Hi,I can confirm, that positional argument can be sometime quite costly.
I had xpath in my xquery, which selects elements according to my criteria and pick up any of them, just to make sure I will get only one item in the result and not a sequence.
Original query was something like
$ltn//PredefinedLocation[Location/TmcLinear/@locationCode = $primarycode]/Location[1]
and replacing it by
($ltn//PredefinedLocation[Location/TmcLinear/@locationCode = $primarycode]/Location)[1]
speeded up my query from 30s to 35ms.
One of reasons could be, my original xpath was simply wrong and was still returning a sequence.
But I realized, that using indexes it is probably quite costly to determine node position.
Jan
2011/2/27 Christian Grün
<christian.gruen@gmail.com>
Hi Phil,
If you know that this node will occur only once, the most efficient
option will be to use a positional predicate:
( //*:siteinfo ) [1]
But you may be surprised that the following query is evaluated very quickly:
count(//*:siteinfo)
This means that the path index has indeed enough information to allow
for a faster evaluation: we're not saving direct references to the
target nodes (as such an index would get very large for e.g. the
Wikipedia page element), but we're saving the number of distinct node
paths. As a result, we could rewrite your query into
( //*:siteinfo ) [position() <= 1]
We haven't included this optimization yet, as the additional predicate
may slow down other queries; but in your case, it would clearly speed
up the evaluation time to a few milliseconds (if at all). I have added
a GitHub issue to remember your thoughts:
https://github.com/BaseXdb/basex/issues#issue/29
> While personally I very much dislike namespaces, they are common,
> and they have to be efficiently handled.
Namespaces, a great topic... It's true that name tests with prefixes
will be evaluated slower than queries without prefixes (i.e., prefix
wildcards). This is something most XQuery implementations suffer from,
as the complex nature of namespaces does not enable simple reference
checks. Indeed, most members of the W3 XML Query Working Group regret
that namespaces have not been specified much simpler; due to all
legacy issues, history cannot be reverted in that aspect.
After all, however, I was surprised to see that your query nearly took
twice the time as the one without namespaces; I'd have expected a
slowdown of maybe 10-15%. To conclude this: if you want faster
queries, you should declare global namespaces, or simply use
wildcards.
Hope this helps,
Christian
_______________________________________________
BaseX-Talk mailing list
BaseX-Talk@mailman.uni-konstanz.de
https://mailman.uni-konstanz.de/mailman/listinfo/basex-talk
--