regexp search spits weird errors on some expressions

[bilbo]

I searched for expression:

name:[ěš]

(name containing either "ě" or "š")

and I got error

The regex "(?:[]|ě|ě|š|š)" had a parse error at offset 15

I remember these expressions worked few hundred revisions ago, so I guess something in the search got broken

Case sensitive and regular expression checkboxes are ticked.

[stoecker]

Situation is this: We use CANON_EQ and it seems this converts different codes into this:
(?:c1|c2), so ěš gets (?:ě|ě|š|š). Now the problem is, that the now empty [] is not removed correctly. This is clearly a bug in Java parsing code.

name:"(?:ě|š)" and name:"(?|[ěš]) fails for the same reason.

You can work around this only by adding something inside the [], so it wont get empty. Or you use name:"[(?|(ěš)]" (i.e. add the (?|...) inside the [] instead of outside as the CANON_EQ does.

@Jiri:
Any comments to this topic? Should we report is as a bug to Java?

[anonymous]

But if it is java bug, how is it possible that it worked before?

Is CANON_EQ needed? If it was not used, can we either remove it or make it optional in preferences?

[stoecker]

We did not use CANON_EQ before :-)

CANON_EQ ensures, that different forms of UTF-8 representation are all found when searching. This is an important feature.

I suggest to use the above workaround (or add more characters inside the [], which have only a single UTF-8, so the brackets don't get empty).

[stoecker]

I tried to enter a bug report for Java, but it seems this did not succeed. As I have no time to care for bugs where reporting is ignored, I'll close this. Nothing we can do about this. Use given workaround until they fix Java.

[bastiK]

Do we need CANON_EQ flag, if both pattern and string are normalized with java.text.Normalizer?

[stoecker]

You can try with the examples in this report to find out.

[bastiK]

How? I cannot find different representations of equivalent unicode string in this report.

[stoecker]

Hmm, seems copy and paste removed the differences. Cou can try [(?|(ěš)] with old josm and CANON_EQ and get the text from the error message :-)