perf: Drop the usages of SupervisedGraphStageLogic to reduce allocation.

[He-Pin]

Motivation:
There is a allocation of Some in SupervisedGraphStageLogic and for performance reason, I want to remove that.
I can choose the Option[T] to OptionVal[T], but that will cause the problem if the user specified function return null.
So I choose to drop the usage of SupervisedGraphStageLogic.

Result:
Reduce allocation, but repeated code.

Status: Mima file is not added yet, open for discussion, and I think should remove the SupervisedGraphStageLogic too.

[laglangyue]

I have reviewed some of your PRs, but due to my lack of familiarity with the project, I did not comment. But I noticed that there is a '=str' in the title all of them, and I don't quite understand what it means.

[He-Pin]

I have reviewed some of your PRs, but due to my lack of familiarity with the project, I did not comment. But I noticed that there is a '=str' in the title all of them, and I don't quite understand what it means.

I was using the old style, maybe chore nowadays ?

=means I added no api change, and str is abbr for stream module. BTW, I see you from China, are you in the wechat group ？

[mdedetrich]

Im with @pjfanning here, I would rather drop this convention (i.e. =str) because no one else is doing it so it kind of defeats the point.

For me, simple git commit messages with active tense as well as making sure commits are atomic as possible is whats most important, i.e. general guidelines as from https://reflectoring.io/meaningful-commit-messages/

[mdedetrich]

but that will cause the problem if the user specified function return null.

Is this a legitimate issue, I mean we can do behavioural changes if its well justified?

[He-Pin]

but that will cause the problem if the user specified function return null.

Is this a legitimate issue, I mean we can do behavioural changes if its well justified?

I think we can't, if I change to make use of OptionVal, then it will just ignore the value and do nothing, which was a Some(null) and throw an NPE.

and I will update the commit message then.eg perf:xxx

[laglangyue]

I was using the old style, maybe chore nowadays ?
=means I added no api change, and str is abbr for stream module. BTW, I see you from China, are you in the wechat group ？

yeah, I am in Hangzhou,Zhejiang, and I am newcommner for pekko/akka，I plan to introduce Pekko into my project in future.
it's my wechatId laglangyue, could you invite me to join the wechat group.
I am not picky about style, and just a simple question.

[He-Pin]

@laglangyue Are you using Pekko at work too, welcome and you are welcome to review this PR too.

[laglangyue]

I don't have a deep understanding of akka/pekko, so I can't comment on this PR.

@laglangyue Are you using Pekko at work too, welcome and you are welcome to review this PR too.

[Roiocam]

My first thought is that this is a trade-off between abstraction and performance.

Too much duplicate code is not good to maintenance and code readability.

Considering that other Operators (such as Fold) in Ops also use inline instead of SupervisedGraphStageLogic, I will not be too picky about the duplicate code brought by this PR.

[He-Pin]

Thank you both from China Scala community to review this, I was plan to add onErrorReturn and OnErrorResume and OnErrorComplete operators, which need very handy control around the supervision.

[He-Pin]

@mdedetrich @pjfanning ping~

[mdedetrich]

Too much duplicate code is not good to maintenance and code readability.

I wouldn't be surprised if the inliner (i.e. #857) will automatically do some of these performance improvements, unfortunately it would only work for Scala 2

[mdedetrich]

lgtm