@gaogaotiantian this PR adds an instruction (CHECK_PERIODIC) after each call. So now pdb.set_trace() sees that as the next instruction, and thinks it's breaking one line earlier. How do you suggest we fix this? Skip the CHECK_PERIODIC instruction (where would the happen?)

The test_pdb failure is not a real failure. We should just fix(remove) the test. It was a corner case where the instruction afterCALL (POP_TOP) does not have a line number associated with it - and we should not stop there because missing line number would frustratepdb. However, if the following instruction isCHECK_PERIODIC and it has a line number, we should just stop there.

CHECK_PERIODIC has the same line number as the CALL of set_trace(). So it will look like we stopped just before the set_trace call. I don't think that's what we want.

That is what we want. The existing test that failed is an exception, which stopped at the next line. We put a lot of effort to makebreakpoint() stop atbreakpoint(), so this is a desired behavior.

It looks like thespecialize_method_descriptor needs to account for theCHECK_PERIODIC before thePOP_TOP, and presumablyCALL_LIST_APPEND will need to skip the two following instructions.

All commit authors signed the Contributor License Agreement.

It looks liketranslate_bytecode_to_trace also needs adjusting to account for the extra instruction.