Mark Shannon
Mark Shannon
I'd rather keep this open, otherwise it gives a false impression of being finished when there's some clean up to do. Or, make new issues for the list above and...
The tracer be abandoning the trace if it encounters an instrumented instruction. And instrumentation should invalid the trace currently being recorded as well as all existing ones. @Fidget-Spinner is this...
We aren't abandoning the trace if we encounter an instrumented instruction. https://github.com/python/cpython/blob/main/Python/bytecodes.c#L5608 should include `opcode >= MIN_INSTRUMENTED_OPCODE`
We also need to abandon the trace if an exception is raised, and before calling any monitoring callback functions on exceptional paths.
This PR has some serious issues: * Moving all the JIT state to the thread does not make things thread safe. It might do the opposite by introducing race conditions....
It's a bug. Want to fix it or shall I?
The changes look good. I'm concerned about the complexity of this for such an obscure use case, but I can't suggest a better way of doing it.
If there is a CPython bug, please report it so we can fix it.
I think we should to stick to the PEP as written. In other words: * `PyEval_GetLocals()` should return a borrowed reference to a `FrameLocalsProxy`. If called multiple times for the...
> returning a FrameLocalsProxy will not decrease the chance that they crash. Hopefully, it will increase the chance of a crash. Crashes are easy to diagnose and fix. I'm more...