[
https://issues.apache.org/jira/browse/GROOVY-8377?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16248826#comment-16248826
]
ASF GitHub Bot commented on GROOVY-8377:
----------------------------------------
GitHub user erdi opened a pull request:
https://github.com/apache/groovy/pull/633
GROOVY-8377 Equals and hash code ast performance
See [GROOVY-8377](https://issues.apache.org/jira/browse/GROOVY-8377) for
the rationale behind this PR.
Given how simple the fix is and its impact (from 3k ops/ms to 30k ops/ms on
my machine) I'd suggest considering backporting of it to 2.4.x, 2.5.x and 2.6.x
lines.
<rant>
Working on this was harder then it should be down to:
- the build takes a minute(!) to configure, I believe it is down to [this
code](https://github.com/apache/groovy/blob/1c42f421edd91f31aa35a3e7b7113910718b35d5/build.gradle#L93)
- the jmh benchmarks for Groovy code are broken and [needed fixing as part
of my
PR](https://github.com/erdi/groovy/commit/fc1d085a120bd7a542edeb7c1ff4b06816b776b2)
I understand that I worked against master, but the experience was highly
unwelcoming.
</rant>
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/erdi/groovy
equals-and-hash-code-ast-performance
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/groovy/pull/633.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 #633
----
commit fc1d085a120bd7a542edeb7c1ff4b06816b776b2
Author: Marcin Erdmann <erd...@gmail.com>
Date: 2017-11-12T10:23:26Z
Fix Groovy dependency in performance tests to be the same as the one used
by Gradle so that jmh benchmarks using Groovy don't fail
commit 3c6d1d462569259aad6ef95b73af4c17837595b0
Author: Marcin Erdmann <erd...@gmail.com>
Date: 2017-11-12T10:25:03Z
Add a benchmark for hashCode() generated using @EqualsAndHashCode when
instance on which hashCode() is called has null fields and properties
commit 2ce280083e30d912829d010592384c72b5b406cf
Author: Marcin Erdmann <erd...@gmail.com>
Date: 2017-11-12T10:26:39Z
Add a test to verify that changes to generating hashCode() for instances
with null fields and properties and EqualsAndHashCodeASTTransformation have no
impact on the value returned
commit bd2b79c2cf67b99020265dca2465e18b2a40b2c8
Author: Marcin Erdmann <erd...@gmail.com>
Date: 2017-11-12T10:29:41Z
Improve performance of generated hashCode() methods for instances with null
fields and properties by avoiding calling NullObject.is() method for which no
call site caching occurs.
----
> hashCode() generated by @EqualsAndHashCode is inefficient for instances with
> null fields or properties
> ------------------------------------------------------------------------------------------------------
>
> Key: GROOVY-8377
> URL: https://issues.apache.org/jira/browse/GROOVY-8377
> Project: Groovy
> Issue Type: Bug
> Affects Versions: 2.4.12, 2.5.0-beta-2, 2.6.0-alpha-2
> Reporter: Marcin Erdmann
> Labels: performance
>
> In a project I work on there is a relatively large hash map for which the key
> type has a {{@EqualsAndHashCode}} generated {{hashCode()}}. After introducing
> an additional property to the key type, for which the value is {{null}} on
> most instances, I noticed a massive degradation in efficiency of adds to the
> aforementioned map.
> After some digging around it turns out that the AST generated by that
> transform performs an identity check using {{is()}} between the
> property/field value and {{this}}. In cases where the property/field value is
> {{null}} this leads to a {{NullObject.is()}} call, for which, as far as I
> understand, no call site caching occurs.
> A PR with a jmh benchmark and a very simple fix to follow.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
