Xapian discuss - Mar 2006 - bug in query parser or wrong usage?

Hi, I noticed this little oddity when I was trying to construct certain 
queries against my various boolean terms.  It can be illustrated just 
with the ParserQuery class alone:

 >>> from xapian import QueryParser
 >>> p = QueryParser()
 >>> p.add_boolean_prefix('x', 'X')
 >>> p.add_boolean_prefix('y', 'Y')

These two work:

 >>> p.parse_query("x:1 AND y:2").get_description()
'Xapian::Query((X1 AND Y2))'
 >>> p.parse_query("x:1 y:2").get_description()
'Xapian::Query((X1 AND Y2))'

but this one is different, seems to be treating OR like a positional term:

 >>> p.parse_query("(x:3 OR x:4) x:1 y:2").get_description()
'Xapian::Query((or:(pos=1) FILTER (X3 AND X4 AND X1 AND Y2)))'

same with this one:

 >>> p.parse_query("(x:3 OR x:4) AND x:1 OR
y:2").get_description()
'Xapian::Query(((or:(pos=1) OR and:(pos=2) OR or:(pos=3)) FILTER (X3 AND 
X4 AND X1 AND Y2)))'

but these two work as expected:

 >>> p.parse_query("x:3 OR x:4").get_description()
'Xapian::Query((X3 OR X4))'
 >>> p.parse_query("x:3 OR x:4 AND x:5").get_description()
'Xapian::Query((X3 OR (X4 AND X5)))'

seems to somehow be related to having the OR subexpression:

 >>> p.parse_query("(x:3 OR x:4) AND x:5").get_description()
'Xapian::Query(((or:(pos=1) OR and:(pos=2)) FILTER (X3 AND X4 AND X5)))'

 >>> p.parse_query("x:5 AND (x:3 OR x:4)").get_description()
'Xapian::Query(((and:(pos=1) OR or:(pos=2)) FILTER (X5 AND X3 AND X4)))'
 >>>

Am I using the parser wrong? or is this a bug?

-Michel

On Fri, Mar 24, 2006 at 04:03:56PM -0800, Michel Pelletier
wrote:
> but this one is different, seems to be treating OR like a positional term:
> 
> >>> p.parse_query("(x:3 OR x:4) x:1
y:2").get_description()
> 'Xapian::Query((or:(pos=1) FILTER (X3 AND X4 AND X1 AND Y2)))'
This is how boolean filter terms currently work.  You can't build
boolean expressions from them.

I'd say this is a bug, but fixing it properly is a little involved 
as I think it requires changes to the Query class itself, so it'll
take a but of time.

A sort-of workaround is to use non-boolean prefixes instead of
boolean ones.

Cheers,
    Olly