Cole Greer created TINKERPOP-3231:
-------------------------------------
Summary: Optimize P.setValue() and P.getValue()
Key: TINKERPOP-3231
URL: https://issues.apache.org/jira/browse/TINKERPOP-3231
Project: TinkerPop
Issue Type: Improvement
Components: process
Affects Versions: 3.8.0
Reporter: Cole Greer
First raised by Pieter Martin in discord:
https://discord.com/channels/838910279550238720/893235809690464277/1470053903742537981
The updated getValue and setValue methods in 3.8.0 result in excessive
processing of collections for within/without predicates. This can be especially
harmful with large collections (millions of literals). Some of this processing
is necessary due to the needed separation between literals and GValues, however
some of it can be improved.
In the case of setValue(), it currently makes 2 complete passes through a
collection, the first to check if GValues are present, and the second to split
the list into literals and GValues. The intention of this 2 pass approach was
to bypass the type sorting in the "no GValue" case, and retain the original
collection type. This is likely unnecessary overhead which is unlikely to
provide justifiable value. The sorting is necessary but can be achieved without
this 2 pass approach.
For getValue(), it is necessary to re-unify the literals and variables
collections to retain the public interface of P. In the case where
`this.variables` is empty, we could likely bypass the cloning step and directly
return `this.literals`. There may be a concern with the method returning a
clone of the collection in some circumstances, and the original collection in
others. This warrants additional consideration, but perhaps is solvable by
wrapping the result in `Collections.unmodifiableCollection()`
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
