On Aug 23, 2012, at 6:09, Christian Heimes <li...@cheimes.de> wrote:

> Am 23.08.2012 06:21, schrieb Andi Vajda:
>> I'm getting errors there too (if I build PyICU into PyLucene's install
>> tree, otherwise these tests are not enabled). I believe the problem lies
>> with the utr30.dat resource file. Their format requirements probably
>> changed with the upgrade to ICU 49.
>> 
>> I remember having a lot of trouble reverse engineering what format these
>> files should be transformed into when taken from the Java version, such as
>> for example, resources/org/apache/lucene/analysis/icu/utr30.dat.
>> 
>> As this feature is optional and rather experimental, I'm not prepared to
>> block the release for that bug. Not that it doesn't need to be fixed but
>> the tranformation of Lucene's Java ICU utr30.dat into C++ ICU 49's
>> utr30.nrm via the /usr/local/sbin/icupkg utility needs to be figured out
>> again for this to work properly.
> 
> I concur with your analysis. The tests are passing when I use Ubuntu's
> default version 48.1.
> 
> With 49.1.1 test suite segfaults somehwere in
> lucene/ICUTransformFilter.py. I'm unable to get the exact line number as
> the values were optimized out by the compiler.
> 
> Perhaps you could disable all ICU features when icu.ICU_VERSION is 49 or
> newer? That would prohibit people from getting segfaults.

I agree, a version check should be added to the tests and a proper warning 
should be emitted in case of mismatch.

Is it worth respinning a release candidate for this or should I go ahead and 
proceed with the release since the vote now has received the required three PMC 
votes for release ?

Andi..

> 
> Christian

Reply via email to