bpo-39166: Fix trace of last iteration of async for loops

[pablogsal]

https://bugs.python.org/issue39166

[pablogsal]

@markshannon Could you take a look at this when you have time? I am not happy about this solution as is just adding more shrinkwrap but solving the general case is not immediate. Do you have a better idea for now?

@nedbat Could you check if this breaks anything else in your test suite?

[nedbat]

All my tests pass with this change, thanks!

[pablogsal]

s/incorrent/incorrect/

Thanks for the catch!

Will correct the other one in a separate PR!

[markshannon]

Unless I am misunderstanding the code, this seems rather fragile as it relies on the compiler emitting the async for exception handler non-aligned with a line and all other exception handlers aligned with a new line.
That seems an unreasonable restraint on the compiler and is not documented anywhere.

Again, it seems like we need a way to signal that instructions are artificial and should be ignored by tracing.

[pablogsal]

all other exception handlers aligned with a new line.

Not exactly, because other exception handlers will have needs_line_update = True.

That seems an unreasonable restraint on the compiler and is not documented anywhere.

Yup, that is why I said:

I am not happy about this solution as is just adding more shrinkwrap

---

Again, it seems like we need a way to signal that instructions are artificial and should be ignored by tracing.

Yup, I fully agree, but is not going to be that immediate, because we need to manage also pushing and poping blocks in the frame in the eval loop from artificial instructions.

How do you want to proceed? Do you want to close this and wait until we have a more resilient solution or you want to merge this (and the tests) meanwhile?

[markshannon]

I'm going to merge this.
It isn't perfect , but it is an improvement and it doesn't prevent further improvements.
@nedbat, @pablogsal Any final comments before I merge?

[nedbat]

None from my side. I'm still wondering if there's a lowish-cost way to transfer some of the coverage.py tests to the CPython test suite, to shrink the feedback loop.

[pablogsal]

I'm still wondering if there's a lowish-cost way to transfer some of the coverage.py tests to the CPython test suite, to shrink the feedback loop.

I have plans to work on this soonish, I will let you know once I have some free cycles. :)

[pablogsal]

Any final comments before I merge?

Nop, thanks for the review! Let's sync in the future to try to work on a more resilient solution :)

[bedevere-bot]

@markshannon: Please replace # with GH- in the commit message next time. Thanks!