Re: [basex-talk] count(//elem) not optimized, even though `elem` is in the index

Hi Gioele,

I wonder if the presence of the namespace somehow confuses the optimizer.

Exactly, that’s the reason. For some historical reason (but not such a wise one, as most quoted “historical reasons” are), we decided to index the node names without considering the namespace URI. As a result, the index:element-names function will yield…

<entry count="2">xml</entry>

…for the following document:

<xml> <xml xmlns='uri'/> </xml>

For the same reason, various optimizations that are based on the database statistics will only get into effect if a document contains no, or at most one global, namespace declaration. In various cases, optimizations could still be made possible (e.g. if we know that the element/attribute names with and without namespace URIs are distinct), but that hasn’t been implemented so far.

Cheers, Christian

I was stressing the BaseX 8.6 planner/optimizer when I noticed that expressions like `count(//elem)` are not optimized at all, even though they are correctly indexed, as demonstrated by `index:element-names()`.

The current database is a 300 MB TEI document. All the elements are in the `http://www.tei-c.org/ns/1.0%60 namespace.

The following test case will report the correct number, but it will take a couple of seconds to run, instead of a few milliseconds.

declare namespace tei="http://www.tei-c.org/ns/1.0";

let $n := index:element-names("monier")[. = 're']/@count

let $c := count(//tei:re)

return <res><in-index>{$n}</in-index><in-doc>{$c}</in-doc></res>

I wonder if the presence of the namespace somehow confuses the optimizer. The same problem can be observed running the same test case with

declare default element namespace "http://www.tei-c.org/ns/1.0";
[...]
let $c := count(//re)

Regards,

-- Gioele Barabucci gioele.barabucci@uni-koeln.de