Meeting minutes
<lisp> ? are those previous/next links intentional or unavoidable?
Scribe?
AndyS: next time look at all SPARQL PRs (item 4)
James: previous meeting dates are strange
External comment
AndyS: that appears to be a feature of the WG - all WG-related meetings are linearly ordered
Andys: one solution is semi-join, one injects bindings
pfps: there is the other part of Boris's message about confusion between syntax and algebra
AndyS: olaf responded to that
pfps: when there is a new version we will respond to Boris
AndyS: sounds right
Outline for changes for PR.
AndyS: I did some work on the PR
AndyS: the flow is better, but there is more work to be done
AndyS: [description of some of the changes]
Olaf: I'm not sure whether it is necessary to make these changes
<olaf> https://
Olaf: [the above link is my solution]
Olaf: [description of definitions in solution]
Olaf: [basic idea is to insert a values clause at the beginning of EXISTS pattern]
pfps: what about injecting a blank node
Olaf: yes, a problem
AndyS: when is the string created
Olaf: for each solution mapping in the simple version, but my advanced version uses a method where only one translation is needed
Olaf: here the current solution mapping is joined to the pattern
AndyS: filters in the EXISTS pattern don't have access to injected variables
AndyS: instead of modifying the pattern you can modify every BGP in the pattern
pfps: there are three options - like MINUS, like OPTIONAL, deep injection
AndyS: this is all semijoin, i.e., like MINUS
Olaf: what about subselects?
AndyS: look at UNION, with a FILTER in one side
AndyS: to get the solution mapping into there you need to inject (into BGPs)
AndyS: variables that are projected out in sub-SELECT are another issue
Andys: you can either affect them or not - one method is to rename them
james: the solution is a strict consraint on the scope of injection
james: andy's solution injects the solution mapping in more places
james: my concern about correlation (injection) is having to re-evaluate for every solution mapping
AndyS: the interior patterns are simple, usually
james: is it possible to have optimized execution even when correlation is needed
AndyS: so the idea is to pass all the rows at once, not just one at a time?
james: yes
AndyS: would this be exactly the same? maybe
olaf: I think it is not possible to define it this way because FILTER is per-solution mapping
AndyS: minus and other negation could be a problem
gtw: aggregation is also an issue
olaf: I'll look at the suggestions
Topic for next time
AndyS: would it be useful to more than just EXISTS, i.e., to look at other PRs
pfps: sounds reasonable
AndyS: there is other work that Olaf is pursuing
james: is there a list of PRs to look at?
AndyS: the ones on SPARQL Query