Till,
On further thought, I think I have a better solution -- one that is more
general and does not require any compiler support.
Add a method to the iterator of the form
int skip(int len);
The contract would be that skip tries to skip len items and returns len
- n as the return value where n is the number of items it could actually
skip (Can be < len, if it reached end-of-sequence).
The default implementation could be:
while(len-- > 0 && in.next() != null);
return len + 1;
The iterator of the Sequence could do this in a smarter way.
The second part is the use of skip in subsequence.
This will give us the speed improvement for the query.
What do you think?
Vinayak
Vinayak Borkar wrote:
In this specific case, the sequence is materialized anyways since its
assigned to a global variable and then used in a loop. If we add a
subsequence() method on XDMSequenece, that creates the subsequence in
constant time.
The compiler could replace the subsequence function with the
subsequence-of-materialized-sequence function when it notices the
materialization condition.
Vinayak
Till Westmann wrote:
How would you optimize this?
Materialize the sequence and index it?
Or evaluate multiple expressions on the sequence during one scan?
Or ...?
Till
On Jan 5, 2010, at 12:55 AM, Till Westmann wrote:
Hmm, is this test part of a later revision of the test suite? For me
test execution never was a real problem ...
How slow is it?
Thanks,
Till
On Jan 4, 2010, at 9:50 PM, Vinayak Borkar wrote:
Its not an infinite loop --
Expressions/Construct/DirectConElem/DirectConElemContent//Constr-cont-document-3.xq
is taking a long time to run. I am pasting the query below.
The query iterates over about 1.1 M integers 70 at a time and calls
subsequence() for the 70 items in the window. This is a quadratic
operation in the engine -- and takes a long time.
Should we optimize this case?
Thanks,
Vinayak
--- Query begin ---
declare variable $codepoints as xs:integer+ := (9, (: 0x9 :)
10,(: 0xA :)
13,(: 0xD :)
32 to 55295, (: 0x20 -
0xD7FF :)
57344 to 65532, (:
0xE000 - 0xFFFD :)
65536 to 1114111 (:
0x10000 - 0x10FFFF :));
declare variable $count as xs:integer := count($codepoints);
declare variable $lineWidth as xs:integer := 70;
<allCodepoints>
<!-- Each <r>-element represents a codepoint range. The 's' attribute
is the start codepoint, the 'e' attribute is the end codepoint.
Note that these are only *Hints*, since the character range
is not contiguous.
-->
{
"
",
"
",
(: The outputted file is rather big, so to make it managable, we
output
a chunk of $lineWidth characters in each element.
:)
for $i in (1 to $count idiv $lineWidth)
let $startOffset := (($i - 1) * $lineWidth) + 1
return (<r s="{$codepoints[$startOffset]}"
e="{$codepoints[$startOffset] + $lineWidth}">
{
codepoints-to-string(subsequence($codepoints,
$startOffset, $lineWidth))
}
</r>, "
")
}
</allCodepoints>
--- Query end ---
Vinayak Borkar wrote:
Till,
Does XTest on the test suite terminate? Some query is throwing the
engine into an infinite loop. Do you see this?
Thanks,
Vinayak
