation.

Thanks for your time, really really appreciate it 😉

[kkos]

I thank for your investigation.