Is the code at the bottom of OutputFileStream.cpp correct? The
gzip_compressor is not being properly flushed in my build. I tried
change it to this->reset(), that didn't help.

On Wed, Jul 4, 2012 at 10:33 AM, Kenneth Heafield <[email protected]> wrote:
> Hi,
>
>         It looks like
>
>    set< string > targetLabelCollection, sourceLabelCollection;
>    map< string, int > targetTopLabelCollection, sourceTopLabelCollection;
>
> are modified by multiple threads without any form of locking.
>
> Kenneth
>
> On 07/04/2012 10:12 AM, Alexander Fraser wrote:
>> In the current version built against boost 1 50, with #undef
>> WITH_THREADS I am getting invalid gzipped output using this option:
>>
>> cat source.00000 target.00000 align.00000
>> 1
>> 1
>> 0-0
>>
>> /home/fraser/src/master/bin/extract-rules target.00000 source.00000
>> align.00000 extract.00000 --GlueGrammar glue-grammar.1 --MaxSpan 10
>> --GZOutput
>>
>> zcat extract.00000.gz
>> 1 [X] ||| 1 [X] ||| 0-0 ||| 1 |||
>>
>> gzip: extract.00000.gz: invalid compressed data--length error
>>
>> BTW, you are writing the glue-grammar to the same filename
>> (model/glue-grammar.1) in all threads, this is likely to cause
>> problems. It would be better to specify this option only for the first
>> run.
>>
>> I am using the multithreaded GNU sort, I'll try the options you
>> specified, thanks.
>>
>> Cheers, Alex
>> _______________________________________________
>> Moses-support mailing list
>> [email protected]
>> http://mailman.mit.edu/mailman/listinfo/moses-support
> _______________________________________________
> Moses-support mailing list
> [email protected]
> http://mailman.mit.edu/mailman/listinfo/moses-support
_______________________________________________
Moses-support mailing list
[email protected]
http://mailman.mit.edu/mailman/listinfo/moses-support

Reply via email to