anaconda/python 3.11 on all supported platforms

AntonOfTheWoods commented 2 years ago

Checklist

[X] I added a descriptive title (including the requested package name, version, and platform).
[X] I searched the defaults channels (Python and R) and did not find the package I'm requesting.
[X] I searched through existing issues and did not find a duplicate request for this package.

Request type

[ ] New (i.e. no versions of this package currently exist on repo.anaconda.com)
[X] Update (i.e. older versions of this package exist and I'm requesting an updated version)
[ ] Rebuild (i.e. something is wrong with an existing version of this package and it needs to be rebuilt)

Name

Python

Version

3.11

Platform(s)

linux-64 (and presumably others)

Location of source code

https://github.com/python/cpython

Conda-Forge feedstock

https://github.com/conda-forge/python-feedstock

Reason for this request

It appears as though 3.10.0 was released (built anyway) a day or so after release on python.org but there doesn't appear to be anything available yet, even though the feedstock updated within a day or so.

(OPTIONAL) Additional information

No response

alexlang74 commented 1 year ago

Hi @AntonOfTheWoods , is there a timeline to integrate this PR? We'd like to start exploring with Python 3.11... Thanks!

wenjuno commented 1 year ago

This is puzzling. Python 3.11.1 has been released but we don't have any Python 3.11 in anaconda channel yet. I'm curious what makes Python 3.11 so difficult to support? Is it because there are much more 3rd party libraries you will need to build with Python 3.11? If so, can we build those libraries gradually?

AntonOfTheWoods commented 1 year ago

@alexlang74 , I'm just a user. There is nothing difficult about implementing this. In fact you can just add condaforge and use that instead. The maintainers don't need this and obviously don't care that we do, or at least don't have the free time to care! It's that simple.

CtrlC-Root commented 1 year ago

This seems to be a blocker for PyTorch vision library Python 3.11 support.

johnnynunez commented 1 year ago

I'm just a user. There is nothing difficult about implementing this. In fact you can just add condaforge and use that instead. The maintainers don't need this and obviously don't care that we do, or at least don't have the free time to care! It's that simple.

dependencies. Conda depends about a lot libraries and conda-forge can be not stable.

AntonOfTheWoods commented 1 year ago

Sure, I am just helping to provide support for recent python versions to a FOSS project that uses anaconda. I would never use it myself, for this very reason. I will suggest that we remove anaconda completely but that takes time and effort...

scdub commented 1 year ago

@AntonOfTheWoods I think this issue can be kept open. I understand the sentiment behind the statement "The maintainers don't need this and obviously don't care that we do, or at least don't have the free time to care! It's that simple" but in the reality is more nuanced than that. While PyPI packages have to maintain their own individual package compatibilities with a new Python release, Anaconda really is a complete ecosystem of packages which cross language boundaries. In many ways is better thought of in the way that Linux distributions are: an intermediate layer between the packages themselves and the system architecture. This implies trade-offs in the speed to build out distributions, and just shipping CPython by itself solves few problems because the ecosystem is built around targeting high-value toolchains like deep learning workflows which encapsulate a significant set of package requirements. I don't work on Anaconda directly but do maintain a set of package for conda and can attest to the level of effort this can entail. Conda-forge is great as a rolling release, automate everything environment but there are trade-offs in ABI compatibility and other areas that Anaconda makes.

AntonOfTheWoods commented 1 year ago

@scdub , so you think they do have the time and still don't even address the issue (here, for example)?

maxnoe commented 1 year ago

Please re-open, it is not solved and important to many people (See the link above to pytorch/vision).

This is currently also a blocker for the next numba release, see

https://github.com/numba/numba/issues/8304

and

https://github.com/numba/llvmlite/pull/869

AntonOfTheWoods commented 1 year ago

If we are completely honest, I think this is a perfect example of why anaconda is a horrible solution for what many people are using it for. If you need a little reactivity in updates of python versions, then you need something else.

I'll reopen and unsubscribe, as I don't need this anymore...

cbouss commented 1 year ago

Thank you for your patience. At Anaconda, we provide a cohesive distribution across the 7 architectures and 4 python versions we support, which does incur some delay in releasing. We are currently working on the build out of the distribution for python 3.11.

johnnynunez commented 1 year ago

Thank you for your patience. At Anaconda, we provide a cohesive distribution across the 7 architectures and 4 python versions we support, which does incur some delay in releasing. We are currently working on the build out of the distribution for python 3.11.

When do you expect a release? At least to be able to create environments. conda create -n py311 python=3.11

cbouss commented 1 year ago

Thank you for your patience. At Anaconda, we provide a cohesive distribution across the 7 architectures and 4 python versions we support, which does incur some delay in releasing. We are currently working on the build out of the distribution for python 3.11.

When do you expect a release? At least to be able to create environments. conda create -n py311 python=3.11

Hi Johnny,

Python 3.11 along with setuptools and pip have been pushed to defaults today. You can now create environments with conda create -n py311 python=3.11 The full release of the distribution is targeted for this quarter.

johnnynunez commented 1 year ago

Thank you for your patience. At Anaconda, we provide a cohesive distribution across the 7 architectures and 4 python versions we support, which does incur some delay in releasing. We are currently working on the build out of the distribution for python 3.11.

When do you expect a release? At least to be able to create environments. conda create -n py311 python=3.11

Hi Johnny,

Python 3.11 along with setuptools and pip have been pushed to defaults today. You can now create environments with conda create -n py311 python=3.11 The full release of the distribution is targeted for this quarter.

@cbouss some dependencies such as numpy and scipy still need to be updated. https://anaconda.org/main/numpy https://anaconda.org/main/scipy problem: https://github.com/pytorch/vision/pull/7108

ZupoLlask commented 1 year ago

scipy is already updated!

burnpanck commented 1 year ago

@ZupoLlask: Is it? Obviously both numpy and scipy both supported python 3.11 within days after it's official release. But that doesn't mean that the anaconda distribution is also shipping python 3.11 binaries of those libraries, and that is what is needed to make a full anaconda python 3.11 release. If you follow the links @johnnynunez is pointing out, you will see that, to date, there are still no python 3.11 binaries available on anaconda for either numpy or scipy.

ZupoLlask commented 1 year ago

@burnpanck Sorry, you are absolutely right. There are no python 3.11 binaries yet for those packages...

RNarayan73 commented 1 year ago

Due to be released on March 14, fingers crossed! https://www.anaconda.com/blog/upcoming-releases-anaconda-distribution-2023-03-and-beyond#:~:text=Python%203.11%20support%20will%20be,of%20Anaconda's%20biggest%20releases%20yet!

bert-pluymers-felfel commented 1 year ago

Do we have reasons to believe this release date is still accurate?

The blog post is from Feb 22 and it also wrongly (I guess) mentions Python 3.10 (rather than 3.11) support in a few places...

cbouss commented 1 year ago

Hello, and thank you for your patience.

The 2023.03 Anaconda Distribution installer release will indeed only come for Python 3.8, 3.9 and 3.10. Not all required packages were available to make a 3.11 release this time.

However we did push today most 3.11 packages on our main channel (including numpy and scipy).

johnnynunez commented 1 year ago

Hello, and thank you for your patience.

The 2023.03 Anaconda Distribution installer release will indeed only come for Python 3.8, 3.9 and 3.10. Not all required packages were available to make a 3.11 release this time.

However we did push today most 3.11 packages on our main channel (including numpy and scipy).

Could you upgrade to 3.11.2 on anaconda channel? The version that exists now it's 3.11 and has some important bugs

cbouss commented 1 year ago

Hello, and thank you for your patience. The 2023.03 Anaconda Distribution installer release will indeed only come for Python 3.8, 3.9 and 3.10. Not all required packages were available to make a 3.11 release this time. However we did push today most 3.11 packages on our main channel (including numpy and scipy).

Could you upgrade to 3.11.2 on anaconda channel? The version that exists now it's 3.11 and has some important bugs

Hello, 3.11.2 is now uploaded to anaconda default channel.

johnnynunez commented 1 year ago

Hello, and thank you for your patience. The 2023.03 Anaconda Distribution installer release will indeed only come for Python 3.8, 3.9 and 3.10. Not all required packages were available to make a 3.11 release this time. However we did push today most 3.11 packages on our main channel (including numpy and scipy).

Could you upgrade to 3.11.2 on anaconda channel? The version that exists now it's 3.11 and has some important bugs

Hello, 3.11.2 is now uploaded to anaconda default channel.

thanks!

johnnynunez commented 1 year ago

Hello, and thank you for your patience. The 2023.03 Anaconda Distribution installer release will indeed only come for Python 3.8, 3.9 and 3.10. Not all required packages were available to make a 3.11 release this time. However we did push today most 3.11 packages on our main channel (including numpy and scipy).

Could you upgrade to 3.11.2 on anaconda channel? The version that exists now it's 3.11 and has some important bugs

Hello, 3.11.2 is now uploaded to anaconda default channel.

python 3.11.3 is out

johnnynunez commented 1 year ago

Python 3.11 support added via https://github.com/conda/conda/pull/12256, this will be included in the May release of conda 23.5.0

ContinuumIO / anaconda-issues