[
https://issues.apache.org/jira/browse/TINKERPOP3-700?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14582282#comment-14582282
]
ASF GitHub Bot commented on TINKERPOP3-700:
-------------------------------------------
GitHub user mhfrantz opened a pull request:
https://github.com/apache/incubator-tinkerpop/pull/76
TINKERPOP3-700 Path getSingle/getList improvements
As requested in the comments of TINKERPOP3-700. I also reversed the sense
of `Pop` to align with `tail(local)`. Thus `Pop.tail` now means the most
recent value.
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/RedSeal-co/incubator-tinkerpop
TINKERPOP3-700-Path-getSingle
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/incubator-tinkerpop/pull/76.patch
To close this pull request, make a commit to your master/trunk branch
with (at least) the following in the commit message:
This closes #76
----
commit e1e1a40b068ffe6da3d321256d6200dee0504074
Author: mhfrantz <[email protected]>
Date: 2015-06-11T16:47:41Z
Add javadoc for Path getList/getSingle
commit 4939cf0a22c4fc44a4f86ed3e023fa30e4e872cc
Author: mhfrantz <[email protected]>
Date: 2015-06-11T17:10:27Z
Reverse sense of Path.getSingle Pop to align with tail(local) step
commit 027d6319f9ff06ca26d785e83bf17b21364c3dca
Author: mhfrantz <[email protected]>
Date: 2015-06-11T17:51:23Z
Optimize MutablePath.getSingle
commit 5e9ad8d7dff6c63b9746147a1d255cbf7258cea3
Author: mhfrantz <[email protected]>
Date: 2015-06-11T17:52:43Z
Optimize ImmutablePath getSingle and getList
----
> WhereStep should "MatchStep" and ConjunctionP should use the BudgetAlgorithm
> ----------------------------------------------------------------------------
>
> Key: TINKERPOP3-700
> URL: https://issues.apache.org/jira/browse/TINKERPOP3-700
> Project: TinkerPop 3
> Issue Type: Improvement
> Components: process
> Reporter: Marko A. Rodriguez
>
> {code}
> g.V.as('a').where(a.knows.b
> a.knows.c
> b.knows.c)
> {code}
> The above can be written as an OrP of the form:
> {code}
> g.V.as('a').or(select(a).knows.b.select(b).knows.c.select(a).knows.where(eq(c)),
>
> select(a).knows.c.select(a).knows.b.select(b).knows.where(eq(c)));
> {code}
> In essence, the where-statements are rewritten in terms of every possible
> permutation. When these permutations are put into an OrP (via
> or(traversals…)), then if any branch returns a result, then the original 'a'
> is emitted (as {{WhereStep}} is a {{FilterStep}}). If {{OrP}} is under the
> BudgetAlgorithm, then {{OrP}} will "thread between" its traversals until a
> value is yielded. *And given that all permutations are the same semantics --
> if one fails, they all fail!*
> What is nice about this, is that arbitrary nesting comes "for free."
> {code}
> g.V.as('a').where(a.knows.b
> a.uncle.b
> and(a.worksFor.c
> b.worksFor.c))
> {code}
> This is rewritten as:
> {code}
> g.V.as('a').or(select(a).knows.b.or(select(a).worksFor.c.select(b).worksFor.where(eq(c)),select(b).worksFor.c.select(a).worksFor.where(eq(c))).select(a).uncle.where(eq(b)),
>
> select(a).uncle.b.select(a).knows.where(eq(b)).or(select(a).worksFor.c.select(b).worksFor.where(eq(c)),select(b).worksFor.c.select(a).worksFor.where(eq(c))))
> {code}
> *IMPORTANT* This is not a "match" in the {{MatchStep}} sense as it doesn't
> return all permutations that bind, it only filters based on a single match.
> What is interesting about this approach:
> 1. The rewrite algorithm seems simple as its just concatenation given
> {{select()}}-projections and {{where(eq)}}-tails.
> 2. The cool thing about the rewrite in all possible permutations is that if
> any one {{FastNoSuchElementException}}, its booted from the {{ConjunctionP}}
> analysis.
> 3. {{ConjunctionP}} has the BudgetAlgorithm and thus can be used for ANY
> step that has conjunctions -- {{HasStep}}, {{IsStep}}, etc.
> 4. It uses the path data structure to maintain the variable bindings.
> {{WhereStep}} has no state! Its all about {{OrP}}.
> 5. Given that the path data is the variable bindings, then this also works
> for OLAP as the traverser contains all the information it needs (no central
> location of analysis!)
> - However, you would only pick one permutation to do as `or()` does not
> exist in OLAP.
> - and with one permutation, {{where().select()}} is then {{MatchStep}}
> which would then work in OLAP!
> - thus, Gremlin OLAP can rewrite {{match()}} to the
> {{where().select()}} form and TADA!
>
> *IMPORTANT* 4 and 5 above are pretty insane consequences. And if any, we
> should at least use this realization to make {{match()}} work in OLAP.
> Next, realize that how {{where()}} should work is that if an {{as()}} is NOT
> in the path data structure, then its a variable bindings for rewrite.
> Moreover, if you don't provide a start {{as()}}, it is assumed to be the
> incoming object (currently how {{where()}} works). For example:
> {code}
> g.V.where(knows.b
> knows.c
> b.knows.c)
> {code}
> This is rewritten as:
> {code}
> g.V.or(x.select(x).knows.b.or(select(b).worksFor.where(eq(a)),select(x).worksFor.where(eq(b))).select(x).uncle.where(eq(b)),
>
> x.select(x).uncle.b.select(x).knows.where(eq(b)).or(select(b).worksFor.where(eq(x)),select(x).worksFor.where(eq(b))))
>
> {code}
> To be sure, the {{as('a').select('a')}} fragments can of course be optimized
> out to just {{as('a')}}.
>
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
