Saxon

I think my expectation would be that the result should be <w>2</w>. My reasoning: there are three candidate windows starting at positions 1, 2, and 3. These have $p=1, $p=2, and $p=3 respectively. For the first and third windows, the condition $p=2 is never true. For the second window, it is true for the first item in the window (in fact, the condition doesn't depend on which item you are looking at), so the window ends immediately.

I believe to remember that sliding windows were introduced by the Zorba guys, so I assumed their result may be correct, but I think they got it wrong: <w>1 2</w> cannot be a correct result, as the position $p is set to 1 for the complete first sliding window.

I now wonder if <w>2</w> shouldn’t be returned for the second sliding window?

• Sliding Window 1: 1 2 3
• Sliding Window 2: 2 3

Here’s a slightly more complex query which doesn’t return <_ b='2'/><_ c='4'/><_ d='3'/> as one might expect:

for sliding window $w in <x><_ a='1'/><_ b='2'/><_ c='4'/><_ d='3'/></x>/_
start $s next $n when $s/@* < $n/@*
only end $e previous $p when $e/@* < $p/@* and $p is $n
return ($s, $n, $e)

…thanks for your feedback; good to know that our expectations are in line.

Saxon is "despatching" a window with the content (2), but because windows have to be output in order of start position, and the window (1,2) is still open at that stage, the window (2) is put on a queue to be output when the disposition of (1,2) is known.

But the logic to empty the queue at the end is failing; the condition windowClause.isIncludeUnclosedWindows() in WindowClausePush line 112 returns false, because of the keyword "end only when...".

Note that we have two versions of the logic, for use in push and pull mode respectively, and the same problem appears to be present in both.

I've created push/pull variants of some of the existing tests.