internals-and-peps

[rose schooner]

oh

that should NOT lead to _requires_builtin

[willow pewter]

though there are 3 following functions with @_require_builtin and all of them returns None

[rose schooner]

replace ```c
if (PyImport_ImportFrozenModule("_frozen_importlib") <= 0) {
return -1;
}

PyObject *importlib = PyImport_AddModuleRef("_frozen_importlib");
if (importlib == NULL) {
    return -1;
}

withc
if (PyImport_ImportModule("importlib") <= 0) {
return -1;
}

PyObject *importlib = PyImport_AddModuleRef("importlib");
if (importlib == NULL) {
    return -1;
}

and see what you get

[willow pewter]

ok

[willow pewter]

so we are directly using importlib instead of frozen importlib here?

[rose schooner]

okayy i guess not

[rose schooner]

yup

but i guess the frozen modules are there for a reason

yeah set it back

we need the frozen imports

[willow pewter]

though does it compile fine for you?

[rose schooner]

yup

the problem for you is that _requires_builtin is getting a string for some reason

[willow pewter]

should I just try to reclone the repo?

[rose schooner]

sure

see if it works

[willow pewter]

how can I directly clone https://github.com/python/cpython/commit/60403a5409ff2c3f3b07dd2ca91a7a3e096839c7 this instead of cloning from main

[rose schooner]

there's https://github.com/python/cpython/releases/tag/v3.13.0

since i don't know how to use git

[willow pewter]

same but it would be easier to switch commits and branch with git

[rose schooner]

i was gonna suggest adding print(old, new) before line 45 in Lib/importlib/_bootstrap.py but you try doing that first i guess

[willow pewter]

cloning....

for some freaking reason visual studio don't want to fetch externals before building

had to use cli to do it

[willow pewter]

did, debug compiled fine, release said can't open input file pcbuild/amd64/python313.lib

ran clean.bat

recompiled

[rose schooner]

uh oh.

python313.lib is supposed to be built with the compilation

[willow pewter]

also just for note

which I think might be important

I am looking at source code because I (and @stable grail) found a bug that creating venv on 3.13 on windows tkinter imports look for tcl in libs/tcl/ instead of tcl/ and some more werid paths and according to @stable grail possibly it could have affacted more things than tkinter

[willow pewter]

what should I do now?

what the python

I executed build.bat instead of visual studio build button

and guess what

it freaking worked

how the hell

[rose schooner]

oh

[willow pewter]

and tkinter works

@stable grail problem is with build system

how do you build installer for python

[willow pewter]

https://basedbin.fly.dev/p/VBRGsj.txt

I already have v143 build tools

[jade raven]

Which is what the documentation says to do

[willow pewter]

it worked fine in 3.13.0a0 though

[willow pewter]

what should I do?

[willow pewter]

ran command in visual studio and it worked

somehow

what is immortal.c

[willow pewter]

never mind it didn't work

I forgot skip build and docs flag and it was generating them so I thought it worked but it didn't

[jade raven]

but that doesn't seem to be the problem

did you restart your terminal after installing build tools?

[willow pewter]

yup

let me try to reboot

[jade raven]

wins key > developer command prompt for VS 2022

then navigate the poaths and try again

[willow pewter]

doesn't work

[willow pewter]

as I said I have tried it in visual studio too

[jade raven]

uninstall whatever it is you have installed and try this, #1215299553914519572 message, alternatively try the second suggestion liste there

[willow pewter]

that's what I have?

[jade raven]

then try the second suggestion there

[willow pewter]

there is just visual studio?

[jade raven]

read the bottom error where it says retarget solution

[willow pewter]

done that

doesn't work

[jade raven]

okay, what exactly is it that you're doing and why?

[willow pewter]

trying to build cpython x64 3.13.0 msi

[jade raven]

why?

[willow pewter]

#internals-and-peps message

i compiled 3.13.0 and tried creating venv with it and its fine so problem must be with installation or build process

problem exists in all(rc, stable, alpha, beta) 3.13.0 windows builds

[jade raven]

let me download the windows build, how do i reproduce your error?

[unkempt rock]

From selenium.webdriver.support.ui import expected_conditions as EC
import error: cannot import name accepted under_conditions from 'selenium.WebDriver.support.ui'

What should I do?

[jade raven]

don't crosspost. i answered you in #python-discussion

[willow pewter]

also it doesn't even belong to this channel

[willow pewter]

wait

[willow pewter]

clone this 60403a5 on from 3.13 branch
install visual studio 2022 community with these components:
python native development tools
.net framework 3.5 development tools
(C++ are mentioned in image)

run Tools/msi/buildrelease.bat -x64

[jade raven]

okay but aren't you saying it's reproducible in the official builds?

[willow pewter]

can be

I can't test on vm because it will take ages, I don't have a fast one

[willow pewter]

but this venv issue is 100% reproducible you can try by creating a new venv with 3.13.0 release**(installed from windows installer from python.org)** and run python -m tkinter in venv

I was thinking about finding the issue in cpython about this and create a PR or issue with possible fix

[jade raven]

i thought your error was going to be related to this one, https://github.com/python/cpython/issues/118973, but doesn't look like it, let me download 3.13.0 from python.org and check it out

[willow pewter]

not at all

[jade raven]

i couldn't repro it

i think your problem might be with something in your $PATH

[willow pewter]

what does py -0p says

did you use venv?

[jade raven]

ah no, one second

indeed reproducible

[willow pewter]

also if you compile cpython the issue is not there

so issue must be with installer or build process

thats why I was trying to build installer which I just can't for some reason

[jade raven]

my best advice i would say is to use windows sandbox so that you have a clean enviroment with no messy path shenanigans or a messy enviroment at all

[willow pewter]

I have tried using offical installer in fresh install of windows 10 vm

I can try in sandbox too

let me try copy pasting the amd64 directory in there

[jade raven]

i mean to build the installer

[willow pewter]

ohh

well at the end its a vm

will try it tomorrow

its 00:25 here

adios

[willow pewter]

tried on fresh windows 10 vm

[stable grail]

you need build tools correctly setup for them to work

[willow pewter]

well I already have v143 build tools installed

also 3.5 .net

[stable grail]

you need your shell to know about the build tools

not the installer

is it added to your path

and is the correct one, the only one in your path?

[willow pewter]

I only have the latest(v143) installed

[willow pewter]

let me see

[willow pewter]

https://basedbin.fly.dev/p/VG9I6o.txt looks like it

[willow pewter]

also

[stable grail]

run the build process with full paths

[willow pewter]

same

[stable grail]

are you following the devguide while doing this?

[willow pewter]

uh well this is what it says in getting started

and I couldn't find much reference about msi

shit I missed this part

this is in tools/msi/readme.txt

[stable grail]

yeah,,, very few does windows building

i had windows in 2018 during pycon, and at that time, there where no working docs for building on windows.

[willow pewter]

what

[stable grail]

the problem with building on windows is that there are so many variables to manage, so writing documentation is v ery miss

[willow pewter]

development of such big codebases on windows is shit

[stable grail]

so what you need from the build tools is... get the 100% correct options. probably only 10 mb of data

[willow pewter]

on linux its much easier with libs available in package managaer, make and gcc

[stable grail]

but what I do, is get them all, download 20GB of build tools, and then, it will work

but nice catch about the ARM build options. that might be the one you need for it all to work without the full install

[willow pewter]

also for some reason microsoft wants to force visual studio just to compile

while its not a even a good editor

[willow pewter]

we may even get a time where devs stop building for windows

they just give source code and say compile yourself

[dusk comet]

just use WSL on windows

[willow pewter]

it requires windows to build for windows

you can't build for windows from linux

[willow pewter]

I tried it and it still doesn't build

also tried installing wix even though ti fetched that automatically but still doesn't work

[unkempt rock]

hi

[rose schooner]

found this

the stuff with universal v143 on it

[willow pewter]

I will try tonight thanks

[willow pewter]

hello

[graceful jungle]

Is Flask the most useful tool for web app development with python? Or is there a better option?

[winged sphinx]

Ask over in #python-discussion plz

[graceful jungle]

Gotcha, thanks

[willow pewter]

looks like its compiling now

thanks

[willow pewter]

@stable grail problem is in installer

the compiled version works fine

but when I built the installer and installed in vm

it doesn't work

[idle ravine]

I think you can download the build tools separately from Microsoft, if it wasn't removed again

https://visualstudio.microsoft.com/ru/downloads/#build-tools-for-visual-studio-2022

[willow pewter]

got fixed in https://github.com/python/cpython/commit/66b8cb114251c3454857bf35355c9365302f4b69

[halcyon trail]

is there a reason why chain.from_iterable is so... weirdly named?

it's the only thing in all of itertools afaics that's nested below something else

I feel like if it was just called flatten instead it would be used more often. both for being much shorter, and for being more clear.

[near aurora]

If I've got an idea for an alternative to a proposed PEP (and the alternative is almost the length of a PEP itself) would it be better to start a new thread on dpo (and reference it in the thread for the proposed PEP) or just post a very long reply?

[neat delta]

from_iterable is an alternate constructor, and most such things are class methods.

while flatten may be more descriptive to one of its common uses, it does have the issue of being slightly inaccurate, as chain.from_iterable only removes a single layer. I would strongly argue that something named flatten should produce a result that is fully flat, with partial flattening being done only at explicit user request, possibly via arguments

[faint river]

Like maybe with a kwarg for how many levels

Default being -1 to signify infinite levels

[feral island]

what if you pass a str

[halcyon trail]

I mean not really, "flatten" isn't "flatten_all" or something like this

flatten already exists and does this exact thing in multiple languages

Whereas I've never seen chain.from_iterable or chain_from_iterable much of anywhere but python

[halcyon trail]

nice

[faint river]

I believe rust has struct from struct type naming for alternate constructors

[halcyon trail]

Lots of languages have that approach in general

I meant this particular name for this particular concept

[wanton flame]

Start a new thread, and drop a link to it on the old thread

[thorn flume]

quick question -
looking at dictobject.c, it seems to be doing this to find the index of objects that collide

        perturb >>= PERTURB_SHIFT;
        i = mask & (i*5 + perturb + 1);```
why does it shift-right `perturb` and add it to `i`?

[rose schooner]

read from this line https://github.com/python/cpython/blob/main/Objects/dictobject.c#L340

[fallen slate]

Objects/dictobject.c line 340

The other half of the strategy is to get the other bits of the hash code```

[rose schooner]

and the other lines below

[thorn flume]

i cannot read owo

[thorn flume]

https://github.com/python/cpython/blob/ce740d46246b28bb675ba9d62214b59be9b8411e/Objects/dictobject.c#L994-1036

why does it do the for loop twice? i know it says manual loop unrolling, but why

[fallen slate]

Objects/dictobject.c line 994

static inline Py_ALWAYS_INLINE Py_ssize_t```

[rose schooner]

for performance

which is by manual loop unrolling

probably something related to it happening 2 times per iteration

[thorn flume]

well no shit

[rose schooner]

i dug through the blame and there's really no particular reason given

except "it's a hotspot and that probably improved the performance"

[spark magnet]

why doesn't that count as a reason?

[rose schooner]

it does

it's just the only reason i found

[spark magnet]

i misunderstood "except"

[jade raven]

what's the best place to request a pathlib property that returns the full path as a string? doing str(Pathobj) isn't pythonic imo

[flat gazelle]

str is the correct operation for that, there's no alternative I've been able to find.

[jade raven]

right, which is why i'm wanting to ask for a property method to be implemented

[flat gazelle]

oh, I misread

[halcyon trail]

Why isn't it pythonic though

[faint river]

I find that quite pythonic

[jade raven]

because pathlib already provides multiple methods / properties which return things as a str object

!d pathlib.PurePath.name for example

[fallen slate]

pathlib.PurePath.name

A string representing the final path component, excluding the drive and root, if any:

>>> PurePosixPath('my/library/setup.py').name
'setup.py'
```  UNC drive names are not considered:

```py
>>> PureWindowsPath('//some/share/setup.py').name
'setup.py'
>>> PureWindowsPath('//some/share').name
''

[flat gazelle]

It is kind of odd that you have a ton of properties that give you parts of the path, but no property that gives you the whole path

[faint river]

what do you suggest it would be named?

[jade raven]

strpath, fullpath idk_naming_is_hard

[faint river]

path.strpath wow, so pythonic

[flat gazelle]

bikeshedding the name is kind of missing the point

[rose schooner]

.fullpath then

[jade raven]

i think i'll open a DPO thread on this

[flat gazelle]

yea, I don't hate .fullpath, tho that should probably be str(path.resolve())

[jade raven]

but then we're back to using str, which is what i want to avoid

[flat gazelle]

yea, my point is that fullpath makes no sense as a replacement for str(path), but it'd be useful as a separate operation.

[jade raven]

going to use this

[faint river]

str(path) not necessarily equals str(path.resolve()), due to relative paths

[flat gazelle]

In [16]: str(Path('a/b'))
Out[16]: 'a\\b'

[faint river]

and fullpath sounds like it would be the latter

(is what I think lakmatiol is getting at)

[flat gazelle]

it is a bit of an odd one to think about, since there's no name for the string representing a path other than a path, which makes the operation a type conversion and thus it should be done via str

[faint river]

path == Path(str(path))

[jade raven]

https://discuss.python.org/t/path-property-that-returns-the-full-path-as-a-string/67787/1

[twilit lotus]

Is there anyone having good resources to learn python decorator?

[dusty oak]

I will advice this book. it is the most rich on Python specific sugarness, its gotchas, how to use and best practices

she dedicated a good chapter to them there

and all within reasonable amount of text too, not too much watery/overloaded

the entire book is just 311 pages

[twilit lotus]

Thank you for the help, l’ll check this book! I am struggling with decorator

[trail cloud]

Is it possible to create chatbots with python and pythons framework ?

[crisp locust]

Discord is full of them

[trail cloud]

Web site chatbots

[cursive field]

Guys can U help me learn how to make neural networks

I can only make chatbots and need help

Ping me if U wanna contact

I'm at school btw 😭😭

[crisp locust]

Ask this in the #data-science-and-ml channel

[crisp locust]

Ask this in #discord-bots channel

[loud jungle]

Yeah? The difference comes down to interfacing with users on your website instead of Discord

#web-development may be the most relevant channel to ask questions about this

[trail cloud]

Got it

[worthy sandal]

Is anyone using structlog? Is it better than the traditional logging library?

Please let me know

[swift imp]

I think the real benefit is it's more machine parsable?

[south kayak]

how do I float small suggestions for the stdlib? Ex: making the name of the positionals and optionals groups in argparse modifiable with a simple reassign

Right now I have to modify the internal value after creating the parser
Code (note: I've done a few other hacks to make mine look tidy, but those are much larger changes that I'm not looking to suggest right now):

parser = WarMACParser(
    usage="warmac <command> [options]",
    ... # more stuff here
)
parser._positionals.title = "commands"  # changing positional header
```This code gives this output (notice red box)

how it normally looks (using pre-commit as an example. This how mine looked before I changed a few things):

[feral island]

Open an issue on CPython, explaining clearly what you want to add

[south kayak]

alright!

[rose schooner]

how would i check if an object is only referenced by the container it's in?

like ```py
class A:
def init(self, x):
self.x = x

a = A(19313)

check if a.x is only referenced by A

gc.get_referrers(a.x) == [a]

x = {4821: 2}

check if 4821 is only referenced by x

which seems to be impossible to check with gc.get_referrers() since it's not directly referenced by the dictionary apparently?

[rose schooner]

other than using gc.get_referrers()

[spark magnet]

What do you want to do based on the outcome? Perhaps this is a time for weakrefs

[rose schooner]

weakrefs don't work on lists

[spark magnet]

wrap them in a proxy class

[rose schooner]

soo how would i detect if the object inside can be garbage collected?

i.e. the only referrer left is the proxy

[spark magnet]

the container would hold a weakref to the proxy, and the proxy would be reclaimed if the container were the only referrer.

[rose schooner]

that sounds a lot like it relies on the proxy being passed around instead of the object but okay

[spark magnet]

yes, you would need to do that.

[spark verge]

Do you expect this to loop forever or halt? And given your choice do you think it's a feature of the language implementation or that it's unspecified?

import gc
import weakref

class A: pass

def main(a):
    lst = []
    a = weakref.ref(a, lambda _: lst.append(True))
    while not lst: gc.collect()

main(A())

[spark verge]

It's tricky because gc.get_referrers can return an empty list when an object is reachable

Eg in the above case ^

[spark verge]

You can use sys.getrefcount but it changes if you're in a generator/coroutine/both

[feral island]

it's unspecified when unreachable objects are collected

[spark verge]

Right but they have to be collected eventually

Implementations aren't allowed to defer collection of unreachable objects forever

in fact when implementations self test they have to fix this function:
https://github.com/python/cpython/blob/ed24702bd0f9925908ce48584c31dfad732208b2/Lib/test/support/__init__.py#L799-L813

[fallen slate]

Lib/test/support/__init__.py lines 799 to 813

def gc_collect():
    """Force as many objects as possible to be collected.

    In non-CPython implementations of Python, this is needed because timely
    deallocation is not guaranteed by the garbage collector.  (Even in CPython
    this can be the case in case of reference cycles.)  This means that __del__
    methods may be called later than expected and weakrefs may remain alive for
    longer than expected.  This function tries its best to force all garbage
    objects to disappear.
    """
    import gc
    gc.collect()
    gc.collect()
    gc.collect()```

[spark verge]

so the demo could be written like this:

import gc
import weakref
import test.support

class A: pass

def main(a):
    a = weakref.ref(a)
    test.support.gc_collect()
    assert a() is None

main(A())

implementations are required to replace test.support.gc_collect with something that will collect all unreachable objects

otherwise they didn't "try their best"

[spark verge]

It's not really possible to do with small integers

[raven ridge]

The answer for small integers is just "no". They're always referenced by the small integer cache, in addition to whatever else is referencing them

[rose schooner]

okay but what if i want to use weakrefs to unmodified list objects

pretty much the only thing i'd need is a weakref.ref() with a callback and works for any object ||(except interpreter-cached values, semantically)||

in other words, something to assign a callback for when the object is finalizing

[rose schooner]

just used sys.getrefcount() i guess

seems to work pretty well

[narrow patio]

can anybody tell me where in cpython the built-in function type is implemented?

like if you look at Python/bltinmodule.c you see a lot of implementations of those functions, but not type

[rose schooner]

Objects/typeobject.c

[lucid oasis]

was wondering what channel to use to ask for advice

[thin fossil]

Holy shit guys wtf is this JIT

I've had this simple prime number benchmark that I do for anything python related. My baseline time is 20s on Cpython

Pypy does it in < 1s

This cpython 3.13 with JIT took 40s

Although without JIT took 17 seconds, they've been doing some nice interpreter optimizations

Graal Python can do it in < 1s

If the JIT isn't ready why even release it...

[jade raven]

because it takes time to build out stuff, 3.13 lays out the groundwork for future improvements.

[thin fossil]

I'm betting 5 years to get to where pypy is right now lol

[jade raven]

a new python is released roughly every year, you might be surprised by what happens between now and then 😛

plus keep in mind the work on python's JIT is pretty recent, it's still in its infancy, so to speak, let it learn how to walk

pypy has been around since 2008 roughly? it's had a lot of time to improve their own thing.

[raven ridge]

What does "release it" mean here? The code exists because it's being actively iterated upon and developed, but it's turned off by default when building the interpreter, and is turned off in every distro I've heard of, and it's described as both "basic" and "experimental" in the release notes.

You literally need to pass a flag with "experimental" in the name just to build an interpreter with the JIT

[gilded flare]

it's like playtesting games

it's not a full release and is only available to the ones who wanna test it out

if you don't want to or it's too disappointing, then use the normal full release

but there's others who wanna see for themselves how it works too

[final geode]

Mind sharing the benchmark?

[spark magnet]

So that interested early adopters will try it and provide feedback, like you will be doing when you give your benchmark to @final geode!

[uneven raptor]

are you sure you built a performance build of cpython?

[thin fossil]

Yes

[thin fossil]

It's nothing special, just something that tests integer math and tight inner loops: https://gist.github.com/veegee/8860afdf289f2a5b5bae9b65f4e26368

I reimplemented that in various languages to get a very very rough idea of efficiency

pypy is one of the greatest open source projects this decade

[thin fossil]

"Release" means make a big stink about it everywhere and have very little to show for it.

[final geode]

Thanks, this is helpful. I’ll dig into it this week. Want me to ping you when I learn more?

[thin fossil]

Of course!

And before anyone screams how benchmarks are worse than lies and damn lies, I KNOW.

But they're still very important. They tell you something and I find that what they tell you (if you know what you're doing and know how things work) is extremely important.

This particular one is showing me how fast the JIT can optimize tight inner loops and integer math, which is the easiest thing to optimise.

[final geode]

Also, while I have you: are your results on main or 3.13.0? And your hardware/os?

[thin fossil]

Installed via:

PYTHON_CONFIGURE_OPTS='--disable-gil --enable-experimental-jit --enable-optimizations --with-lto' PYTHON_CFLAGS='-march=native -mtune=native' pyenv install 3.13.0

Hardware:

Architecture:             x86_64
  CPU op-mode(s):         32-bit, 64-bit
  Address sizes:          46 bits physical, 48 bits virtual
  Byte Order:             Little Endian
CPU(s):                   56
  On-line CPU(s) list:    0-55
Vendor ID:                GenuineIntel
  Model name:             Intel(R) Xeon(R) CPU E5-2697 v3 @ 2.60GHz
    CPU family:           6
    Model:                63
    Thread(s) per core:   2
    Core(s) per socket:   14
    Socket(s):            2
    Stepping:             2
    CPU(s) scaling MHz:   38%
    CPU max MHz:          3600.0000
    CPU min MHz:          1200.0000```

[uneven raptor]

nogil is probably the bottleneck, not the JIT

[final geode]

Oh, I totally missed that in the flags. Yeah, disabling the GIL results in huge slowdowns right now. Try without that.

The JIT isn’t even enabled when the GIL is disabled.

[thin fossil]

I'll try again without that

pyenv must not be using the -j flag for make lmao. I have 56 threads on this dual socket system and it's taking forever to build.

Or it might be the linker :\

[final geode]

I'm building too, it's using -j 8 for me, it's the two LTO link steps that take forever (and can't be parallelized).

[thin fossil]

Yeah LTO is the issue

Ok just ran the test again, (verified JIT is enabled, and did not disable GIL):

....
[0] Found 664579 prime numbers
[1] Found 664579 prime numbers
[2] Found 664579 prime numbers
[3] Found 664579 prime numbers
[4] Found 664579 prime numbers
[5] Found 664579 prime numbers
[6] Found 664579 prime numbers
[7] Found 664579 prime numbers
[8] Found 664579 prime numbers
[9] Found 664579 prime numbers
Total time: 15.577s```

[loud jungle]

JIT enabled performs faster than JIT disabled on your machine

[thin fossil]

@final geode so there's a very slight improvement vs 3.12 (which does 19s on the test)

[final geode]

I mean, 20% sure.

[thin fossil]

pypy:

➜ pypy3 bench3.py
....
[0] Found 664579 prime numbers
[1] Found 664579 prime numbers
[2] Found 664579 prime numbers
[3] Found 664579 prime numbers
[4] Found 664579 prime numbers
[5] Found 664579 prime numbers
[6] Found 664579 prime numbers
[7] Found 664579 prime numbers
[8] Found 664579 prime numbers
[9] Found 664579 prime numbers
Total time: 1.326s```

LMAO 1.326s

That's nuckin futz

The JVM implementation (using int primitives, not boxed Integer, with custom array implementation) does it in 1.173s

That's how absurdly impressive pypy is

Yes it does have some disadvantages in very complex logic that is hard to JIT, but compared to the hotspot VM which probably has $billions of funding, I'm consistently blown away by how good pypy is (and on that note, postgresql)

On another note, the only reason I have

#res = list(filter(lambda x: x or False, s))```
in there is to try both methods of building the list. It used to be that many years ago (like 10 years ago when I first wrote this), generators and generator expressions were not optimized in pypy. But that quickly changed.

But for 3.13.0, with the list comprehsion used instead of the generator expression, **the time further decreases from 15.577s to 13.051s**

[thin fossil]

So yeah, you're definitely right that nogil was the issue initially. And maybe the above little tidbit about list comprehensions vs generator expressions may help the devs out. I wish I had time to contribute to the JIT myself, but there's no way in hell these days 😦

[final geode]

Yeah, the comprehension is much easier to JIT than the functional version.

[thin fossil]

It's definitely a step in the right direction. I still think it'll take years before we get to anywhere near what pypy can do. But as long as it's not a performance regression, it's all good. It might be at the moment, so I won't be using it until they turn it on by default.

[final geode]

It's not our goal to compete with PyPy's performance.

[thin fossil]

There are not many JITs in the world that can compete with pypy performance lol, so fair enough. Jokes aside, I hear what you're saying and I agree. Making everything a bit faster without introducing runtime performance regressions would be huge.

Thank you for the work that you're doing and the time you're contributing for the betterment of humankind ❤️

[final geode]

Thanks.

Appreciate you trying it out and sharing your experience.

[thin fossil]

Ain't no thang, I'm going to be watching it closely. I use python a lot. It's my go-to for prototyping things.

[glass mulch]

Using sys.addaudithook(print) works fine everywhere except PyREPL on Windows, because PyREPL spams time.sleep(0.01) events due to a lack of polling/select in that platform :/

And sys.addaudithook(None) breaks all kinds of REPLs in many versions and OSes
https://github.com/python/cpython/issues/125852

[short fiber]

sorry for pulling back an old msg -- does this fix https://github.com/python/cpython/issues/118973 ? having trouble building the installer locally, i can't replicate your build on my box (definitely a VS noob issue, lol)

[willow pewter]

no I managed to compile python and installer also the issue was in path prefix which is now fixed

[short fiber]

ah, gotcha. but that only fixes importing tkinter in a venv?

[unkempt rock]

who knows pyqt5

[crimson drum]

It's my goal 🙂

[rose schooner]

oh it's mark

[unkempt rock]

https://paste.pythondiscord.com/ZSIA

who can fix my bot code

[safe basalt]

See #❓｜how-to-get-help

[supple beacon]

hey

i dont have persmission to talk in rooms

!!

[glass mulch]

There's an assert that is very easy to trigger (in debug builds, obviously) with sys.audit:

Python 3.14.0a1+ (heads/main:c5b99f5c2c, Oct 26 2024, 12:35:53) [GCC 13.2.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import sys
>>> sys.audit(9)
python: ./Python/sysmodule.c:522: sys_audit: Assertion `PyUnicode_Check(args[0])' failed.
Aborted (core dumped)

Passing a non-string there results in a clean error in non-debug builds.

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: expected str for argument 'event', not int

Is this a common way of doing checks in debug builds? IOW, is this OK or should it avoid aborting for a simple TypeError?

[feral island]

this seems like a bug

[merry bramble]

Yeah, I don't think that's supposed to happen

[crimson drum]

That's definitely a bug. Could you make a CPython issue? Thanks

[gilded flare]

1 core dev was probably enough

[merry bramble]

Any message including the phrase "Aborted (core dumped)" is a core dev magnet

[glass mulch]

https://github.com/python/cpython/issues/126018

[spark verge]

How does this test run? It's not in a class https://github.com/python/cpython/pull/126042/files#diff-1cab14f7e86d7fddf1d9a2cb12c3f5ef50844cf932eb67a97f7fb6acff8330c2R570

[merry bramble]

you can see that the same PR made changes to Lib/test/test_audit.py which execute that function from a unittest test method

I guess the setup for the audit tests is a bit baroque

[feral island]

I didn't check but I assume it runs the test in a fresh process so it can test the audit system in isolation

[merry bramble]

correct

[spark verge]

Ah could do with comments why it doesn't use assertRaises

[grave jolt]

that embed 💀

[wanton flame]

it's Miss Islington: https://www.youtube.com/watch?v=iGx1hiSJbCo

[grave jolt]

yeah I know

it's just funny with github using an avatar in the embed

[heavy egret]

hallo

this is genral chat?

My english is very bad

[astral gazelle]

no #python-discussion

[white nexus]

!e () = []

someone recently brought this to my attention and its wild that it doesn't break

[fallen slate]

:warning: Your 3.13 free threaded eval job has completed with return code 0.

[No output]

[faint river]

makes sense

[unkempt rock]

hey how could i remove a ban

[spark verge]

That's off topic for this channel

[little token]

Hi

[glass mulch]

Is a segfault in _tkinter interesting, or is it a "here be dragons" module?

I took vstinner's old fuzzer, fusil, for a ride and it stumbled upon:

import  _tkinter 
_tkinter.create(None, '', '\U0010FFFF', None)

[spark verge]

Looks like you can call it via https://github.com/python/cpython/blob/3.13/Lib/tkinter/__init__.py#L2459

[fallen slate]

Lib/tkinter/__init__.py line 2459

self.tk = _tkinter.create(screenName, baseName, className, interactive, wantobjects, useTk, sync, use)```

[feral island]

please report a bug

[spark verge]

If you can call it via tkinter.Tk() even better

...or worse depending on your perspective

!e

import tkinter
tkinter.Tk(screenName=None, baseName='', className='\U0010FFFF')

[fallen slate]

:x: Your 3.12 eval job has completed with return code 1.

001 | Traceback (most recent call last):
002 |   File "/home/main.py", line 1, in <module>
003 |     import tkinter
004 |   File "/snekbin/python/3.12/lib/python3.12/tkinter/__init__.py", line 38, in <module>
005 |     import _tkinter # If this fails your Python may not be configured for Tk
006 |     ^^^^^^^^^^^^^^^
007 | ImportError: libtk8.6.so: cannot open shared object file: No such file or directory

[spark verge]

Ah it's not installed

[feral island]

Python 3.14.0a0 experimental free-threading build (heads/gh-125331:c315120a535, Oct 30 2024, 11:22:31) [Clang 15.0.0 (clang-1500.3.9.4)] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> import tkinter
... tkinter.Tk(screenName=None, baseName='', className='\U0010FFFF')
... 
zsh: segmentation fault  ./python.exe

[spark verge]

Nice

[spark verge]

Please raise a ticket and post the repro with tkinter.Tk also

[glass mulch]

I will, as soon as I get back from an errand. It found another in the meantime:

import sre_constants
sre_constants._makecodes("", {}, 10)

[spark verge]

Hmm that's deprecated, can you get to that via re?

[spark verge]

Could be troubling actually

Can you reach it via re._constants _NamedIntConstant?

!e

from re._constants import _NamedIntConstant

def _makecodes(*names):
    items = [_NamedIntConstant(i, name) for i, name in enumerate(names)]
    return items
_makecodes("", {}, 10)

[fallen slate]

:warning: Your 3.13 free threaded eval job has completed with return code 0.

[No output]

[spark verge]

!e

import sre_constants
sre_constants._makecodes("", {}, 10)

[spark verge]

Hmm can't reproduce

What python version are you on? It's a worrying one because it all looks like pure python

[fallen slate]

:x: Your 3.13 free threaded eval job has completed with return code 1.

001 | /home/main.py:1: DeprecationWarning: module 'sre_constants' is deprecated
002 |   import sre_constants
003 | Traceback (most recent call last):
004 |   File "/home/main.py", line 2, in <module>
005 |     sre_constants._makecodes("", {}, 10)
006 |     ~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^
007 |   File "/snekbin/python/3.13t/lib/python3.13t/re/_constants.py", line 74, in _makecodes
008 |     globals().update({item.name: item for item in items})
009 |                       ^^^^^^^^^^^^^^^
010 | TypeError: unhashable type: 'dict'

[feral island]

I also get the unhashable TypeError for that code, tried 3.12 and main

[glass mulch]

main

[spark verge]

Commit hash?

[glass mulch]

Wait, I'm running a python from october 8th 😦

Python 3.14.0a0 (heads/main:31516c98dd7, Oct  8 2024, 17:52:24) [GCC 11.4.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import sre_constants
... sre_constants._makecodes("", {}, 10)
...
<python-input-0>:1: DeprecationWarning: module 'sre_constants' is deprecated
  import sre_constants
Segmentation fault

WIll re-run all cases on real main and report back later

[spark verge]

Well now your reward is that you have to git-bisect it

[feral island]

I checked out that commit and still couldn't reproduce (on macos)

can't repro on linux either

[unkempt rock]

and what?

[spark verge]

How to remove a ban is detailed when you join the discord and agree to the rules

[unkempt rock]

and what?

i am not a discord nerd here i just didn't know so i asked
any problems?

[spark verge]

This isn't the correct channel to ask in

[unkempt rock]

Ok i didn't know dude

[spark verge]

You can DM @summer lichen

[nocturne lance]

I am dealing with ctypes, and getting NameError. there is a structure that points to its parent
something of the lines of:

class RunHeader(ctypes.Structure):
    _fields_ = [
            ...
            ('mv_header', ctypes.POINTER(MvHeader)),
            ...
    ]

class MvHeader(ctypes.Structure):
    _fields_ = [
            ...
            ('run_header', ctypes.POINTER(RunHeader)),
            ...
    ]

is there a efficient way to deal with this? that isn't so hacky

[raven ridge]

I answered you in #c-extensions

[glass mulch]

It repros for me in a JIT build only (I should be more organized with my builds)

[glass mulch]

JIT debug main:
```py`
Python 3.14.0a1+ (heads/main:d467d9246c, Oct 30 2024, 22:52:43) [GCC 11.4.0] on linux
Type "help", "copyright", "credits" or "license" for more information.

import sre_constants; sre_constants._makecodes('', {}, 10)
<python-input-0>:1: DeprecationWarning: module 'sre_constants' is deprecated
import sre_constants; sre_constants._makecodes('', {}, 10)
python: Python/ceval.c:966: _PyEval_EvalFrameDefault: Assertion `STACK_LEVEL() >= level' failed.
Aborted

[feral island]

nice, definitely report this as a bug, I'm guessing @final geode will be very interested. (I haven't tried to repro)

[glass mulch]

Aborts on JIT+debug, segfaults on JIT main.

[final geode]

Thanks! https://github.com/python/cpython/issues/126222

[unkempt rock]

Yeah i did thanks

[glass mulch]

ISTM that tracing of audit hooks with sys.monitoring doesn't work like it should: adding a __cantrace__ attribute to the audit hook still doesn't trace it's CALL events:

import inspect
import sys

sys.monitoring.use_tool_id(1, "monitor")
sys.monitoring.set_events(1, sys.monitoring.events.CALL)

def monitor(code, offset, function, obj):
    print(f"MONITORED: {function}")

sys.monitoring.register_callback(1, sys.monitoring.events.CALL, monitor)


def g(*args):
    print()

def f(*args):
    print(args)

f.__cantrace__ = True

class F:
    __cantrace__ = True
    def __init__(self):
        self.__cantrace__ = True
    def __call__(self, *args, **kwargs):
        print(args)

sys.addaudithook(F())
sys.addaudithook(f)

for x in range(3):
    g.__code__
    g()

# Output:
# ('sys.addaudithook', ())
# ('sys.monitoring.register_callback', (<function monitor at 0x7f985301b590>,))
# ('sys.monitoring.register_callback', (<function monitor at 0x7f985301b590>,))
# MONITORED: <class 'range'>
# MONITORED: <built-in function print>
# ('object.__getattr__', (<function g at 0x7f9853f168d0>, '__code__'))
# MONITORED: <built-in function print>
# ('object.__getattr__', (<function g at 0x7f9853f168d0>, '__code__'))
# MONITORED: <function g at 0x7f9853f168d0>
# MONITORED: <built-in function print>
# 
# MONITORED: <built-in function print>
# ('object.__getattr__', (<function g at 0x7f9853f168d0>, '__code__'))
# MONITORED: <built-in function print>
# ('object.__getattr__', (<function g at 0x7f9853f168d0>, '__code__'))
# MONITORED: <function g at 0x7f9853f168d0>
# MONITORED: <built-in function print>
# 
# MONITORED: <built-in function print>
# ('object.__getattr__', (<function g at 0x7f9853f168d0>, '__code__'))
# MONITORED: <built-in function print>
# ('object.__getattr__', (<function g at 0x7f9853f168d0>, '__code__'))
# MONITORED: <function g at 0x7f9853f168d0>
# MONITORED: <built-in function print>

Which seems to go against the intention of the C code:

        /* Disallow tracing in hooks unless explicitly enabled */
        PyThreadState_EnterTracing(ts);
        while ((hook = PyIter_Next(hooks)) != NULL) {
            PyObject *o;
            int canTrace = PyObject_GetOptionalAttr(hook, &_Py_ID(__cantrace__), &o);
            if (o) {
                canTrace = PyObject_IsTrue(o);
                Py_DECREF(o);
            }
            if (canTrace < 0) {
                break;
            }
            if (canTrace) {
                PyThreadState_LeaveTracing(ts);
            }
            PyObject* args[2] = {eventName, eventArgs};
            o = _PyObject_VectorcallTstate(ts, hook, args, 2, NULL);
            if (canTrace) {
                PyThreadState_EnterTracing(ts);
            }
            if (!o) {
                break;
            }
            Py_DECREF(o);
            Py_CLEAR(hook);
        }

Can you see any obvious error?

[merry bramble]

cc. @crimson drum

[glass mulch]

As an aside, not monitoring these audit hooks makes my attempt to hack them after they are set harder, which might be good. I just ask because it seems the code intends to make them monitorable (and there are no tests of this that I can find).

Hmm, I think this is about tracing only, not monitoring. Which would mean you cannot monitor CALL events of sys.audit hooks at all.

[ruby elm]

I've recently been implementing python's regex flavor from scratch, and the explanations given in both the re __init__ docstring and on the re documentation page don't match the actual behavior. The ones I've found so far:

• \0 refers to the null character, not capture group 0
• \YYY refers to octal character YYY, not capture group YYY
• [\b] matches \x08. not b
• \xYY is hex character YY
• The maximum m and n in {m} and {m,n} is 4294967294, or 2**32-2
  Ideally this inconsistency wouldn't exist, but I'm not 100% sure where these notes would go, or what the process would be to add them.

[spark magnet]

\0 refers to capture group 0 if you are in an r-string.

[gilded flare]

1. documented

Groups are numbered starting from 1.

2. documented

This special sequence can only be used to match one of the first 99 groups. If the first digit of number is 0, or number is 3 octal digits long, it will not be interpreted as a group match, but as the character with octal value number.

3. documented, right after point no. 4

(Note that \b is used to represent word boundaries, and means “backspace” only inside character classes.)

4. documented

Most of the escape sequences supported by Python string literals are also accepted by the regular expression parser:

\a      \b      \f      \n
\N      \r      \t      \u
\U      \v      \x      \\

5. undocumented
   https://docs.python.org/3/library/re.html

[spark magnet]

in other words, you have to get \ 0 to the regex engine, where it will mean group 0. r"\0 is one way to do that, "\\0 is another.

[spark magnet]

there is a capture group 0, it's the entire matched string.

[gilded flare]

that works?

i've only used it in re.Match objects

[ruby elm]

!e print(__import__("re").fullmatch(r"\0", "\0"))

[fallen slate]

:white_check_mark: Your 3.12 eval job has completed with return code 0.

<re.Match object; span=(0, 1), match='\x00'>

[spark magnet]

it only makes sense in the replacement for re.sub. Perhaps the docs need to clarify that.

[gilded flare]

ah
#bot-commands message

docs actually match there

[ruby elm]

!e Also no print(__import__("re").sub(r"a", "a", "\0"))

[fallen slate]

:white_check_mark: Your 3.12 eval job has completed with return code 0.

�

[gilded flare]

!e print(__import__("re").sub(r"a", r"\0", "a"))

[fallen slate]

:white_check_mark: Your 3.12 eval job has completed with return code 0.

�

[spark magnet]

did you read what I said about r-strings?

[gilded flare]

there is no such thing as a \0 group

not even in r-strings

[spark magnet]

@ruby elm in any case, you can make a pull request to improve the docs if you like.

[ruby elm]

Ah, I see I missed those in the re docs. I suppose then the question is should the re __init__ be updated to match that, since it gives an incomplete explanation?

[spark magnet]

there is: the complete matched string.

[ruby elm]

!e print(__import__("re").sub(r"a", "a", r"\0"))

[gilded flare]

0 index in match objects, sure, but \0 does not exist specially

[fallen slate]

:white_check_mark: Your 3.12 eval job has completed with return code 0.

\0

[gilded flare]

it's re.sub(from_, to, string) btw

[feral island]

!e print(__import__("re").sub(r"a", r"\0", r"a"))

[fallen slate]

:white_check_mark: Your 3.12 eval job has completed with return code 0.

�

[spark magnet]

i see what you mean.

you can use \g<0> though

[feral island]

!e print(__import__("re").sub(r"a", r"\g<0>", r"a"))

[fallen slate]

:white_check_mark: Your 3.12 eval job has completed with return code 0.

a

[gilded flare]

tbf \g is undocumented

[spark magnet]

it's in the docs

[gilded flare]

for one small part?

[spark magnet]

what does that mean?

[gilded flare]

i don't see a dedicated section for it

considering \0 vs \g<0>

and also \g<name>

[spark magnet]

i see what you mean. That could also be added.

[ruby elm]

So with the limit on m and n in {m} and {m,n}, should that be listed somewhere as a concrete number? It comes from here MAXREPEAT = _NamedIntConstant(MAXREPEAT, 'MAXREPEAT') but does that mean it could change based on the implementation? Since it think those come from `here ```c

define SRE_MAXREPEAT (~(SRE_CODE)0)

...

define SRE_MAXREPEAT ((SRE_CODE)PY_SSIZE_T_MAX)```

[spark magnet]

i wouldn't list the limits, since they are far larger than anyone will ever hit.

it feels like trivia

[ruby elm]

\g is explained in the re.sub part of the docs https://docs.python.org/3/library/re.html#re.sub

[gilded flare]

Unlike \number, \g<number> supports a group number starting at 0, referring to the entire matched string.
or something similar

[ruby elm]

The backreference \g<0> substitutes in the entire substring matched by the RE.

[gilded flare]

i mean that's not the section i expected it to be in

but it is re.sub()-only behavior

[ruby elm]

The only other question I have is with the docstring in the re module's __init__ Which gives a lot less detail than the documentation page, and is what I was mainly going off of. Should it have a link back to the docs saying "see here for the full explination"? Be kept in sync with the docs? Not try to explain and just link to the docs?

[feral island]

I think it should give a concise summary, aiming for correctness but not comprehensiveness

Could be a good idea to add a line like "refer to the Python docs page for details"

[spark magnet]

i don't think we need to say "refer to the docs". We'd add that to every docstring.

[ruby elm]

Oh no subjectivity. Where is "correctness but not comprehensiveness"? Since it currently doesn't have anything about substitution behavior, or \number being either \x00 or \group or \octal. It also doesn't have the \b is boundary, except in [\b] where it's backspace. Should those things be added to it? It also says " If the ordinary character is not on the list, then the resulting RE will match the second character" but that's not right since it also accepts the normal python escape sequences.

[feral island]

I might actually trim it down significantly and mention only the most commonly used regex features

The idea would be to let new users who use help(re) get a quick sense of what you can do with it

[ruby elm]

Interesting, I didn't know help(module) linked back to the module docs. I'm not sure how much you would want to shrink it down, but at least the default terminal size on windows gives very little space before the -- More --. Also, where's the help on help? I think I found the controls through trial and error, but help(help) leads to _Helper, not an actual explination. It would also suck if someone got trapped into help without knowing how to quit through q or ctr+c, since for me the re help text is over 800 lines long.

[dusk comet]

.__doc__ vs .__doc_extra__

[spark magnet]

we should fix help(help) if we can.

[spark magnet]

help() is helpful

also help

and in 3.13, help does what help() did in previous versions

and help(help) doesn't natter on about Helper, so that's already taken care of.

oops, it does

[feral island]

it does have some useful info ``` | Calling help() at the Python prompt starts an interactive help session.
| Calling help(thing) prints help for the python object 'thing'.

[ruby elm]

Interesting, but that also doesn't list the controls when stuck in the -- More -- state. Are those actually listed anywhere?
Something else fun is that "help mode" help> is not useful for re, since doing modules re gives at least 100 results.

[feral island]

But I think it'd be nicer if help(help) gave you help about the help system instead

[ruby elm]

help>re does work, but that wasn't my first instinct.

Interestingly, help>help is already covered, so I wonder why help(help) isn't.

[feral island]

it seems like it just recursively creates a nested help session

which I'm not sure is the most useful thing to do 🙂

wait no, that's not true; it only prints the help intro

[ruby elm]

This wasn't my first instinct because help>x and help(x) work different. help(re) fails if re isn't in scope, but succeeds regardless of what re is in help>re

I see, so it's help('re') that does the same thing.

[spark verge]

What's help>?

[ruby elm]

The help utility thing from help or help()

Another fun thing I guess you could do is help(help()), which puts you into help>, then prints the help for NoneType once you exit help>. Since help(object) returns None, you can infinitely chain help(help(...)) to get however many NoneType help texts printed in a row.

[spark verge]

I just got this:

>>> help>help
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: '>' not supported between instances of '_Helper' and '_Helper'

[feral island]

first type help, which gives you a help> prompt

then type help (or anything) in that prompt

[spark verge]

oh there's a help repl TIL

[ruby elm]

I don't think anyone would actually do this unless they are purposefully trying to break it, but it is silly that it works.

Another thing you can do is (help(), help()) which makes you have to exit the help repl twice.

[spark verge]

This is the _Helper.__repr__ is a repl?

[ruby elm]

And I did actually find something bad. If you do [help() for _ in iter(int, 1)] or any other infinite iterator, you get the terminal stuck in the help repl permanently.

[feral island]

That's kind of asking for it, isn't it?

If you do [input() for _ in itertools.count()] you also get stuck

[ruby elm]

I mean yes, but normally even if you get the repl stuck you can ctrl+c to get out, but this is immune to even that.

ie with [input() for _ in __import__('itertools').count()], sure the terminal hangs takes input forever, but ctrl+c still works to get everything back to normal. [help() for _ in iter(int, 1)] seems to be permanent, you have to close the terminal.

The closest thing I can think of is if you caught the ctrl+c, ie ```py
while True:
try:
input()
except:
pass

[wispy grotto]

anyone have strong opinions on docs signatures and language around args/kwargs? i'm curious if anyone has feedback on the discussions https://github.com/python/cpython/issues/125897 and https://github.com/python/cpython/pull/125945#discussion_r1816853024

[jade raven]

maybe you'd be interested in: https://discord.gg/ygjd3XwJ

[uneven raptor]

yeah, this issue is way in depth than i originally anticipated when reviewing it. extra eyes would be appreciated

[glass mulch]

Is this worth an issue, or is it meh?

For a debug non-gil build (even with GIL enabled) of main on Linux, I get an abort at finalization for this code:

./python -c "import gc; gc.freeze(); gc.is_finalized(lambda: None); gc.collect(); gc.get_count(); gc.collect(); gc.unfreeze()"

Error looks like:

Python/gc_free_threading.c:550: validate_refcounts: Assertion "!gc_is_unreachable(op)" failed: object should not be marked as unreachable yet
Enable tracemalloc to get the memory block allocation traceback

object address  : 0x20000360f10
object refcount : 1152921504606847590
object type     : 0x55d414c25660
object type name: dict
object repr     : {'__name__': 'builtins', '__doc__': [...]}

Fatal Python error: _PyObject_AssertFailed: _PyObject_AssertFailed
Python runtime state: finalizing (tstate=0x000055d414ca77e0)

Current thread 0x00007f222228a740 (most recent call first):
  Garbage-collecting
  <no Python frame>
Aborted

[uneven raptor]

you could report it, but i don't know if it will get fixed. breaking things with gc is typically your fault

[feral island]

That sounds like a bug to me. Silly tricks with gc.get_referrers aren't always fixable but this looks like a pretty normal way to use gc.

[glass mulch]

Here's one that happens before finalization, it's the same error but with a shorter repr:

import gc
gc.freeze()
gc.is_finalized(lambda: None)
gc.collect()
gc.unfreeze()
gc.collect()

Error looks like:

Python/gc_free_threading.c:550: validate_refcounts: Assertion "!gc_is_unreachable(op)" failed: object should not be marked as unreachable yet
Enable tracemalloc to get the memory block allocation traceback

object address  : 0x200007373f0
object refcount : 1152921504606846977
object type     : 0x55adadc26660
object type name: dict
object repr     : {'__name__': '__main__', '__doc__': None, '__package__': None, '__loader__': <_frozen_importlib_external.SourceFileLoader object at 0x20000330a20>, '__spec__': None, '__builtins__': <module 'builtins' (built-in)>, '__file__': '/home/fusil/python-61/gc-assertion-abort-2/source.py', '__cached__': None, 'gc': <module 'gc' (built-in)>}

Fatal Python error: _PyObject_AssertFailed: _PyObject_AssertFailed
Python runtime state: initialized

Current thread 0x00007f1b47a64740 (most recent call first):
  Garbage-collecting
  File "/home/fusil/python-61/gc-assertion-abort-2/source.py", line 6 in <module>
Aborted

[glass mulch]

A simple one:

import curses
curses.napms(37)

Fatal Python error: _Py_CheckFunctionResult: a function returned a result with an exception set
Python runtime state: initialized
_curses.error: must call initscr() first

The above exception was the direct cause of the following exception:

SystemError: <built-in function napms> returned a result with an exception set

Current thread 0x00007fe41a8c3740 (most recent call first):
  File "/home/fusil/python-73/curses-fatal/source.py", line 3 in <module>
Aborted

[feral island]

Weird that is_finalized() would have an effect here, I took a quick look at the code and don't see how it could have a side effect

[feral island]

Report a bug 🙂 That one is likely easy to fix

[glass mulch]

This one is probably a "don't do that then": a segfault by calling tracemalloc on different threads, only happens with PYTHON_GIL=0.

# This segfaults with PYTHON_GIL=0, works with PYTHON_GIL=1
from threading import Thread
import tracemalloc

alive = [
    Thread(target=tracemalloc.take_snapshot, args=()),
    Thread(target=tracemalloc.clear_traces, args=()),
]

tracemalloc.start()
import inspect # This seems to be important, has to happen after .start()

for obj in alive:
    print('START', obj)
    obj.start()

[glass mulch]

A different gc assert, happens with PYTHON_GIL=0 only:

from threading import Thread
import gc

gc.freeze()
alive = [Thread(target=gc.freeze, args=())]
gc.collect()
gc.unfreeze()
gc.collect()

Python/gc_free_threading.c:554: validate_refcounts: Assertion "op->ob_tid == 0" failed: merged objects should have ob_tid == 0
Enable tracemalloc to get the memory block allocation traceback

object address  : 0x20000278210
object refcount : 1152921504606846994
object type     : 0x5616274befc0
object type name: type
object repr     : <class '_thread.lock'>

Fatal Python error: _PyObject_AssertFailed: _PyObject_AssertFailed
Python runtime state: initialized

Current thread 0x00007f34ed160740 (most recent call first):
  Garbage-collecting
  File "/home/fusil/python-104/gc-abort-assertion/source.py", line 8 in <module>
Aborted

[uneven raptor]

you can report that, i'll take a look

[glass mulch]

It's also possible to hit this different assert, but I haven't minimized it:

Python/gc_free_threading.c:558: validate_refcounts: Assertion "op->ob_tid != 0" failed: unmerged objects should have ob_tid != 0p

[glass mulch]

Which one do you want first? gc aborts, tracemalloc segfault, curses abort or a strange grp one?

[uneven raptor]

gc abort, but you can go ahead and report the other two. worst case scenario they're wontfixes

well, i think i found part of the problem. the reference count of a lambda on the free threaded build is somehow significantly higher than that of an immortal object (1152921504606846976 for lambda: None on my end, but None itself has a refcnt of 3221225472, so that shouldn't be possible)

[glass mulch]

https://github.com/python/cpython/issues/126312

[glass mulch]

Not sure this helps, but I can't get a repro if I use a stored lambda, seems to work only with a fresh one:

import gc
gc.freeze()
l = lambda x: None
gc.is_finalized(l)
gc.collect()
gc.unfreeze()
gc.collect()
# No abort

[uneven raptor]

there's definitely something shady going on with the reference counts

i suspect DRC has something to do with it

[glass mulch]

If you pass (x for x in range(10)) to is_finalized it also aborts, but if you store it in a variable it doesn't. So it looks like it has to be a fresh object with no references... not sure of what kinds yet.

[spark verge]

I think the error message for a missing file in a package could be improved

currently it looks like this:

 graingert@conscientious  ~/projects/demo_imports   main  tree
.
└── demo_imports
    ├── __init__.py
    └── __pycache__
        └── __init__.cpython-312.pyc

3 directories, 2 files
 graingert@conscientious  ~/projects/demo_imports   main  cat demo_imports/__init__.py 
from . import _version
 graingert@conscientious  ~/projects/demo_imports   main  python -m demo_imports 
/usr/bin/python: Error while finding module specification for 'demo_imports.__main__' (ImportError: cannot import name '_version' from partially initialized module 'demo_imports' (most likely due to a circular import) (/home/graingert/projects/demo_imports/demo_imports/__init__.py))
 ✘

demo_import/_version.py is missing

demo_import/__init__.py tries to import from . import _version

but the error message seems to imply a cyclic import

here's the context: #1302360304562339961 message

[limpid forum]

Because of the dual from package import module and from module import name
Since file is missing, python tries to interpret it as second thing. So . becomes current package's init file... where this import was, hence circular import error

[glass mulch]

Here's a way to get a SystemError (or abort in debug builds) by accessing a memoryview from threads. It sometimes segfaults on a no-gil debug build, but I'm not sure being no-gil is relevant.

Is it worth filling an issue?

from threading import Thread

av = memoryview(b"something")
av.release()

alive = [
    Thread(target=av.release, args=('/bin/sh')),
    Thread(target=av.__iter__, args=())
]

for obj in alive:
    try:
        print('START', obj)
        obj.start()
    except Exception:
        pass

Rare segfault:

Exception in thread Thread-1 (release):
Fatal Python error: _Py_CheckFunctionResult: a function returned a result with an exception set
Python runtime state: initialized
ValueError: operation forbidden on released memoryview object

The above exception was the direct cause of the following exception:

SystemError: <method-wrapper '__iter__' of memoryview object at 0x200006a6050> returned a result with an exception set

Current thread 0x00007f750b5d0640 (most recent call first):
  File "/home/danzin/projects/mycpython/Lib/threading.py", line 992 in run
  File "/home/danzin/projects/mycpython/Lib/threading.py", line 1041 in _bootstrap_inner
  File "/home/danzin/projects/mycpython/Lib/threading.py", line 1012 in _bootstrap

Thread 0x00007f750bdd5640 (most recent call first):
  File "<frozen importlib._bootstrap_external>", line 142 in _path_split
Python/traceback.c:979:18: runtime error: member access within misaligned address 0x000000000001 for type 'struct _PyInterpreterFrame', which requires 8 byte alignment
0x000000000001: note: pointer points here
<memory cannot be printed>
Segmentation fault

Common abort:

Fatal Python error: _Py_CheckFunctionResult: a function returned a result with an exception set
Python runtime state: initialized
ValueError: operation forbidden on released memoryview object

The above exception was the direct cause of the following exception:

SystemError: <method-wrapper '__iter__' of memoryview object at 0x200006a6050> returned a result with an exception set
Aborted

[feral island]

yes

[glass mulch]

https://github.com/python/cpython/issues/126341

[heady maple]

Hello! I want to ask, what does Python do during lexical analysis, when the indentation level is not in the stack? Does it look ahead for next indents to get the nearest context it can have, or just propagate to the parser?

[spark magnet]

maybe idk what you mean. When I try it, I get, "IndentationError: unindent does not match any outer indentation level"

[heady maple]

On the stream of tokens level, that is. I'm not sure if that indentation error is raised by the lexer or the parser. I have no idea.

Does it try to 'round off' to the nearest indentation level below for error recovery?

[spark magnet]

it doesn't seem to do any error recovery. A dedent to a level not seen before is an error. Done.

[heady maple]

So if it encounters this error, it panics. Hmm.

[quick snow]

Doesn't this dedent to an unseen level?

if True:
               x = (
  2
)

I'm pretty sure it doesn't; the lexer doesn't emit a dedent inside parens etc. But how does it know? Does it count?

[heady maple]

Grouping symbols ignore whitespace, even if it's after a line break

[quick snow]

And that doesn't happen in the parser?

[heady maple]

I actually have no idea, to be honest.

[quick snow]

That means the lexer must have a stack of open grouping characters (or at least a grouping count)

[spark magnet]

i don't have any insight into the inner workings. I just tried a bad dedent and saw what happened.

[rose schooner]

i think it has both, yeah

it only generates indents/dedents when in any type of grouping

[spark magnet]

You can experiment with python -m tokenize and type code at it

[quick snow]

When outside one, you mean.

[rose schooner]

oh yeah

[glass mulch]

fusil found :D:

python: ./Include/internal/pycore_stackref.h:99: _PyStackRef_FromPyObjectSteal: Assertion `obj != NULL' failed.
Aborted

me having to reduce a huge file with a bug that only triggers sometimes: :/

[glass mulch]

Ok, reduced. For a no-gil main debug build on Linux, I get an abort with the following code. Do you think it's worth filling an issue?

from threading import Thread
from collections import ChainMap, Counter, defaultdict

dictionary = dict.fromkeys(range(100))
counter = Counter(range(100))
chainmap = ChainMap(dictionary, counter)

def mutate_removing():
    for x in range(10):
        dictionary.pop(list(dictionary.keys()).pop())
        counter.pop(list(counter.keys()).pop())
        chainmap.pop(list(chainmap.keys()).pop())

def mutate_adding():
    for x in range(50):
        dictionary[max(dictionary) + x + 1] = x
        counter[max(counter) + x + 1] = x
        chainmap[max(chainmap) + x + 1] = x

alive = []

for x in range(55):
    alive.append(Thread(target=mutate_removing, args=()))
    alive.append(Thread(target=mutate_adding, args=()))

for obj in alive:
    try:
        print('START', obj)
        obj.start()
    except Exception:
        pass

Abort looks like:

python: ./Include/internal/pycore_stackref.h:99: _PyStackRef_FromPyObjectSteal: Assertion `obj != NULL' failed.
Aborted

[feral island]

yes, pure Python code like that should never crash

that includes the nogil build

Do you need all three of the dict, the Counter, and the ChainMap, though?

worth trying if it also crashes with three dicts or something like that

[glass mulch]

Thanks! Let me check.

Turns out the dict isn't necessary, we can get the same abort with two ChainMaps that reference a Counter. Doesn't seem to work with a dict in the place of the Counter or of the ChainMap.

[glass mulch]

This is getting silly 😄 . New issue, I guess?

./python -c "from unittest.mock import MagicMock as MM; from _pickle import load; load(MM())"
python: Objects/typeobject.c:5583: _PyType_LookupRef: Assertion `!PyErr_Occurred()' failed.
Aborted (core dumped)

[feral island]

go for it! new PR would be even better 😄

that one feels like it shouldn't be hard to fix

[glass mulch]

Can anyone confirm that the following segfaults? I'm suspicious it might be something in my system.

python -c "from _ssl import _SSLSocket; s =_SSLSocket(); s.do_handshake()"

[quick snow]

Can confirm (x64 Linux, 3.10-3.13)

[glass mulch]

Thank you!

[spark verge]

!e

from _ssl import _SSLSocket; s =_SSLSocket(); s.do_handshake()

[fallen slate]

:warning: Your 3.13 free threaded eval job has completed with return code 139 (SIGSEGV).

[No output]

[spark verge]

139 SIGSEGV

[glass mulch]

Also

python -c "from _pyrepl._minimal_curses import tigetstr; tigetstr('')"

[crimson drum]

Please file issues for these. Thanks

[gray galleon]

is there a proposal for Private and Protected type hints

[flat gazelle]

Probably not, the __ and _ prefixes fill that role reasonably well.

[fossil pagoda]

where is the gerneal chat?

looking for a beginner python discord server

[jovial flame]

#python-discussion is for general python discussions. Off topic discussions can go in one of the three off topic channels like #ot0-psvm’s-eternal-disapproval

[gray galleon]

__ = Private
_ = Protected (probably?)
so yeah

[dusk comet]

_private
__i_need_this_only_in_this_class
__protected__

[spark magnet]

don't make up your own __dunder__ names: those are for Python

[ruby elm]

I might have actually found something not in the re docs. The part about char sets [] says

Character classes such as \w or \S (defined below) are also accepted inside a set, although the characters they match depend on the flags used.
And that is true for all of the escapes listed, except \A and \Z, which give a "bad escape" error.

[quick snow]

Makes sense to me, those aren't characters or character classes themselves. I don't see how it would make sense to put them in a character class.
Although coming up with a good reason to put \b inside one is also hard.

[ruby elm]

Then I'd say it's missing in the other direction, since octal/unicode/hex escapes work, despite not being a character class like \w or \S.

\B also doesn't work, so it looks to be all the zero width escapes

Which makes sense, since it doesn't make sense for a character set to have a zero width match, but it is a thing not mentioned.

[quick snow]

Octal/unicode/hex escapes don't work, they get parsed as the corresponding character by Python itself.
It's not all zero-width escapes though, since \b is also one. Interesting.

[ruby elm]

I'd say that they work, since they get transformed correctly even inside an r-string ```pycon

re.fullmatch(r"[\100]", "@")
<re.Match object; span=(0, 1), match='@'>

If they didn't work, I'd expect this to instead get parsed as an escaped `1`, followed by two `0`s. 

That's because specifically in char groups, `\b` means backspace instead of word boundary.

[quick snow]

Huh, indeed; mea culpa.

Interesting, didn't know \b meant backspace inside of brackets.

[spark verge]

People seem to sort of make up their own dunder names a lot

Eg __version__ or __{pypi_name}__

[spark magnet]

yes. they shouldn't.

[spark verge]

__version__ is particularly weird, as it's banned by docs.python.org but the pep 396 has a note saying it might be ok

[flat gazelle]

_name is ill-suited for private, since it cannot exist in a subclass without conflicting.

Usually it is fine since your class won't be subclassed

[dusk comet]

__mylib_mything__ has a very low chance of collision with future cpython names

[faint river]

I would use sunders

[wanton flame]

Where do the docs ban __version__?

[gray galleon]

so its protected

[spark verge]

Any use of * names, in any context, that does not follow explicitly documented use, is subject to breakage without warning.

https://docs.python.org/3/reference/lexical_analysis.html#reserved-classes-of-identifiers

[grave jolt]

Not every concept will map neatly from Java to Python

A single underscore usually means that a name is not public and should not be relied upon. It's not "protected" in the sense that subclasses can access it (in fact, it is also used for module-level functions and local/global variables, where subclasses don't exist). If a library class is using some _underscored members, it means that they can change in the next version without informing you

There are other modes of visibility like "public but only within this library" and "public but only within this module/subpackage", that also don't have a convention. If you care about it, you can leave a comment or something like that.

But normally _underscored means "private", as in "if you are not the maintainer of this code, don't use this variable/class/module"

[quick snow]

or maybe "warranty void if broken", i.e. "if you touch this, all promises are off"

[grave jolt]

yep

[spark verge]

Also sealed

So classes can be subclassed in one context but not another

[glass mulch]

New one, confirmations welcome.

python3 -c "import time; time.pthread_getcpuclockid(2)"
Segmentation fault

[feral island]

Traceback (most recent call last):
  File "<string>", line 1, in <module>
    import time; time.pthread_getcpuclockid(2)
                 ^^^^^^^^^^^^^^^^^^^^^^^^^^
AttributeError: module 'time' has no attribute 'pthread_getcpuclockid'
``` on mac

but segfaults on 3.12 on linux

[glass mulch]

And

python -c "from sys import maxsize; from itertools import count; count(maxsize)"
python: ./Modules/itertoolsmodule.c:3325: itertools_count_impl: Assertion `(cnt != PY_SSIZE_T_MAX && long_cnt == NULL && fast_mode) || (cnt == PY_SSIZE_T_MAX && long_cnt != NULL && !fast_mode)' failed.
Aborted (core dumped)

[glass mulch]

Thanks!

[feral island]

reproes on main+macos but not 3.12+linux

[glass mulch]

main+linux aborts too

[jade raven]

petition to add a segfault hunter as a role like core dev and award it to @glass mulch

never seen so many segfaults for python in my life than what i've seen @glass mulch push out in the last month

[glass mulch]

This is all thanks to Victor Stinner, he made a tool (fusil) in 2008 that found many segfaults, I just rediscovered it, added a few things and am running it when time allows. Talked to him and will publish it in a bit.

[glass mulch]

Aw, time.pthread_getcpuclockid is documented as being segfaulty:
https://docs.python.org/3/library/time.html#time.pthread_getcpuclockid

[feral island]

Interesting, that's not great for a Python-level function. Wonder if it's possible to validate the thread id so we can avoid the segfaults.

[glass mulch]

Looks like it's not possible:
https://github.com/python/cpython/issues/115378

[uneven raptor]

theoretically, you could iterate over the active thread states and check if the passed ID is in there, but that would only work for threads that have called PyGILState_Ensure or something similar

[raven ridge]

Just install a SIGSEGV handler!

[feral island]

depending on how the function works in C you might be able to do more interesting things than get a segfault

[raven ridge]

I doubt it, I'd expect it's all reads and no writes, and anyway this is nothing you couldn't do with ctypes in a typical CPython build that isn't trying to do any sandboxing

the interpreter could maybe guard against the segfault by using mincore, but that's not enough - it's not just that the address needs to be mapped, but also that the region it's mapped into needs to be large enough to hold a pthread structure, and that size is unknown since the structure is opaque

it would be easy enough to make it less likely to segfault, but not at all easy to make it never segfault, I think

and any attempt to validate it could still fall afoul of TOCTOU

[uneven raptor]

i wrote something a while back that converted SEGV signals into exceptions that you could literally try/except and then proceed with execution like nothing happened

[jovial flame]

!cban 1136816274303885322 We are not a server for hiring

[fallen slate]

:incoming_envelope: :ok_hand: applied ban to @near fiber permanently.

[glass mulch]

Here's a cute abort:

python -c "b''.__buffer__(-2**31 - 1)"
python: Objects/typeobject.c:9321: wrap_buffer: Assertion `_Py_STATIC_CAST(Py_ssize_t, _Py_STATIC_CAST(int, (flags))) == (flags)' failed.
Aborted

[feral island]

oops I wrote that code. Please send a bug and I'll work on it

[glass mulch]

https://github.com/python/cpython/issues/126594

[uneven raptor]

you keep giving me and bénédikt things to do 😅

[real herald]

Hii is there anyone who is preparing for PECP or had PECP?

[merry venture]

disclaimer: no bugs will be harmed during fixing the issue

[glass mulch]

There's something interesting going on. With a lot of code (35 lines) involving interpreters (hundreds) and threads (thousands), it may non-deterministically result in one of the following errors, from most common to rarest:

1. tstate_delete_common assertion, which seems to indicate a thread creation failed, see https://github.com/python/cpython/issues/109746

python: Python/pystate.c:1739: tstate_delete_common: Assertion `tstate->_status.cleared && !tstate->_status.finalized' failed.

2. Fatal Python error, seems to be https://github.com/python/cpython/issues/113148

Fatal Python error: Py_EndInterpreter: not the last thread
Python runtime state: initialized

Current thread 0x00007f09af37e640 (most recent call first):
  <no Python frame>

Thread 0x00007f09b0380640 (most recent call first):
  <no Python frame>
Aborted

3. heap_pop: This one might be new and worth a look?

python: Python/index_pool.c:92: heap_pop: Assertion `heap->size > 0' failed.
Aborted

Does anyone want to take a look at the code so we can try to figure out enough to create an issue?

[glass mulch]

Got a bt of the heap_pop one:

#0  __pthread_kill_implementation (no_tid=0, signo=6, threadid=140729519150656) at ./nptl/pthread_kill.c:44
[...]
#4  0x00007ffff7cc67f3 in __GI_abort () at ./stdlib/abort.c:79
#5  0x00007ffff7cc671b in __assert_fail_base (fmt=0x7ffff7e7b130 "%s%s%s:%u: %s%sAssertion `%s' failed.\n%n",
    assertion=0x555555aceaf4 "heap->size > 0", file=0x555555aceae0 "Python/index_pool.c", line=92, function=<optimized out>)
    at ./assert/assert.c:92
#6  0x00007ffff7cd7e96 in __GI___assert_fail (assertion=assertion@entry=0x555555aceaf4 "heap->size > 0",
    file=file@entry=0x555555aceae0 "Python/index_pool.c", line=line@entry=92,
    function=function@entry=0x555555aceb40 <__PRETTY_FUNCTION__.1> "heap_pop") at ./assert/assert.c:101
#7  0x00005555558caa1c in heap_pop (heap=<optimized out>) at Python/index_pool.c:92
#8  0x00005555558cac33 in _PyIndexPool_AllocIndex (pool=pool@entry=0x7ffff780f498) at Python/index_pool.c:173
#9  0x000055555568d9f0 in _Py_ReserveTLBCIndex (interp=interp@entry=0x7ffff780b020) at Objects/codeobject.c:2752
#10 0x0000555555951c33 in new_threadstate (interp=interp@entry=0x7ffff780b020, whence=whence@entry=5) at Python/pystate.c:1516
#11 0x0000555555953994 in _PyThreadState_NewBound (interp=interp@entry=0x7ffff780b020, whence=whence@entry=5)
    at Python/pystate.c:1578
#12 0x0000555555896ee9 in _enter_session (session=session@entry=0x7ffe24ff8740, interp=interp@entry=0x7ffff780b020)
    at Python/crossinterp.c:1548
#13 0x000055555589b9fb in _PyXI_Enter (session=session@entry=0x7ffe24ff8740, interp=interp@entry=0x7ffff780b020,
    nsupdates=nsupdates@entry=0x0) at Python/crossinterp.c:1711
#14 0x00007ffff7c3d217 in _run_in_interpreter (interp=interp@entry=0x7ffff780b020, codestr=0x200006182c8 "f()", codestrlen=3,
    shareables=shareables@entry=0x0, flags=1, p_excinfo=p_excinfo@entry=0x7ffe24ff8870) at ./Modules/_interpretersmodule.c:461

Seems to require PYTHON_GIL=0

Ah, the heap_pop code was added last week in https://github.com/python/cpython/pull/123926

[uneven raptor]

@glass mulch FWIW, the issue is that a subinterpreter can start running after the final destruction check, which breaks a lot

[glass mulch]

Got another subinterpreters abort, seems to be in 3.13t_d but not main:

python: Python/pystate.c:1292: _PyInterpreterState_IDDecref: Assertion `interp->id_refcount != 0' failed.
Aborted (core dumped)

[uneven raptor]

oh yeah, i noticed that a little while back. i don’t think we’re doing anything with IDs right now so let’s leave it

[glass mulch]

Well, this one reduced to a lot less than I thought it would:

import _interpreters

interpr = _interpreters.create()
_interpreters.decref(interpr)

Should I leave it be then?

[glass mulch]

Got another one on main:

python: ./Modules/_interpretersmodule.c:462: _run_in_interpreter: Assertion `!PyErr_Occurred()' failed.
Aborted (core dumped)

[uneven raptor]

that’s probably the exception problem i described in your other issue

[uneven raptor]

yeah, leave it be. i’m not sure what eric is envisioning for incref and decref right now, so we’ll fix it later

[glass mulch]

This has a short repro, might help:

import _interpreters

_interpreters.exec(False, "aaaa", 1)

[uneven raptor]

that looks like a different bug, report it

[glass mulch]

Oh, it segfaults on non-debug builds

[glass mulch]

I want to publish updated fusil this week. Hopefully we can add new kinds of fuzzing input to it to help discover more bugs. If anyone wants to take a peek before it's cleaned up (or run it on some different environment or implementation) ping me and I'll add you to the (for now) private repository.

Thanks all that have evaluated, diagnosed and fixes issues fusil has found. And thanks Victor for fusil 🙂

[jade raven]

could i be added? i want to see how it works

thank you!

[glass mulch]

Done. fusil_messy is what I have been running against CPython and now PyPy. fusil is the repo I want to make public, where I'll reconstruct the main fuzzer from fusil_messy, fusil_python_threaded.

[jade raven]

i saw, thank you, fine w you if i fork for internal testing?

[glass mulch]

The way it works is it imports a module, lists all functions classes and objects in it, and generates random calls to functions and methods with invalid data. Also adds such calls to threads. The fuzzer is simple, the handling of processes, using a different user to run the fuzzing, matching successful runs and other features of the library is where the magic hides.

[glass mulch]

Sure. Feel free to send open or address issues of you want, anything really.

[jade raven]

awesome!

[glass mulch]

The one you want to run is fusil_messy, the working fuzzer is fusil-python-threaded. You need to install python-ptrace. And be sure to create a limited user called fusil and run the fuzzer as root, so it can change the process user to this limited user and keep your precious filesystem safe from wacky calls to pathlib, open, os, etc.

I've been running it on Linux, might work on Mac.

I'll be away for some hours, but let me know if you need any assistance or have any doubts. Happy fuzzing 🙂

[spark verge]

Is there a way to sandbox it?

[jade raven]

<@&831776746206265384>

[spark verge]

I have a weird issue in an issue, there's a reproducer for an issue that should only work on pypy but it works on 3.12.3 and 3.13.0 does anyone know what's causing it: https://github.com/python/cpython/issues/126775 see "Important Meta issue" heading

[flat rune]

hi

[pine wraith]

Hey folks, I was looking at PEP-0589: https://peps.python.org/pep-0589/ and found the concept of totality. I was wondering how is this implemented internally:

class Movie(TypedDict, total=False):
    name: str
    year: int

Here total looks like a kwarg. How is it passed to the metaclass? Can I pass custom kwargs to mine own metaclasses or is it something specific for the use-case of TypedDict?

[flat gazelle]

it is passed to __init_sublcass__ (and to the metaclass objects __call__)

[quick snow]

__call__, really? That would imply it's called with the class already.

It should be __new__

[flat gazelle]

!e

class Foo(metaclass=print, end='hello, world\n'): pass

[fallen slate]

:white_check_mark: Your 3.12 eval job has completed with return code 0.

Foo () {'__module__': '__main__', '__qualname__': 'Foo'}hello, world

[rose schooner]

yes i'm pretty sure it calls __call__

the secret to any class initialization is a __call__() call

[flat gazelle]

it does __call__ of the object that is kept around as the metaclass (ideally an instance of type), which then calls __new__ and __init__ as is proper.

idk whether the kwarg goes to __prepare__ tbh

[rose schooner]

TIL i don't need to specify type as a base when using metaclasses

[flat gazelle]

it is slightly horrifying that this works tbh

[rose schooner]

esoteric python!!

[quick snow]

Well it calls the metaclass of course, but that would go via the metaclasses metaclasses __call__, wouldn't it?

Which for type is type, sure..

[rose schooner]

yes

you never know how many layers of objects you're gonna pass through

[flat gazelle]

yea, that's what I meant with the metaclass object.

it does metaclass.__call__, not directly metaclass.__new__

[rose schooner]

1. TypedDict is not a metaclass, it's a superclass
2. it's passed to TypedDict.__init_subclass__()

[pine wraith]

Thanks @flat gazelle ! That was super quick 🙂
I will check the specifics of __init_subclass__.

[rose schooner]

(type(metaclass).__call__)

[flat gazelle]

ah yea, doing so directly gives you the descriptor smh

[quick snow]

It does type(metaclass).__call__

[pine wraith]

Yes, I meant this metaclass: https://sourcegraph.com/github.com/python/cpython/-/blob/Lib/typing.py?L3125

[quick snow]

!e note how Meta.__call__ is never called.

class Meta(type):
    def __init__(self, *args, **kwargs):
        print("Called meta __init__ with", args, kwargs)

    def __new__(self, *args, **kwargs):
        print("Called meta __new__ with", args, kwargs)
        return super().__new__(self, *args)

    def __call__(self, *args, **kwargs):
        print("Called meta __call__ with", args, kwargs)
        return super().__call__(self, *args)


class Foo(metaclass=Meta):
    pass

print("Before class definition")
class Bar(Foo, ham="spam"):
    pass

[fallen slate]

:white_check_mark: Your 3.12 eval job has completed with return code 0.

001 | Called meta __new__ with ('Foo', (), {'__module__': '__main__', '__qualname__': 'Foo'}) {}
002 | Called meta __init__ with ('Foo', (), {'__module__': '__main__', '__qualname__': 'Foo'}) {}
003 | Before class definition
004 | Called meta __new__ with ('Bar', (<class '__main__.Foo'>,), {'__module__': '__main__', '__qualname__': 'Bar'}) {'ham': 'spam'}
005 | Called meta __init__ with ('Bar', (<class '__main__.Foo'>,), {'__module__': '__main__', '__qualname__': 'Bar'}) {'ham': 'spam'}

[rose schooner]

ah
that's TypedDict's metaclass

or at least something to imitate TypedDict

since it doesn't necessarily use that

what

[fallen slate]

Lib/typing.py lines 3337 to 3338

_TypedDict = type.__new__(_TypedDictMeta, 'TypedDict', (), {})
TypedDict.__mro_entries__ = lambda bases: (_TypedDict,)```

[flat gazelle]

yea, I forgot that you get the attribute before the descriptor protocol in this case.

TypedDict being a def-made object is a bit silly, what.

[rose schooner]

a what.

TIL TypedDict is not a class

[merry bramble]

Take a look at the typing_extensions backport if you want to find things that are even more fun

https://github.com/python/typing_extensions/blob/e888dfdabcfcd03330477cb5de3c25c21029f425/src/typing_extensions.py#L1042

[fallen slate]

src/typing_extensions.py line 1042

@_ensure_subclassable(lambda bases: (_TypedDict,))```

[merry bramble]

That decorator has to do some fun things because the hack that cpython uses to make the TypedDict function subclassable doesn't work on old versions of PyPy

on old versions of PyPy, typing_extensions.TypedDict is therefore a subclassable, callable instance of an anonymous class instead of a function

[pine wraith]

Actually, it seems that the __init_subclass__ hook is not used at all. I ran this code:

from typing import TypedDict


class A(TypedDict):
    a: int

and ran it with python -m trace --trace main.py > out.txt
This is the relevant part:

main.py(4):  class A(TypedDict):
 --- modulename: typing, funcname: <lambda>
typing.py(3319):  TypedDict.__mro_entries__ = lambda bases: (_TypedDict,)
 --- modulename: main, funcname: A
main.py(4):  class A(TypedDict):
main.py(5):      a: int
 --- modulename: typing, funcname: __new__

My suspicion is that it works because __new__ defines a total arg.

[flat gazelle]

yea, the metaclass gets called, which then calls __new__

[swift imp]

Reading discourse on pep 765. What on earth is GvR getting at with the for a[i] in ...: ... comment?

https://discuss.python.org/t/pep-765-disallow-return-break-continue-that-exit-a-finally-block/71348/4

[spark magnet]

I think he's referring to the idea that for a[i] in range(10): is a suprise to people: that the loop target can be something other than a simple variable.

[swift imp]

It is a surprise to me and when trying to Google it the AI said it's syntax error.

It does like an anti pattern

This seems so useless and convoluted

#bot-commands message

[flat gazelle]

I have very rarely used it in cases like

config = {...}
for config['optlevel'] in 0, 1, 2, 3:
    do_tests(config)

[sonic flax]

i want a way to directly fund work on the PEP541 (pypi name retirement/takeover) backlog

it looks like it's over a year long at this point

i will go the entire next week without grabbing a coffee so i can fund this

[dusk comet]

for x, y in zip(X, Y): ...

is a common example of that
zip() produces tuples, and they are unpacked to x, y

for assignment target can be anything that you can put to the left side of =
same applies to some other situations: ```py
with foo as TARGET: ...

I can't recall any other cases sadly :(

[rose schooner]

if for can take tuple targets, why not non-name targets?

for {}() in range(10):
    print("Ta da! Unused loop variable")

[dusk comet]

!e ```py
for [] in [[]]+[[]]+[[]]:
print('foo')

[fallen slate]

:white_check_mark: Your 3.12 eval job has completed with return code 0.

001 | foo
002 | foo
003 | foo

[spark magnet]

THis is Guido's point: Python has primitive constructs that can be combined together in ways that "don't make sense" but you don't forbid all those ways, they just don't get used.

BTW, I have used it. in a class with a turn counter: for self.turn in range(self.nturns):

and doesn't it make sense? for i in seq is like running i = seq[0] etc over and over. The loop target is an assignment target.

[swift imp]

That makes more sense. As does @flat gazelle example

[spark magnet]

I posted an even better example to Bluesky: https://bsky.app/profile/did:plc:w225kiqaw67ddpex4jmoo7rs/post/3lb5ehqs3qc2c

[swift imp]

<@&831776746206265384>

[somber thicket]

please don't advertise in this server

[opaque wagon]

Hello

[unkempt rock]

Woahh

[unkempt rock]

Since when 👁

[glass mulch]

I've just republished fusil, the tool by Victor Stinner that has been finding crashers in CPython and PyPy:
https://mastodon.social/@danzin/113510116056712424
https://github.com/devdanzin/fusil

[spark verge]

Nice, could you update the change log?

[jade raven]

@dusk comet should i focus on that repo then?

[worthy sandal]

Is anyone using python and rust combined in their project? If so what are the places you applied rust and why? I am a python lover and trying to make use of rust but did not find appropriate scenario... thanks in advance!

[dusk comet]

hm? what repo?

[jade raven]

https://github.com/devdanzin/fusil
as in, you're no longer going to work on those other private repos

[inland acorn]

We did for a little while, but that was simply because two different devs working on different components with different preferences

(now the python part is in js, we still got the rust code)

But a more general answer, rust is nice for software you need to distribute to end users, which is a bit harder to do with python

(also rust is just amazing and should be used for everything /hj )

[inland acorn]

(ironically in this project python was the user facing software and rust the backend... Which is why we moved it to js/web)

[dusk comet]

you must be confusing me with someone else

[jade raven]

indeed, i was confusing you with @glass mulch , oops

[bold marsh]

iteration over set isn't ordered by insertion. does this mean iteration over set is O(buckets) instead of O(entries)?

[bold marsh]

answer: yes

[dusk comet]

since buckets/entries=O(1) it doesn't matter much

[bold marsh]

well no since you can add then remove loads of items

[spark verge]

It's buckets+items, and because buckets<items it boils down to just items

[glass mulch]

Yes, that repo was a private one and now it's been made public. So many issues, so little time 🙂

[glass mulch]

I've added basic changes made to the changelog and expect to do it in a more rigorous way before releasing 1.6, thanks for the heads up.

Does anyone know if a commit in, say, the last four days could be responsible for a much increased rate of hitting python: Python/pystate.c:1620: PyThreadState_Clear: Assertion 'tstate->_status.initialized && !tstate->_status.cleared' failed.? It has increased significantly in fuzzer runs. I'm not 100% sure it has to come from code change and not something in the fuzzing machine, but it seems to be stable across reboots. Sorry for asking before building an older revision and testing myself 😬

[jade raven]

right, so no more private stuff, just focus on the public one? i'll open a PR or two soon

[glass mulch]

That's right. Wow, thank you very much! 🙂

[bold marsh]

"because buckets<items" this isn't ever true

from sys import getsizeof
s = set()
print(getsizeof(s))
for i in range(256):
    s.add(i)
print(getsizeof(s))
for i in range(256):
    s.remove(i)
print(getsizeof(s))

also, buckets isn't bounded by items by the above code

[glass mulch]

Ah, turns out to be caused by a fix to the fuzzer code. It means a lot more threads are created now, and I guess it hits https://github.com/python/cpython/issues/109746 because of that. Sorry for the noise.

[glass mulch]

Applying the fix from https://github.com/python/cpython/pull/109761 solves the issue.

[fallen slate]

:incoming_envelope: :ok_hand: applied timeout to @thorny ridge until <t:1732388190:f> (9 minutes and 59 seconds) (reason: duplicates spam - sent 4 duplicate messages).

The <@&831776746206265384> have been alerted for review.

[flint burrow]

Hello

[grave jolt]

@raw night Hello, this server is not the right place to advertise your linkedin profile.

[raw night]

@grave jolt may I know the server which is right place for this ??

[grave jolt]

No, I don't know

[fallen slate]

:incoming_envelope: :ok_hand: applied timeout to @sturdy hound until <t:1732460828:f> (10 minutes) (reason: duplicates spam - sent 4 duplicate messages).

The <@&831776746206265384> have been alerted for review.

[rose schooner]

unpacking operator has inconsistent precedence and yield takes an implicit tuple as a value, which might cause some confusion

why?

!e ```py
import traceback

for i, line in enumerate((
r'print("abc" if 1 else "123")',
r'print(["abc" if 1 else "123"])',
), start=1):
print(f"[{i}]:", line)
try:
exec(line)
except SyntaxError as e:
traceback.print_exc()
print()

def f(*g):
a, b = yield 'a', *g
yield a, b

x = f([1, 2, 3])
print(x.send(None)) # receives first yield
print(x.send('xy')) # sends in place of first yield, receives second yield

[fallen slate]

:white_check_mark: Your 3.12 eval job has completed with return code 0.

001 | [1]: print(*"abc" if 1 else "123")
002 | a b c
003 | 
004 | [2]: print([*"abc" if 1 else "123"])
005 | Traceback (most recent call last):
006 |   File "/home/main.py", line 9, in <module>
007 |     exec(line)
008 |   File "<string>", line 1
009 |     print([*"abc" if 1 else "123"])
010 |                   ^^
... (truncated - too many lines)

Full output: https://paste.pythondiscord.com/UPDGJKH44P4XL4TSPCJYVYOI7Y

[merry venture]

those yields work as expected

unpacking too

not sure where's the confusion here, but maybe i'm too used to it

[rose schooner]

unpacking has 2 different meanings in seemingly same contexts

yield i find more reasonable to be expected

but i feel like it could be interpreted that a = yield 'a' and b, = *g,

[merry venture]

oh, i see

you would a, b = (yield 'a'), *g instead

even print(yield) is an invalid syntax, which makes sense because what would print(yield 1, 2) mean?

and you need either print((yield 1, 2)) or print((yield 1), 2)

[astral gazelle]

This seems entirely unrelated to the channel and server

[grave jolt]

!cban @brittle stag spam

[fallen slate]

:incoming_envelope: :ok_hand: applied ban to @brittle stag permanently.

[merry venture]

you meant the deleted spam, not messages above?

[astral gazelle]

<@&831776746206265384> scam

[jovial flame]

!cban 1285208360328761477 scam

[fallen slate]

:incoming_envelope: :ok_hand: applied ban to @brittle sedge permanently.

[rose schooner]

sometimes i wonder if the server should have self-role access for this channel

[regal glen]

While I really like that idea, a decent amount of python contributors are not the most discord literate. Wouldn’t want to lock them out of here. (If you want to continue this, let’s not here and move it to #community-meta)

[candid vessel]

yes

[jade raven]

is there a reason you're sending random messages in different channels?

[candid vessel]

yes for Voice Verification need 50 msg to complete

[jade raven]

it needs to be organic, not spammy.

[candid vessel]

sorry man not trying to spam just trying to entry conversation

[molten island]

This breaks the #rules number 9. Don't advertise work or solicit paid work

[raven ridge]

Because it's a thin wrapper around the C function https://en.cppreference.com/w/c/program/system and that's how it behaves

[spark verge]

You can run two os.system calls in different threads

Better to just use subprocess.run though

[hallow tulip]

question, I'm running 3.12.5 (dk if it matters as much) and ran this in an ipynb, but the interpreter returns a syntax error at:
print([*"abc" if 1 else "123"])
^^
SyntaxError: invalid syntax

[rose schooner]

that's intended

the interpreter returns a SyntaxError when the expression being unpacked has a precedence lower than the "bitwise OR" (|) operator

[broken forum]

[coarse stone]

Multiplication

Sorry, wrong chat

[gray galleon]

why do floor division with floats return floats

[merry bramble]

Because of possibly unfortunate decisions made a long time ago that cannot realistically be reversed now https://discuss.python.org/t/make-float-i-r-floordiv-return-an-int/24959/1

[final geode]

The real fun is that the type of the result of exponentiation depends on the sign of the exponent.

[crude anvil]

do you have an example i don't think i got you right.

[quick snow]

2**2 is 4 (an int), 2**-2 is 0.5

[crude anvil]

yeah that's what i first though but that makes sense i guess ...

[sour thistle]

you can even get complex numbers from int ** float: ```py

[(-2)**n for n in (2, -2, 0.5)]
[4, 0.25, (8.659560562354934e-17+1.4142135623730951j)]

[crude anvil]

now it's fun lol

[rose schooner]

FPE casually adding an insignificantly low number for the real side

[dusk comet]

>>> ((-1)**-1e-99*-1e99).imag
3.141592653589793
``` 😳

[wild trout]

What is actually going on there? (too lazy to think about it)

[rose schooner]

(-1) ** -x as x approaches 0, the resulting imaginary part (without leading zeroes) approaches pi

no idea why tho :p

[wild trout]

yeah that's not right tho, it should approach 0

lim (-1)^(-x) as x -> 0 should be 1

[dusk comet]

#esoteric-python message

[rose schooner]

cool

[rose schooner]

it does approach 0

but the leading zeroes in the imaginary part also get infinitely many

[dusk comet]

infinitely big zero 🤯

[wild trout]

Oh oops I read the second operation as a power, not a multiply

[rose schooner]

so as it approaches 0, the imaginary part essentially gets wiped out

[wild trout]

I guess it works for any large power

the only assumption is sin(x) = x

but x doesn't need to be that small for sin(x) = x to be roughly true

1       0.0
2       2.0
5       2.938926261462366
10      3.090169943749474
20      3.1286893008046173
100     3.141075907812829
200     3.141463462364135
1000    3.1415874858795636

for i in [1, 2, 5, 10, 20, 100, 200, 1000]:
    print(f"{i}\t{((-1)**(1/i)*(1*i)).imag}")

[grave jolt]

Question: is it specified anywhere that the __iter__ of an iterator should only return self, and shouldn't e.g. reset its state or do other things?
I think that's something that's informally expected of iterators (stuff like the recipe here would not work otherwise https://docs.python.org/3/library/itertools.html#itertools.batched), but is it formally written down anywhere?

[feral island]

I hope so because I recently approved a change to 3.13 that relies on this property. Let me look it up.

https://docs.python.org/3/glossary.html#term-iterator "Iterators are required to have an iter() method that returns the iterator object itself"

[wild trout]

Also here https://docs.python.org/3/library/stdtypes.html#iterator.__iter__

[grave jolt]

Yeah, but this doesn't mean that it doesn't do e.g. ```py
def iter(self):
self.side_effect()
missiles.launch()
self.counter = 0
return self

[feral island]

True. I guess we could explicitly say that, I'd support a docs PR to make that change

[wild trout]

It can do other things, can't it?

[feral island]

It can but it shouldn't

[wild trout]

Why not?

Is that just the semantic of iterators, that the function only does that and not anything else?

That seems to impose a restriction on the programmer that might not be necessary

But yeah definitely the current docs don't make it sound like the programmer shouldn't do other things in __iter__

[flat gazelle]

I do think it'd be reasonable to have that be part of the semantics, yea.

[wild trout]

That seems overly restrictive to me, and not necessary

[grave jolt]

Well, "not doing anything else at all" might be too restrictive
But not mutating the iterator seems reasonable

[shy grove]

Well you can always ignore recommendations

[flat gazelle]

IG you could put like logs or breakpoint() in there

[wild trout]

They're saying to make it a requirement for something to be called an iterator, not a recommendation

[wild trout]

Yeah exactly

[shy grove]

Sounds good to me

Would be weird if my duck stopped quacking

My iterator should behave like an iterator

[grave jolt]

unfortunately, ducks do not live forever

[wild trout]

I'm not smart enough to know if mutating the iterator in __iter__ is enough to no longer behave as an iter, if you are, then fair enough, but you're much smarter than me 😄

[flat gazelle]

that I would argue should not be allowed

[shy grove]

You can always implement your custom methods, having standard methods with standard semantics and definitions does not stop you from doing that

[flat gazelle]

if it were allowed, it would force every correct function handling iterators to carefully manage their iter calls

[feral island]

The advantage of saying __iter__ should have no side effects is that the interpreter is free to call it multiple times if that's convenient for the runtime

(Plus what @flat gazelle said)

You can still have side effects in __iter__, they might just happen multiple times when you might not expect it

[grave jolt]

The reason mutating the iterator is potentially bad is that it makes it problematic to pass the iterator to other functions. See the example I linked from the docs:

def batched(iterable, n, *, strict=False):
    # batched('ABCDEFG', 3) → ABC DEF G
    if n < 1:
        raise ValueError('n must be at least one')
    iterator = iter(iterable)
    while batch := tuple(islice(iterator, n)):
        if strict and len(batch) != n:
            raise ValueError('batched(): incomplete batch')
        yield batch

First, it calls iter on the iterable (let's assume the iterable is already an iterator). Then, this iterator is passed to islice on each iteration. islice expects an iterable, so it calls iter on the argument. This looks like reasonable code, but if the iterator changes its state when its __iter__ is called, it's not going to work as expected

[final geode]

https://github.com/python/cpython/issues/127809

[austere totem]

hey so im pretty new to Python and coding in general, is there something you would recommend to actually use it like projects i could do?

[winged sphinx]

Ask in #python-discussion , that's a better channel for this question.

[austere totem]

ok thanks

[boreal umbra]

Has there been any discussion of adding methods to the generator type, like map, filter, and limit?

[spark verge]

But there's already syntax support for map and filter

Also the Generator type is a bare minimum ABC, so adding methods would break it

We'd need to fix the whole collections.abc.Generator Vs types.GeneratorType issue first

[boreal umbra]

the builtin map and filter aren't part of syntax, but I see what you mean about generators being a protocol rather than an exact type.

[spark verge]

You don't need the built in map or filter

You just use generator expressions

[boreal umbra]

ah, you are right (including about the syntax part--my apologies)

[grave jolt]

I think a bigger issue is that other iterators won't automatically get that

You'd probably also want generator.map(lambda x: x * 2) to return something with map, filter and other methods, so map and filter objects now probably also get these methods. And then you'll have an inconsistency in the ecosystem as to which iterators support which helper methods

With that limitation, a better solution would probably be some kind of "pipe", so that you don't have to bolt on a map method to every iterator type

[spark verge]

And you'd want list.map to return a list and gen.map to return a generator and set.map to return a set maybe?

[grave jolt]

In fact, you can already do something like ```py
Pipe(generator) | (map, lambda x: x * 2) | (filter, lambda y: y > 0) | (limit, 69)

but realistically, just using generator expressions or loops will be the solution in Python

[rose schooner]

i had an idea

nvm

def f(XY or is_XY):
    print(XY, is_XY)

f(True)  # True True
f(False)  # False False

it's not as practical as when i first had it in mind

[dusk comet]

parameter aliases?

[rose schooner]

yup

i think it's better explained with these examples ```py
f(is_XY=True) # True True
f(XY=False) # False False

[grave jolt]

but why

[rose schooner]

i had an idea with py class Point: def __init__(coord0: Number, coord1: Number, XY or is_XY: bool = False): if is_XY: self.x = coord0 self.y = coord1 else: self.y = coord0 self.x = coord1 ... but i guess it could be better made with this ```py
class XYPoint:
def init(x: Number, y: Number):
self.x = x
self.y = y
...

class YXPoint(XYPoint):
def init(y: Number, x: Number):
super().init(x, y)

[grave jolt]

I mean, why do you want two different ways to name a parameter?

If you think the old name (in a library) is really bad, maybe it's better to deprecate it, which will require some extra runtime stuff

@deprecationlib.rename_param(new="is_xy", old="xy")
def __init__(self, c0: int, c1: int, /, *, is_xy: bool = True) -> None:

[rose schooner]

convenience

no idea

[rose schooner]

that's why i said this

[grave jolt]

it would be cooler to have an and

def launch_missiles(i_am_sure and really_sure):
    ...

launch_missiles(i_am_sure=True, really_sure=True)

[rose schooner]

hmm

parameters that are only allowed when both are given..?

[grave jolt]

no, they are just anded

for those of us activating weapons of mass destruction from IPython

[rose schooner]

okay wait

i had this problem with my Vector class where i wanted to support giving x and y OR giving magnitude and direction

[grave jolt]

I would split it into Vector(x, y) and Vector.polar(4206.9, pi)

[rose schooner]

that would be nicer, yes

[grave jolt]

I guess there are some functions that have groups of parameters that are either all provided or all not provided

but perhaps that is better replaced with a parameter object

[rose schooner]

unfortunately i already implemented manual argument parsing to support both

[grave jolt]

if you add enough of those groups to a function, it starts to get bloated

[rose schooner]

https://github.com/thatbirdguythatuknownot/sniplections/blob/main/vectors.py#L80-L150

yeah i guess there are better ways than this

[grave jolt]

💀

[quick snow]

Was reading PEP 768 and wondering whether this would be useful for something like memray, until I read the list of authors :)

[raven ridge]

It will! 🤣

[rose schooner]

more ideas ```py
f() => g()

instead of

(f(), g())[1]

also

f() =< g()
(f(), g())[0]

def add_(x)and(y):
return x + y

def combine_[T](*x: T)and[U](*y: U) -> list[tuple[T, U]]:
return list(zip(x, y))

[dusk comet]

1,2 - reminds me of a comma operator in C
3,4 - there was either a discussion or a pep about it, if I am not dreaming it up

[steel solstice]

iirc there was a time there was a way to not have to use stacklevel in warnings and just go to a frame with a name is that a thing or am i just misremembering?

[feral island]

I think that's a newly added feature

Oh not quite what you say, skip_file_prefixes (added in 3.12) was what I was thinking of https://docs.python.org/3.14/library/warnings.html#warnings.warn

[steel solstice]

ah that was indeed it

thank you

[opaque solar]

Hi

How can I learn code python

[crisp locust]

Tutorials

[winged sphinx]

Ask in #python-discussion

[deft kiln]

@opaque solar I can teach you if you want and I will help you to earn certificate from Harvard online CS50

[winged sphinx]

Please don't solicit people to contact you directly like this. If you want to help, see #1035199133436354600

[opaque solar]

Can you tell me how to get started from scratch learning python

[boreal umbra]

has str? ever been considered as an alternative to str | None?

[steel solstice]

In terms of default arguments?

[shrewd dew]

just curiouse guys the people that make ''rats'' that like logs your whole imformation like saved cookies and stuff idk i think its like a exe file they run is that extreamly hard to make?

[spark verge]

This isn't the right channel

[shrewd dew]

wich one is

[spark verge]

If you're not sure on a channel use #❓｜how-to-get-help

[feral island]

Yes. I think there was a discussion on the typing-sig mailing list

[boreal umbra]

What did people hate about it?

[feral island]

I haven't looked up the discussion, but new syntax is always a hard sell, and I don't like that it special-cases None too much

[verbal escarp]

what is necessary to formally get a PEP accepted?

[crisp locust]

That’s life with computer languages

[rose schooner]

the steering council thinking it's a good idea, i guess?

usually you'd precede PEP writing with a message to the Ideas category of discuss.python.org

[sour thistle]

the process is documented in https://peps.python.org/pep-0001/

[quick snow]

Get it to be a PEP: Be a core dev or have one sponsor it. Have it accepted: For most types of PEPs: Have steering council decision in support.

[verbal escarp]

it's quite fuzzy 🫤

[verbal escarp]

politics 😐

[merry bramble]

The Steering Council (at least in recent years) generally hasn't approved PEPs if it's clear that there's a majority (or even a significant minority) of core devs who are uncomfortable with the change. So I would say it's more than just "convince the steering council"; it's really "forge consensus among the core developer team as a whole"

[verbal escarp]

more politics sigh

okay

@rose schooner i understand why you don't write PEPs :p

[quick snow]

What kind of process do you have in mind that doesn't involve "politics"?

[crisp locust]

Why assume politics in technical subjects?

[verbal escarp]

i'm pretty sure that would derail the topic here, but let's say "focus on merrit of the idea, not the person"

[rose schooner]

if i were to write a PEP it would be about a great feature that isn't any of the toys i currently have amassed in cpython4

(some of them might be great features but i don't like making arrangements with the public in general)

[verbal escarp]

question is whether it might be better to get approval for a toy first and gradually make bolder proposals or make the boldest proposal first in hopes of getting less bold proposals approved instead..

if it were based on merrit of the idea instead of person, the order wouldn't matter, but alas..

[rose schooner]

these would be pretty difficult toys to justify :p ```pycon

x = []
((x=x) => x)() += 5,
x
[5]

oh right