Feature request: Change a Dependency Package Version During Package Initiation

A lot of the Python code we use in production are used directly as imports in other python distributions (such as the python comes with the finite element software Abaqus and MSC Marc), many packages (such as matplotlib, numpy) that may have varying versioned dependencies. I was wondering if this could be expanded to allow a version to be set within a package and have that propagate to all modules in that package. For example in the root init.py if I set multiversion(tornado, 2.2.1) then all modules in that package will use tornado 2.2.1 when I import tornado. See a relevant issue on github: https://github.com/mitsuhiko/multiversion/issues/1 Thank you! Qiang

A lot of the Python code we use in production are used directly as imports in other python distributions (such as the python comes with the finite element software Abaqus and MSC Marc), many packages (such as matplotlib, numpy) that may have varying versioned dependencies, which makes it is a pain to get all these dependencies met in those close source python distributions. The work around is to let a version to be set within a package and have that propagate to all modules in that package. For example in the root init.py if I set tonado version to be 2.2.1 then all modules in that package will use tornado 2.2.1 when I import tornado. See a relevant issue of a similar package on github: https://github.com/mitsuhiko/multiversion/issues/1 Thank you! Qiang

Thanks for the idea but there are currently no plans to support such a feature. If you would like to see it then you will need to write a PEP with a proof-of-concept to demonstrate how you would expect such a feature to work. On Fri., May 17, 2019, 07:55 Q via Python-Dev, <python-dev@python.org> wrote:
A lot of the Python code we use in production are used directly as imports in other python distributions (such as the python comes with the finite element software Abaqus and MSC Marc), many packages (such as matplotlib, numpy) that may have varying versioned dependencies.
I was wondering if this could be expanded to allow a version to be set within a package and have that propagate to all modules in that package. For example in the root init.py if I set multiversion(tornado, 2.2.1) then all modules in that package will use tornado 2.2.1 when I import tornado.
See a relevant issue on github: https://github.com/mitsuhiko/multiversion/issues/1
Thank you! Qiang
_______________________________________________ Python-Dev mailing list Python-Dev@python.org https://mail.python.org/mailman/listinfo/python-dev Unsubscribe: https://mail.python.org/mailman/options/python-dev/brett%40python.org

This sounds exactly like what people used to do with eggs. You could have multiple versions of a package on the path as eggs and then require a version at runtime. The approach has problems. Ruby also abandoned a strategy where random app code depends on package management code at runtime. One better strategy is to set up a python path in a wrapper script. On Fri, May 17, 2019, 11:27 Brett Cannon <bcannon@gmail.com> wrote:
Thanks for the idea but there are currently no plans to support such a feature. If you would like to see it then you will need to write a PEP with a proof-of-concept to demonstrate how you would expect such a feature to work.
On Fri., May 17, 2019, 07:55 Q via Python-Dev, <python-dev@python.org> wrote:
A lot of the Python code we use in production are used directly as imports in other python distributions (such as the python comes with the finite element software Abaqus and MSC Marc), many packages (such as matplotlib, numpy) that may have varying versioned dependencies.
I was wondering if this could be expanded to allow a version to be set within a package and have that propagate to all modules in that package. For example in the root init.py if I set multiversion(tornado, 2.2.1) then all modules in that package will use tornado 2.2.1 when I import tornado.
See a relevant issue on github: https://github.com/mitsuhiko/multiversion/issues/1
Thank you! Qiang
_______________________________________________ Python-Dev mailing list Python-Dev@python.org https://mail.python.org/mailman/listinfo/python-dev Unsubscribe: https://mail.python.org/mailman/options/python-dev/brett%40python.org
_______________________________________________ Python-Dev mailing list Python-Dev@python.org https://mail.python.org/mailman/listinfo/python-dev Unsubscribe: https://mail.python.org/mailman/options/python-dev/dholth%40gmail.com

From the viewpoint of the package user, people don't want to change every import to "require a version at runtime". If "set up a python path in a wrapper script" is a better strategy, can you please give an example for the following use case:
The abaqus python distribution has libA 1.0, the external package to be installed is LibB, which requires libA2.0, and we don't want to change every import in LibB to be import libA 1.0, because the source code may have not been provided. In emacs, you can always advice a command to change its behaviour, which makes it very user friendly, and that what all the following suggestion is about. I have no idea at all how it could be implemented though. https://github.com/mitsuhiko/multiversion/issues/1 "The work around is to let a version to be set within a package and have that propagate to all modules in that package. For example in the root init.py if I set tonado version to be 2.2.1 then all modules in that package will use tornado 2.2.1 when I import tornado." ---- 在 星期五, 2019-05-17 23:38:55 Daniel Holth <dholth@gmail.com> 撰写 ---- This sounds exactly like what people used to do with eggs. You could have multiple versions of a package on the path as eggs and then require a version at runtime. The approach has problems. Ruby also abandoned a strategy where random app code depends on package management code at runtime. One better strategy is to set up a python path in a wrapper script. On Fri, May 17, 2019, 11:27 Brett Cannon <mailto:bcannon@gmail.com> wrote: Thanks for the idea but there are currently no plans to support such a feature. If you would like to see it then you will need to write a PEP with a proof-of-concept to demonstrate how you would expect such a feature to work. On Fri., May 17, 2019, 07:55 Q via Python-Dev, <mailto:python-dev@python.org> wrote: A lot of the Python code we use in production are used directly as imports in other python distributions (such as the python comes with the finite element software Abaqus and MSC Marc), many packages (such as matplotlib, numpy) that may have varying versioned dependencies. I was wondering if this could be expanded to allow a version to be set within a package and have that propagate to all modules in that package. For example in the root init.py if I set multiversion(tornado, 2.2.1) then all modules in that package will use tornado 2.2.1 when I import tornado. See a relevant issue on github: https://github.com/mitsuhiko/multiversion/issues/1 Thank you! Qiang _______________________________________________ Python-Dev mailing list mailto:Python-Dev@python.org https://mail.python.org/mailman/listinfo/python-dev Unsubscribe: https://mail.python.org/mailman/options/python-dev/brett%40python.org _______________________________________________ Python-Dev mailing list mailto:Python-Dev@python.org https://mail.python.org/mailman/listinfo/python-dev Unsubscribe: https://mail.python.org/mailman/options/python-dev/dholth%40gmail.com

Sorry, there is an error in my last post. It's corrected as: The abaqus python distribution has libA 1.0, the external package to be installed is LibB, which requires libA2.0, and we don't want to change every import in LibB to be import /somewhere/libA2.0, because the source code may have not been provided. ---- 在 星期六, 2019-05-18 10:26:04 Q <qiang.fang@zoho.com.cn> 撰写 ----
From the viewpoint of the package user, people don't want to change every import to "require a version at runtime". If "set up a python path in a wrapper script" is a better strategy, can you please give an example for the following use case:
The abaqus python distribution has libA 1.0, the external package to be installed is LibB, which requires libA2.0, and we don't want to change every import in LibB to be import libA 1.0, because the source code may have not been provided. In emacs, you can always advice a command to change its behaviour, which makes it very user friendly, and that what all the following suggestion is about. I have no idea at all how it could be implemented though. https://github.com/mitsuhiko/multiversion/issues/1 "The work around is to let a version to be set within a package and have that propagate to all modules in that package. For example in the root init.py if I set tonado version to be 2.2.1 then all modules in that package will use tornado 2.2.1 when I import tornado." ---- 在 星期五, 2019-05-17 23:38:55 Daniel Holth <dholth@gmail.com> 撰写 ---- This sounds exactly like what people used to do with eggs. You could have multiple versions of a package on the path as eggs and then require a version at runtime. The approach has problems. Ruby also abandoned a strategy where random app code depends on package management code at runtime. One better strategy is to set up a python path in a wrapper script. On Fri, May 17, 2019, 11:27 Brett Cannon <mailto:bcannon@gmail.com> wrote: Thanks for the idea but there are currently no plans to support such a feature. If you would like to see it then you will need to write a PEP with a proof-of-concept to demonstrate how you would expect such a feature to work. On Fri., May 17, 2019, 07:55 Q via Python-Dev, <mailto:python-dev@python.org> wrote: A lot of the Python code we use in production are used directly as imports in other python distributions (such as the python comes with the finite element software Abaqus and MSC Marc), many packages (such as matplotlib, numpy) that may have varying versioned dependencies. I was wondering if this could be expanded to allow a version to be set within a package and have that propagate to all modules in that package. For example in the root init.py if I set multiversion(tornado, 2.2.1) then all modules in that package will use tornado 2.2.1 when I import tornado. See a relevant issue on github: https://github.com/mitsuhiko/multiversion/issues/1 Thank you! Qiang _______________________________________________ Python-Dev mailing list mailto:Python-Dev@python.org https://mail.python.org/mailman/listinfo/python-dev Unsubscribe: https://mail.python.org/mailman/options/python-dev/brett%40python.org _______________________________________________ Python-Dev mailing list mailto:Python-Dev@python.org https://mail.python.org/mailman/listinfo/python-dev Unsubscribe: https://mail.python.org/mailman/options/python-dev/dholth%40gmail.com

It probably makes the most sense to formally outline all of the behavioral changes, expectations, and constraints you are asking for as Brett said because dependency and package management are tricky and it's not clear (to me at least) what the request is, what problem is being encountered, and how it is proposed that it be resolved. First glance makes it sound a bit like the a prerequisite for this request would be allowing multiple versions of the same package to be installed in a given python environment to begin with, or else some runtime hacking of paths would be necessary. I may be misunderstanding, but that's why a formal approach for something like this might make some sense Dan Ryan gh: @techalchemy // e: dan@danryan.co From: Q [mailto:qiang.fang@zoho.com.cn] Sent: Friday, May 17, 2019 10:32 PM To: Daniel Holth Cc: Brett Cannon; Python-Dev Subject: Re: [Python-Dev] Feature request: Change a Dependency Package Version During Package Initiation Sorry, there is an error in my last post. It's corrected as: The abaqus python distribution has libA 1.0, the external package to be installed is LibB, which requires libA2.0, and we don't want to change every import in LibB to be import /somewhere/libA2.0, because the source code may have not been provided. ---- 在 星期六, 2019-05-18 10:26:04 Q <qiang.fang@zoho.com.cn> 撰写 ----
From the viewpoint of the package user, people don't want to change every import to "require a version at runtime". If "set up a python path in a wrapper script" is a better strategy, can you please give an example for the following use case: The abaqus python distribution has libA 1.0, the external package to be installed is LibB, which requires libA2.0, and we don't want to change every import in LibB to be import libA 1.0, because the source code may have not been provided. In emacs, you can always advice a command to change its behaviour, which makes it very user friendly, and that what all the following suggestion is about. I have no idea at all how it could be implemented though. https://github.com/mitsuhiko/multiversion/issues/1 "The work around is to let a version to be set within a package and have that propagate to all modules in that package. For example in the root init.py if I set tonado version to be 2.2.1 then all modules in that package will use tornado 2.2.1 when I import tornado."
---- 在 星期五, 2019-05-17 23:38:55 Daniel Holth <dholth@gmail.com> 撰写 ---- This sounds exactly like what people used to do with eggs. You could have multiple versions of a package on the path as eggs and then require a version at runtime. The approach has problems. Ruby also abandoned a strategy where random app code depends on package management code at runtime. One better strategy is to set up a python path in a wrapper script. On Fri, May 17, 2019, 11:27 Brett Cannon <bcannon@gmail.com> wrote: Thanks for the idea but there are currently no plans to support such a feature. If you would like to see it then you will need to write a PEP with a proof-of-concept to demonstrate how you would expect such a feature to work. On Fri., May 17, 2019, 07:55 Q via Python-Dev, <python-dev@python.org> wrote: A lot of the Python code we use in production are used directly as imports in other python distributions (such as the python comes with the finite element software Abaqus and MSC Marc), many packages (such as matplotlib, numpy) that may have varying versioned dependencies. I was wondering if this could be expanded to allow a version to be set within a package and have that propagate to all modules in that package. For example in the root init.py if I set multiversion(tornado, 2.2.1) then all modules in that package will use tornado 2.2.1 when I import tornado. See a relevant issue on github: https://github.com/mitsuhiko/multiversion/issues/1 Thank you! Qiang

I suspect this discussion would be better moved to python-ideas, since it seems clear there needs to be some work on determining the exact requirements On Sat, May 18, 2019 at 4:03 PM Dan Ryan <dan@danryan.co> wrote:
It probably makes the most sense to formally outline all of the behavioral changes, expectations, and constraints you are asking for as Brett said because dependency and package management are tricky and it's not clear (to me at least) what the request is, what problem is being encountered, and how it is proposed that it be resolved.
First glance makes it sound a bit like the a prerequisite for this request would be allowing multiple versions of the same package to be installed in a given python environment to begin with, or else some runtime hacking of paths would be necessary. I may be misunderstanding, but that's why a formal approach for something like this might make some sense
Dan Ryan gh: @techalchemy // e: dan@danryan.co
From: Q [mailto:qiang.fang@zoho.com.cn] Sent: Friday, May 17, 2019 10:32 PM To: Daniel Holth Cc: Brett Cannon; Python-Dev Subject: Re: [Python-Dev] Feature request: Change a Dependency Package Version During Package Initiation
Sorry, there is an error in my last post. It's corrected as: The abaqus python distribution has libA 1.0, the external package to be installed is LibB, which requires libA2.0, and we don't want to change every import in LibB to be import /somewhere/libA2.0, because the source code may have not been provided.
---- 在 星期六, 2019-05-18 10:26:04 Q <qiang.fang@zoho.com.cn> 撰写 ----
From the viewpoint of the package user, people don't want to change every import to "require a version at runtime". If "set up a python path in a wrapper script" is a better strategy, can you please give an example for the following use case: The abaqus python distribution has libA 1.0, the external package to be installed is LibB, which requires libA2.0, and we don't want to change every import in LibB to be import libA 1.0, because the source code may have not been provided. In emacs, you can always advice a command to change its behaviour, which makes it very user friendly, and that what all the following suggestion is about. I have no idea at all how it could be implemented though. https://github.com/mitsuhiko/multiversion/issues/1 "The work around is to let a version to be set within a package and have that propagate to all modules in that package. For example in the root init.py if I set tonado version to be 2.2.1 then all modules in that package will use tornado 2.2.1 when I import tornado."
---- 在 星期五, 2019-05-17 23:38:55 Daniel Holth <dholth@gmail.com> 撰写 ----
This sounds exactly like what people used to do with eggs. You could have multiple versions of a package on the path as eggs and then require a version at runtime. The approach has problems. Ruby also abandoned a strategy where random app code depends on package management code at runtime.
One better strategy is to set up a python path in a wrapper script.
On Fri, May 17, 2019, 11:27 Brett Cannon <bcannon@gmail.com> wrote:
Thanks for the idea but there are currently no plans to support such a feature. If you would like to see it then you will need to write a PEP with a proof-of-concept to demonstrate how you would expect such a feature to work.
On Fri., May 17, 2019, 07:55 Q via Python-Dev, <python-dev@python.org> wrote:
A lot of the Python code we use in production are used directly as imports in other python distributions (such as the python comes with the finite element software Abaqus and MSC Marc), many packages (such as matplotlib, numpy) that may have varying versioned dependencies.
I was wondering if this could be expanded to allow a version to be set within a package and have that propagate to all modules in that package. For example in the root init.py if I set multiversion(tornado, 2.2.1) then all modules in that package will use tornado 2.2.1 when I import tornado.
See a relevant issue on github: https://github.com/mitsuhiko/multiversion/issues/1
Thank you! Qiang
_______________________________________________ Python-Dev mailing list Python-Dev@python.org https://mail.python.org/mailman/listinfo/python-dev Unsubscribe: https://mail.python.org/mailman/options/python-dev/steve%40holdenweb.com
participants (5)
-
Brett Cannon
-
Dan Ryan
-
Daniel Holth
-
Q
-
Steve Holden