Bug #13671: Regexp with lookbehind and case-insensitivity raises RegexpError only on strings with certain characters

• Status: Closed
• Assignee: duerst (Martin Dürst)
• Target version: -
• ruby -v: 2.4.1
• Backport: 3.2: REQUIRED, 3.3: DONE, 3.4: REQUIRED
• Tags: regexp, notbug

[dschweisguth (Dave Schweisguth)]

Here is a test program:

def test(description)
  begin
    yield
    puts "#{description} is OK"
  rescue RegexpError
    puts "#{description} raises RegexpError"
  end
end

test("ass, case-insensitive, special") { /(?<!ass)/i =~ '✨' }
test("bss, case-insensitive, special") { /(?<!bss)/i =~ '✨' }
test("as,  case-insensitive, special") { /(?<!as)/i  =~ '✨' }
test("ss,  case-insensitive, special") { /(?<!ss)/i  =~ '✨' }
test("ass, case-sensitive,   special") { /(?<!ass)/  =~ '✨' }
test("ass, case-insensitive, regular") { /(?<!ass)/i =~ 'x' }

Running the test program with Ruby 2.4.1 (macOS) gives

ass, case-insensitive, special raises RegexpError
bss, case-insensitive, special raises RegexpError
as,  case-insensitive, special is OK
ss,  case-insensitive, special is OK
ass, case-sensitive,   special is OK
ass, case-insensitive, regular is OK

The RegexpError is "invalid pattern in look-behind: /(?<!ass)/i (RegexpError)"

Side note: in the real code in which I found this error I was able to work around the error by using (?i) after the lookbehind instead of //i.

Running the test program with Ruby 2.3.4 does not report any RegexpErrors.

I think this is a regression, although I might be wrong and it might be saving me from an incorrect result with certain strings.

[Hanmac (Hans Mackowiak)]

did some checks on my windows system to check how deep the problem is.
i used "ä" as variable.

the same problem happens when you try to use match function too:
/(?<!ass)/i.match('ä')
also happen for
Regexp.union(/(?<!ass)/i, /ä/)

but i still don't understand why it does crash with ass, while ss works.
might have something todo how regexp are stored internal

[naruse (Yui NARUSE)]

I created a ticket in upstream: https://github.com/k-takata/Onigmo/issues/92

[gotoken (Kentaro Goto)]

I encountered a non ss case. Is this a same problem?

% ruby -ve '"".match(/(?<=ast)/ui)'
ruby 2.6.0dev (2018-08-27 trunk 64549) [x86_64-linux]
-e:1: invalid pattern in look-behind: /(?<=ast)/i

It was reproduced in version 2.4 and 2.5.
#14838 seems to be duplicate.

[znz (Kazuhiro NISHIYAMA)]

You can use (?:s) instead of s for workaround.

$ ruby -ve '/(?<=ast)/iu'
ruby 2.5.1p57 (2018-03-29 revision 63029) [x86_64-darwin17]
-e:1: invalid pattern in look-behind: /(?<=ast)/i
-e:1: warning: possibly useless use of a literal in void context
$ ruby -ve '/(?<=a(?:s)t)/iu'
ruby 2.5.1p57 (2018-03-29 revision 63029) [x86_64-darwin17]
-e:1: warning: possibly useless use of a literal in void context

[znz (Kazuhiro NISHIYAMA)]

• Related to Bug #14838: RegexpError with double "s" in look-behind assertion in case-insensitive unicode regexp added

[gotoken (Kentaro Goto)]

Thanks znz. The workaround is helpful. And I understood what was happened.
https://github.com/k-takata/Onigmo/issues/92#issuecomment-373981492 shows how some combinations of letters are variable length.

For example, "ss" and "st" are mapped "ß" ("\u00DF") and "st" ("\uFB06").
Those combinations are listed in ftp://ftp.unicode.org/Public/UNIDATA/SpecialCasing.txt

By the way, this expansion by //i option looks over kill for me.
I wish case sensitivity and SpecialCasing mapping were separated...

[shyouhei (Shyouhei Urabe)]

gotoken (Kentaro Goto) wrote:

By the way, this expansion by //i option looks over kill for me.
I wish case sensitivity and SpecialCasing mapping were separated...

I know how you feel. Too bad we are just doing what Unicode specifies to do.

See also http://unicode.org/faq/casemap_charprop.html#11

[gotoken (Kentaro Goto)]

Thanks shyouhei for your pointing out.

I imagine another Rexexp option, say //I, which is almost the same as //i except for never-applying SpecialCasing mapping.
This change extends Unicode matching indeed but does not introduce incompatibilities, IMHO.
A difficulty is the implementation is on the upstream library and cruby is just a user.

[duerst (Martin Dürst)]

gotoken (Kentaro Goto) wrote:

For example, "ss" and "st" are mapped "ß" ("\u00DF") and "st" ("\uFB06").
Those combinations are listed in ftp://ftp.unicode.org/Public/UNIDATA/SpecialCasing.txt

By the way, this expansion by //i option looks over kill for me.
I wish case sensitivity and SpecialCasing mapping were separated...

I still have to verify this, but currently I strongly suspect that the problem is NOT in SpecialCasing, but in how Onigmo (/Oniguruma?) implement it.

[mauromorales (Mauro Morales)]

FYI The issue has been addressed in Onigmo https://github.com/k-takata/Onigmo/pull/116 and has already been released in version 6.2.0. I tried it by applying the changes using Ruby 2.6.6 and it works as expected.

[mauromorales (Mauro Morales)]

Unfortunately, the problem persists in Ruby 2.7.2 and 3.0.0

[Eregon (Benoit Daloze)]

It seems ruby master as of today still uses Onigmo 6.1.3, but https://github.com/k-takata/Onigmo/releases/tag/Onigmo-6.2.0 is needed to fix this bug.

Who can update Onigmo to latest?

[duerst (Martin Dürst)]

Eregon (Benoit Daloze) wrote in #note-12:

It seems ruby master as of today still uses Onigmo 6.1.3, but https://github.com/k-takata/Onigmo/releases/tag/Onigmo-6.2.0 is needed to fix this bug.

Who can update Onigmo to latest?

If nobody else wants to urgently do this, feel free to assign this to me.

[mame (Yusuke Endoh)]

• Status changed from Open to Assigned
• Assignee set to duerst (Martin Dürst)

[Eregon (Benoit Daloze)]

@duerst (Martin Dürst) Could you take a look at this? It's not fixed yet in 3.2.0.

[nobu (Nobuyoshi Nakada)]

• Backport changed from 2.2: UNKNOWN, 2.3: UNKNOWN, 2.4: UNKNOWN to 3.2: REQUIRED, 3.3: REQUIRED, 3.4: REQUIRED

[k_takata (Ken Takata)]

• Status changed from Assigned to Closed

Applied in changeset git|bfbbcf34557b0aad4f5ed045df3e4e86b11c9a8c.

---

[Bug #13671] Fix that "ss" in look-behind causes syntax error

Fixes k-takata/Onigmo#92.

This fix was ported from oniguruma:
https://github.com/kkos/oniguruma/commit/257082dac8c6019198b56324012f0bd1830ff4ba

https://github.com/k-takata/Onigmo/commit/b1a5445fbeba97b3e94a733c2ce11c033453af73

[nagachika (Tomoyuki Chikanaga)]

• Backport changed from 3.2: REQUIRED, 3.3: REQUIRED, 3.4: REQUIRED to 3.2: REQUIRED, 3.3: DONE, 3.4: REQUIRED

ruby_3_3 f0feca1a8495eba2706a7914f0c4f8128c281366.