Introducing PeepholeCombineArrayPush

[cshung]

Fixes issue #3212

[tjgq]

Created Google internal issue b/130801558.

Assigning to @brad4d for review.

[brad4d]

Overall this optimization doesn't strike me as a very useful one.
It's not likely that a human author would put a lot of array.push(), right next to each other,
so I'd think this would only be useful if other optimizations ended up putting them together, which
then allowed this optimization to do something useful. That also seems likely to be a rare thing to me.

Execution time is precious for large projects. Do you have a use-case in which the code size savings from this optimization will justify the time spent in looking for places where it can be done?

[cshung]

Overall this optimization doesn't strike me as a very useful one.

I agree. I was hunting for an issue to contribute to a compiler related project, and this one seems like a challenge that I would like to get started on, so I did it. It was marked help wanted and was commented as a good first issue, so I took it on, without a second thought. In retrospect, I probably should have considered the cost/benefit for the optimization.

Do you have a use-case in which the code size savings from this optimization will justify the time spent in looking for places where it can be done?

@dan1wang, can you help us to understand why do we want this optimization?

[dan1wang]

I think the optimization is useful only if the original array is pretty large. I don't know of a use-case where that's the case (I did find this ugly code, though: https://github.com/nemac/seldon/blob/master/areas.js).

I think it would be more worthwhile to address issue #3211 (string.concat & array.concat) because it's more likely one of the string/array is fairly large.

[brad4d]

Sounds like we've agreed to close this PR.
@cshung thanks for submitting it anyway.
Hopefully you still found the experience a valuable one and can apply it to creating another?
:)

[cshung]

Hopefully you still found the experience a valuable one and can apply it to create another?

@brad4d Can you please recommend an issue that is good?

[brad4d]

@cshung As it happens I've just been looking at #3343 which would probably be a good first PR.