[
https://issues.apache.org/jira/browse/BEAM-134?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15216882#comment-15216882
]
Scott Wegner commented on BEAM-134:
-----------------------------------
TL;DR: 39 classes in the SDK would be candidates for AutoValue, 12 of which are
sufficiently complex. I recommend we do the work to incorporate AutoValue.
I did some analysis over the current SDK codebase, looking at classes which act
as value classes with boilerplate equals/hashCode implementations; these are
good candidates for AutoValue. Of those, I categorized them based on how much
they would value from being converted to AutoValue:
* "low": equality is represented by a single field. Little value in converting.
* "medium": equality represented by 2 fields; AutoValue would remove some
complexity and ensure equals/hashCode are implemented "correctly"
* "high": equality represented by 3 or more fields; at this point, the
boilerplate code is larger and equality is easy to get wrong or break over
time. These would benefit from AutoValue.
Based on those definitions, I categorized the SDK codebase, limiting to
non-test, non-runner code, and found:
* Low value: 16 classes
* Medium value: 21 classes
* High value: 12 classes
Based on this, I think AutoValue would be worth including and converting these
existing classes.
> Investigate use of AutoValue
> ----------------------------
>
> Key: BEAM-134
> URL: https://issues.apache.org/jira/browse/BEAM-134
> Project: Beam
> Issue Type: Improvement
> Components: sdk-java-core
> Reporter: Scott Wegner
> Assignee: Scott Wegner
> Priority: Minor
> Attachments:
> 0001-Mark-classes-which-might-benefit-from-AutoValue.patch
>
>
> The initial PR for [BEAM-118] added a dependency on AutoValue to
> auto-implement equality semantics for a new POJO. We decided to remove the
> dependency because the cost of adding the dependency for this feature may not
> be worth it for the value.
> However, we could use AutoValue for all of our POJO's, it might be worth it.
> The proposal here is to follow-up with an investigation on whether we would
> gain significant value to porting our code to use AutoValue instead of
> hand-written POJO's.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
