rvesse commented on PR #1616:
URL: https://github.com/apache/jena/pull/1616#issuecomment-1318422052
> my concern would be that this completely different path flatterning in
#1620 has never been put to wide testing and per the comment in the source file,
And my point is that neither is yours. You're changing the default
behaviour for all existing users based on testing on your own datasets.
As the recent flurry of issues around another new transform that was
introduced has shown optimisations that benefit one dataset don't necessarily
benefit all, and ultimately led to it being disabled by default (#1604). Hence
the suggestion of enabling an opt-in to the alternative transform rather than
changing the existing default.
> `It does not produce very nice execution structures so ARQ uses a
functional equivalent, but different, transformation.`. One obvious instance
from your tests, `:p1{2}` now becomes a join instead of a simple bgp (although
I guess the other optimisers will turn that back into a sequence which maybe
isn't quite as bad?)
Yes the full optimiser converts joins of adjacent BGPs or triple operators
back into a single BGP, expanded my test cases to explicitly cover this now.
>
> Another thought about Std should be given w.r.t inverse paths like
`^:p1+`, they are reversed by the current optimiser but remain in reverse in
the Std implementation. This might then cause worse behaviour in the
determineUngroundedStartingSet which does not handle them at the moment afaik
(fall-back to allNodes)
Well if both ends of the path are ungrounded whether we reverse it or not is
irrelevant, and regardless we're conflating the issue of deficiencies in the
path evaluator with algebra transformation.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
