New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
bpo-45256: Remove the usage of the C stack in Python to Python calls #28488
Conversation
|
If you want to schedule another build, you need to add the " |
Python/ceval.c
Outdated
| static InterpreterFrame* | ||
| _PyEval_FrameFromPyFunctionAndArgs(PyThreadState *tstate, PyObject* const *args, int nargs, PyObject *function) { | ||
| assert(PyFunction_Check(function)); | ||
| size_t nargsf = nargs | PY_VECTORCALL_ARGUMENTS_OFFSET; |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
nargsf and vector_nargs below are only used in asserts.
I think all the asserts and assignments to nargsf and vector_nargs could be replaced with assert(nargs == 0 || args != NULL);
|
When you're done making the requested changes, leave the comment: |
|
@markshannon Damn, unfortunately something is going on with Windows: Apparently that is a |
|
@markshannon Oh, I found the source of the Windows failure: it was an existing bug in the That was challenging to find indeed! |
|
If you want to schedule another build, you need to add the " |
Python/ceval.c
Outdated
| Py_DECREF(function); | ||
| _PyFrame_SetStackPointer(frame, stack_pointer); | ||
| new_frame->depth = frame->depth + 1; | ||
| tstate->frame = frame = new_frame; |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Note to self, no need to pop frame here because it's done on eval frame exit at exit_eval_frame.
|
I cannot reproduce the ASAN bug. :( |
|
There're refleaks but I can't pinpoint where :( |
Ths commit inlines calls to Python functions in the eval loop and steals all the arguments in the call from the caller for performance.
|
Commit 0ddc46e fixes the reference leaks (but it needs some cleanup as it requires simplification): |
|
If you want to schedule another build, you need to add the " |
I hadn't realized the docs were so sparse. Probably best to do it in another PR. |
|
@markshannon what do you think of 0ddc46e for now? I dislike this approach (is correct, but is difficult to reason about if you don't have the full picture in mind). Should we do the refactor now, should we do a smaller refactor or should we just do some celanup of 0ddc46e ? |
Just do the minimal cleanup of 0ddc46e that you are happy with for now. |
|
If you want to schedule another build, you need to add the " |
|
@markshannon I had to fix some merge conflicts and I have done the cleanup. I am building again with the buildbot fleet but this should be ready for a first version. |
|
If you want to schedule another build, you need to add the " |
The C changes LGTM. Some minor formatting nits at https://github.com/python/cpython/pull/28488/files#r717573235 and below.
Huge disclaimer: I am not a Python-GDB expert! It would be better to have someone else reviewing those changes.
| PyObject *function = PEEK(oparg + 1); | ||
| if (Py_TYPE(function) == &PyFunction_Type) { | ||
| PyCodeObject *code = (PyCodeObject*)PyFunction_GET_CODE(function); | ||
| PyObject *locals = code->co_flags & CO_OPTIMIZED ? NULL : PyFunction_GET_GLOBALS(function); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The code from here onwards has waay more than 80 characters.
|
@Fidget-Spinner Thanks for the review! I am not going to push anything new until the buildbots pass to not restart the long refleak builds yet again. I will fix the formatting issues in a new PR unless @markshannon wants to change something fundamental (I also plan to rename the |
I was just about to mention that. Good call! |
| // *valid* arguments (i.e. the ones that fit into the frame). | ||
| PyCodeObject *co = (PyCodeObject*)con->fc_code; | ||
| const Py_ssize_t total_args = co->co_argcount + co->co_kwonlyargcount; | ||
| for (Py_ssize_t i = 0; i < Py_MIN(argcount, total_args); i++) { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
GHA is warning:
comparison of integer expressions of different signedness: ‘Py_ssize_t’ {aka ‘long int’} and ‘long unsigned int’ [-Wsign-compare
and
'>': signed/unsigned mismatch [D:\a\cpython\cpython\PCbuild\pythoncore.vcxproj]
| for (Py_ssize_t i = 0; i < Py_MIN(argcount, total_args); i++) { | |
| for (size_t i = 0; i < Py_MIN(argcount, (size_t)total_args); i++) { |
|
Cleanups happening here: #28836 @markshannon After that is merged, I will create a PR to address the counter to convert it to a "entry frame" flag. |
|
This is really cool, nice work @pablogsal ;-) |
https://bugs.python.org/issue45256