Yep, it can also be removed by doing DELETE /dbname/docid?rev=...

I think the workaround patch needs to be at a lower level than the view 
updater, as I believe replication will also break when it encounters the bad 
document.  Regards,

Adam

On Sep 1, 2010, at 2:40 PM, Jan Lehnardt wrote:

> Thanks Adam for finding this one. I ran into it a couple of times and I 
> thought I'm crazy.
> 
> I think the view server should skip the invalid doc and print a warning in 
> the log file with the doc id when it does.
> 
> I believe a _bulk_doc request with a _deleted:true member still does allow 
> removal of that doc, but I haven't tried in a while.
> 
> Cheers
> Jan
> -- 
> 
> 
> On 31 Aug 2010, at 07:25, Adam Kocoloski wrote:
> 
>> It turns out that mochijson2 will incorrectly decode an invalid UTF-8 string 
>> if the illegal byte sequence in the string occurs after an escaped character 
>> (COUCHDB-875).  This means that one can store documents which will never be 
>> successfully retrieved or indexed in CouchDB 1.0.  Moreover, once one of 
>> these documents makes it into the DB a view build on that DB will never 
>> complete.
>> 
>> I wonder what we should do to circumvent that problem?  At the very least it 
>> might make sense for the view indexer to skip documents which contain 
>> invalid UTF-8.
>> 
>> Adam
>> 
> 

Reply via email to