19

From the documentation:

The embedded distribution is a ZIP file containing a minimal Python environment.

Sounds great! The 64-bit Windows embedded v3.6.5 of Python is only 13MB. As an alternative to compiling, I would like to zip some python scripts together with the minimum needed to run them on a Win10 machine that doesn't have Python installed.

Now, I almost always need to import additional packages to provide functionality. But I can't determine how I should do this if I want to send out a python script together with this embedded version of Python. For example, if my script uses numpy, how can I include that library in this "embed?" I.e., so that on any Win10 machine I can unzip the single deployment file and immediately execute my scripts?

(A regular pip install numpy appears to create a Lib subdirectory that's over 50MB! But for an "embedded" deployment I don't need any support for debugging or whatever else is encompassed in that mass of files.)

feetwet
  • 2,914
  • 7
  • 41
  • 80
  • Is there a reason you can't use a `requirements.txt` then have them run `pip install -r requirements.txt` before they execute the script? – Aaron N. Brock Apr 09 '18 at 16:42
  • @AaronN.Brock: The documentation warns: *Using pip to manage dependencies as for a regular Python installation is not supported with this distribution, though with some care it may be possible to include and use pip for automatic updates.* I guess I could give it a try if I'm reading too far into that warning.... – feetwet Apr 09 '18 at 16:45
  • *modulefinder* might come in handy. Check https://stackoverflow.com/questions/39539089/what-files-are-required-for-py-initialize-to-run/39541474#39541474 to see how I used it. – CristiFati Apr 09 '18 at 17:00
  • Related: [pip with embedded python](https://stackoverflow.com/questions/42666121/pip-with-embedded-python) – idbrii Jul 31 '21 at 01:59

3 Answers3

9

There is a way to extend Python embedded installation. I managed to create Flask-ready package, that I can just unzip on target machine and run code. The trick is to install EXACT same python version (normal full blown python) as your target embedded small python. Not only version but x86, x64 has to match as well.

Then install modules from pip on normal python, go to NormalPython\Lib\site-packages and copy all new files that appear after installing the package to EmbeddedPython\Lib finally add Lib and Lib\site-packages to pythonXX._pth inside Embedded python folder.

It's extremely important to fully test your application in case you miss some package. Also this would not work for packages that also add .exe to Scripts folder. You could still probably copy the exe's to Script folder and add it to path which could do the trick.

Zhu Weiji
  • 57
  • 8
Michał Rawluk
  • 133
  • 1
  • 7
  • How would you do this without ._pth files given their [potential deprecation in Python 3.8](https://bugs.python.org/issue33944)? – Todd Feb 12 '19 at 03:30
  • [I found a more explicit script-based explanation](https://gist.github.com/rgl/72b60fc12f91c93c0d286d56ae19e324) that seems to follow this pattern, but zips up the dependencies. (It also uses _pth.) – idbrii Jul 20 '21 at 18:42
  • if you're not sure what file you need to copy into the embedded installation: import then .__file__ should do the trick – Tom Grundy Nov 19 '21 at 23:04
  • @Todd - We're now well into 3.10 and still no sign of deprecating _pth files. That discussion, as well as the resulting [PEP](https://www.python.org/dev/peps/pep-0648/) (which happened to be rejected anyway), never seemed to be focused on getting rid of those files entirely anyway (other than a few heat-of-the-moment posts at the beginning of the thread). Their original intended use, to allow adjustment to `sys.path`, is something that the vast majority of devs still think is a good idea, and definitely isn't going away any time soon, even if the arbitrary code execution does get removed. – John Y Jan 29 '22 at 02:07
2

There is a way to extend Python embedded installation. The trick is to install the same python version( I will call it NormalPython) as your target embedded python(I will call it EmbeddedPython). The version as well as architecture has to match exactly.

You then install modules from pip on NormalPython. You can find pip on NormalPython\Scripts.

Go to NormalPython\Lib\site-packages and copy all files that appear after installing whatever you want to install through pip to EmbeddedPython\Lib\

Then add Lib to pythonXX._pth inside Embedded python folder.

This worked for me on windows 10 after downloading the latest python embed and python install through https://www.python.org/downloads/windows/ . I used Michal Rawluk's answer to do it, but it was somewhat hard to understand so I am trying to explain it a bit differently.

Julio Peguero
  • 132
  • 4
1

There is a similar trick.

  1. Install pip in the embedded version:

    curl -sSL https://bootstrap.pypa.io/get-pip.py -o get-pip.py

    python get-pip.py

  2. Edit pythonXX._pth by adding

    Lib

    Lib\site-packages

  3. Use pip to install other packages

Mike
  • 11
  • 1
  • 1
    you might want to make the reference to "similar trick" more precise. Are you referring to another answer? Then you can note "trick similar to answer of xyz" – Christopher Hamkins Oct 21 '21 at 09:41