Mailman 3 March 2022 - Python-Dev

Python multithreading without the GIL
by Sam Gross 31 Jul '23

31 Jul '23

Hi, I've been working on changes to CPython to allow it to run without the global interpreter lock. I'd like to share a working proof-of-concept that can run without the GIL. The proof-of-concept involves substantial changes to CPython internals, but relatively few changes to the C-API. It is compatible with many C extensions: extensions must be rebuilt, but usually require small or no modifications to source code. I've built compatible versions of packages from the scientific Python ecosystem, and they are installable through the bundled "pip". Source code: https://github.com/colesbury/nogil Design overview: https://docs.google.com/document/d/18CXhDb1ygxg-YXNBJNzfzZsDFosB5e6BfnXLlej… My goal with the proof-of-concept is to demonstrate that removing the GIL is feasible and worthwhile, and that the technical ideas of the project could serve as a basis of such an effort. I'd like to start a discussion about these ideas and gauge the community's interest in this approach to removing the GIL. Regards, Sam Gross colesbury(a)gmail.com / sgross(a)fb.com

24 48

Mixed Python/C debugging
by Skip Montanaro 26 Apr '23

26 Apr '23

Having tried comp.lang.python with no response, I turn here... After at least ten years away from Python's run-time interpreter & byte code compiler, I'm getting set to familiarize myself with that again. This will, I think, entail debugging a mixed Python/C environment. I'm an Emacs user and am aware that GDB since 7.0 has support for debugging at the Python code level. Is Emacs+GDB my best bet? Are there any Python IDEs which support C-level breakpoints and debugging? Thanks, Skip

5 6

PEP 448 review
by Guido van Rossum 29 Mar '23

29 Mar '23

I'm back, I've re-read the PEP, and I've re-read the long thread with "(no subject)". I think Georg Brandl nailed it: """ *I like the "sequence and dict flattening" part of the PEP, mostly because itis consistent and should be easy to understand, but the comprehension syntaxenhancements seem to be bad for readability and "comprehending" what the codedoes.The call syntax part is a mixed bag on the one hand it is nice to be consistent with the extended possibilities in literals (flattening), but on the other hand there would be small but annoying inconsistencies anyways (e.g. the duplicate kwarg case above).* """ Greg Ewing followed up explaining that the inconsistency between dict flattening and call syntax is inherent in the pre-existing different rules for dicts vs. keyword args: {'a':1, 'a':2} results in {'a':2}, while f(a=1, a=2) is an error. (This form is a SyntaxError; the dynamic case f(a=1, **{'a': 1}) is a TypeError.) For me, allowing f(*a, *b) and f(**d, **e) and all the other combinations for function calls proposed by the PEP is an easy +1 -- it's a straightforward extension of the existing pattern, and anybody who knows what f(x, *a) does will understand f(x, *a, y, *b). Guessing what f(**d, **e) means shouldn't be hard either. Understanding the edge case for duplicate keys with f(**d, **e) is a little harder, but the error messages are pretty clear, and it is not a new edge case. The sequence and dict flattening syntax proposals are also clean and logical -- we already have *-unpacking on the receiving side, so allowing *x in tuple expressions reads pretty naturally (and the similarity with *a in argument lists certainly helps). From here, having [a, *x, b, *y] is also natural, and then the extension to other displays is natural: {a, *x, b, *y} and {a:1, **d, b:2, **e}. This, too, gets a +1 from me. So that leaves comprehensions. IIRC, during the development of the patch we realized that f(*x for x in xs) is sufficiently ambiguous that we decided to disallow it -- note that f(x for x in xs) is already somewhat of a special case because an argument can only be a "bare" generator expression if it is the only argument. The same reasoning doesn't apply (in that form) to list, set and dict comprehensions -- while f(x for x in xs) is identical in meaning to f((x for x in xs)), [x for x in xs] is NOT the same as [(x for x in xs)] (that's a list of one element, and the element is a generator expression). The basic premise of this part of the proposal is that if you have a few iterables, the new proposal (without comprehensions) lets you create a list or generator expression that iterates over all of them, essentially flattening them: >>> xs = [1, 2, 3] >>> ys = ['abc', 'def'] >>> zs = [99] >>> [*xs, *ys, *zs] [1, 2, 3, 'abc', 'def', 99] >>> But now suppose you have a list of iterables: >>> xss = [[1, 2, 3], ['abc', 'def'], [99]] >>> [*xss[0], *xss[1], *xss[2]] [1, 2, 3, 'abc', 'def', 99] >>> Wouldn't it be nice if you could write the latter using a comprehension? >>> xss = [[1, 2, 3], ['abc', 'def'], [99]] >>> [*xs for xs in xss] [1, 2, 3, 'abc', 'def', 99] >>> This is somewhat seductive, and the following is even nicer: the *xs position may be an expression, e.g.: >>> xss = [[1, 2, 3], ['abc', 'def'], [99]] >>> [*xs[:2] for xs in xss] [1, 2, 'abc', 'def', 99] >>> On the other hand, I had to explore the possibilities here by experimenting in the interpreter, and I discovered some odd edge cases (e.g. you can parenthesize the starred expression, but that seems a syntactic accident). All in all I am personally +0 on the comprehension part of the PEP, and I like that it provides a way to "flatten" a sequence of sequences, but I think very few people in the thread have supported this part. Therefore I would like to ask Neil to update the PEP and the patch to take out the comprehension part, so that the two "easy wins" can make it into Python 3.5 (basically, I am accepting two-thirds of the PEP :-). There is some time yet until alpha 2. I would also like code reviewers (Benjamin?) to start reviewing the patch <http://bugs.python.org/issue2292>, taking into account that the comprehension part needs to be removed. -- --Guido van Rossum (python.org/~guido)

7 17

PEP 638: Syntactic macros
by Mark Shannon 05 Feb '23

05 Feb '23

Hi everyone, I've submitted my PEP on syntactic macros as PEP 638. https://www.python.org/dev/peps/pep-0638/ All comments and suggestions are welcome. Cheers, Mark

16 25

How about using modern C++ in development of CPython ?
by redradist＠gmail.com 19 Oct '22

19 Oct '22

Hi all, What about of using modern C++ in developing CPython ? With C++ we can get the following benefits: 1) Readability - less code, most code is hidden by abstraction without losing performance 2) Maintainability - no code duplication in favor of using reusable classes 3) RAII - Resource Acquisition Is Initialization, predictable allocation and free resources ...

23 51

PEP 505 (None-aware operators) for Python 3.11
by Doug Swarin 21 Sep '22

21 Sep '22

Hello, I've been following PEP 505 (https://www.python.org/dev/peps/pep-0505/) since it was first proposed for Python 3.8. It's been deferred for some time and I'm interested in seeing it in Python 3.11, but I know there were also a number of objections which resulted in it being deferred (including by one of the original authors, Mr. Dower). I did email both Mr. Dower and Mr. Haase and they graciously gave me their permission to bring it up on this list for discussion and hopefully final pronouncement one way or the other. I personally believe that the PEP will result in a significant reduction in boilerplate code, and it is substantially similar to the same operators now found in a number of other languages, especially C# and JavaScript (https://wikipedia.org/wiki/Null_coalescing_operator and https://wikipedia.org/wiki/Safe_navigation_operator). I believe strong and valid arguments can be made about the use of None being a fundamental flaw in some types of coding (and that adding additional support for it to the language will increase the use of None in this way), but I also believe there are many use cases in programming where it is by far the simplest way to express various semantics, and the fact exists that None is already used extensively in large quantities of code, and further that there is already a great deal of code written to constantly test against None and break out of a statement without throwing an error. I also understand the argument that especially the maybe-dot (?.) and maybe-subscript (?[) operators can decrease readability of code and also believe these are valid arguments against it. While I believe the existence and use of these operators in other languages definitely helps the case that these can be used and understood successfully, I think it is entirely valid to either consider other syntax (though I prefer the chosen syntax of PEP 505), or even to reduce PEP 505 to having only the coalesce operator (??) and the maybe-assign operator (??=). Separately, I have implemented a pure-Python solution for PEP505 (which is definitely rather beta) which might help test the waters for a final implementation in CPython (though the CPython implementation would of course be much more efficient). It can be found at https://pypi.org/project/pep505/ Thanks, Doug

22 53

PEP 563 and Python 3.10.
by Thomas Wouters 22 Jul '22

22 Jul '22

(Starting a new thread so as not to derail any of the ongoing discussions.) Thanks, everyone, for your thoughts on Python 3.10 and the impact of PEP 563 (postponed evaluation of annotations) becoming the default. The Steering Council has considered the issue carefully, along with many of the proposed alternatives and solutions, and we’ve decided that at this point, we simply can’t risk the compatibility breakage of PEP 563. We need to roll back the change that made stringified annotations the default, at least for 3.10. (Pablo is already working on this.) To be clear, we are not reverting PEP 563 itself. The future import will keep working like it did since Python 3.7. We’re delaying making PEP 563 string-based annotations the default until Python 3.11. This will give us time to find a solution that works for everyone (or to find a feasible upgrade path for users who currently rely on evaluated annotations). Some considerations that led us to this decision: - PEP 563’s default change is clearly too disruptive to downstream users and third-party libraries to happen right now. We can’t risk breaking even a small subset of the FastAPI/pydantic users, not to mention other uses of evaluated type annotations that we’re not aware of yet. - PEP 563 provides no warning to users of the feature it’s disabling. Without that, we can’t expect users to be aware of the upcoming breakage. The lack of a warning was by design, and made sense in a world where type annotations were only consumed by static type checkers --- but that’s not actually the situation we’re in. There are clearly existing real-world, run-time uses of type annotations that would be adversely affected by this change. - Originally, PEP 563 was scheduled to take effect in Python 4, and this changed recently (after the discussion in the Language Summit of 2020). It's possible that third-party libraries and users didn’t plan to react in the current time frame as they were not aware of this change in timing. - There isn’t enough time to properly discuss PEP 649 or any of the alternatives before the beta 1 deadline, and we really need to make sure we don’t compound errors here. We need to look for a long term solution, which isn’t possible while still maintaining the release deadlines of Python 3.10. That means we’re also deferring PEP 649 to Python 3.11. In the Steering Council’s unanimous opinion, rolling back the default flip for stringified annotations in Python 3.10 is the least disruptive of all the options. We need to continue discussing the issue and potential solutions, since this merely postpones the problem until 3.11. (For the record, postponing the change further is not off the table, either, for example if the final decision is to treat evaluated annotations as a deprecated feature, with warnings on use.) For what it’s worth, the SC is also considering what we can do to reduce the odds of something like this happening again, but that’s a separate consideration, and a multi-faceted one at that. For the Steering Council, Thomas. -- Thomas Wouters <thomas(a)python.org> Hi! I'm an email virus! Think twice before sending your email to help me spread!

6 5

compiled python3.10 is unable to find _ssl
by Sandeep Gupta 29 May '22

29 May '22

I am having compilation issues again with python3.10 with ssl . The ./configure was invoked with ssl options and ssl modules seems to be build successfully. """ The following modules found by detect_modules() in setup.py, have been built by the Makefile instead, as configured by the Setup files: _abc _hashlib _ssl pwd time """ However, when I do import ssl from python, I get the error: Traceback (most recent call last): File "<stdin>", line 1, in <module> File "/home/shared/Builds/Python-3.10.0/lib/python3.10/ssl.py", line 98, in <module> import _ssl # if we can't import it, let the error propagate ModuleNotFoundError: No module named '_ssl' I can't find _ssl.so in build or install directory. Thanks -S

6 12

C API: Move PEP 523 "Adding a frame evaluation API to CPython" private C API to the internal C API
by Victor Stinner 25 Apr '22

25 Apr '22

Hi, I proposed two PRs to move the private C API (Include/cpython/) of PEP 523 "Adding a frame evaluation API to CPython" to the internal C API (Include/internals/): * https://github.com/python/cpython/pull/32052 * https://github.com/python/cpython/pull/32054 API: * _PyFrameEvalFunction type * _PyInterpreterState_GetEvalFrameFunc() * _PyInterpreterState_SetEvalFrameFunc() * (undocumented) _PyEval_EvalFrameDefault() The private API to get/set the eval function *is* documented at: https://docs.python.org/dev/c-api/init.html#c._PyInterpreterState_GetEvalFr… I added the Get/Set functions so debuggers don't have to access directly to the PyInterpreterState structure which has been moved to the internal C API in Python 3.8. This API causes me multiple issues: * It's a private API and I'm trying to remove the private API from the public C API header files. * The _PyFrameEvalFunction type is not stable: it got a new "tstate" parameter in Python 3.9 and the type of the second parameter changed from PyFrameObject* to _PyInterpreterFrame* in Python 3.11. * These functions use the _PyInterpreterFrame type which is part of the internal C API. While Pyston didn't bring a JIT compiler to Python with PEP 523, debuggers were made faster by using this API. Debuggers like pydevd, debugpy and ptvsd use it. I propose to move theses API to the internal header files (Include/internals/) to clarify that it's not part of the public C API and that there is no backward compatibility warranty. The change is being discussed at: https://bugs.python.org/issue46850 -- PEP 523 API added more private functions for code objects: * _PyEval_RequestCodeExtraIndex() * _PyCode_GetExtra() * _PyCode_SetExtra() The _PyEval_RequestCodeExtraIndex() function seems to be used by the pydevd debugger. The two others seem to be unused in the wild. I'm not sure if these ones should be moved to the internal C API. They can be left unchanged, since they don't use a type only defined by the internal C API. Victor -- Night gathers, and now my watch begins. It shall not end until my death.

7 20

Re: C API: Move PEP 523 "Adding a frame evaluation API to CPython" private C API to the internal C API
by Jason Ansel 20 Apr '22

20 Apr '22

The PyTorch team plans to use PEP 523 as a part of PyTorch 2.0, so this proposal may break the next major release of PyTorch. The related project is TorchDynamo, which can be found here: https://github.com/facebookresearch/torchdynamo We will likely move this into the core of PyTorch closer to release. If the changed happens, would PyTorch still be able to use the eval frame API? Or would it prevent from being used entirely?

11 25