I'd rather see https://bugs.python.org/issue17611 finished than review another bytecode PR.

I prefer to fix separate issues by separate commits. Fixing this issue will make patches for bpo-17611 simpler. Currently they contain changes for fixing this issue (not completely).

In particularly this could limit changes in PyCompile_OpcodeStackEffect by only changes related to new and changed opcodes.

OTOH reviewing two bytecode and compiler PRs is more work than reviewing a single one. And I would like to see how Mark's approach works with this.

Isn't reviewing reviewing two separate PRs for two issues easier than reviewing the single PR that mixes fixes of two issues in the same code? Or rather reviewing three PRs every of them fixing the same two issues by slightly different code?

Well, those three PRs will have to be regenerated because of conflicts after this PR gets merged.

In any case, if you want to go forward with this, I would like to see more comments (especially on opcodes with a variable stack effect).

In this PR I use slightly different approach than in #5006, and it looks more correct to me. Now I better understand this code. I'll update #5006 accordingly.

What information do you want be added?

Could you please take a look again @pitrou? All stack effects are now defined in a single place.

This looks ok to me. I can't guarantee that the compiler changes for async for and try...except are right, but the test suite validates them.