Re: xpath processing brain dead

Andrew Dunstan wrote:
>
>
> Tom Lane wrote:
>> Hmm, does this proposal require adding a test of well-formed-ness to
>> a code path that doesn't currently have one? If so, is that likely
>> to contribute any noticeable slowdown?
>>
>> I can't offhand see an objection to this other than possible performance
>> impact.
>>
>>
>>
>
> Yeah, testing the well-formedness might cost a bit. We could
> short-circuit the test by applying some comparatively fast heuristic
> tests.
>
> Or we could decide that we'll just fix the xpath prefix part for 8.3
> and keep the wrapping. I don't want to spend a huge effort on fixing
> something I regard as fundamentally broken.
>
> I'll do some tests to see what the cost of extra xml parsing might be.

The extra cost appears to be fairly negligible.

regression=# create table xpathtest3 as select xmlconcat(xmlelement(name
unique1, unique1), '\n\t',xmlelement(name unique2, unique2),
'\n\t',xmlelement(name two, two), '\n\t',xmlelement(name four,
four),'\n\t',xmlelement(name ten,ten),'\n\t',xmlelement(name
twenty,twenty),'\n\t',xmlelement(name
hundred,hundred),'\n\t',xmlelement(name
thousand,thousand),'\n\t',xmlelement(name
twothusand,twothousand),'\n\t',xmlelement(name
fivethous,fivethous),'\n\t',xmlelement(name
tenthous,tenthous),'\n\t',xmlelement(name
odd,odd),'\n\t',xmlelement(name even,even),'\n\t',xmlelement(name
stringu1,stringu1),'\n\t',xmlelement(name
stringu2,stringu2),'\n\t',xmlelement(name string4,string4),'\n') from tenk1;

regression=# select count(*) from (select
xpath('//two[text()="0"]/text()',xmlconcat) as elems from xpathtest3,
generate_series(1,10) ) x ;
count
--------
100000
(1 row)

Time: 27.722 ms

Proposed patch for 8.3 attached. (Note: it only reparses in the
non-document case)

cheers

andrew