Embedded Python
Kit comes with embedded Python. Regular CPython 3.7 is used with no modifications.
Kit initializes the Python interpreter before any extension is started. Each extension can then add their own folders (or subfolders) to the sys.path
(using [[python.module]]
definitions). By subclassing as IExt
, extensions get an entry point into Python code. They can then execute any code, import other extensions, use any API they provide. More info: Extensions.
For most applications this is all you need to know about Python in Kit. Examining sys.path
at runtime is the most common way to debug most issues. This doc provides more advanced details on Python integration.
Hello Python
Run > kit.exe --exec your_script.py
to run your script using Kit Python.
Using system Python
When the Python interpreter is initialized, system-defined environment variables (like PYTHONHOME
, PYTHONPATH
) are ignored. Instead, the following setting is used for python home:
/plugins/carb.scripting-python.plugin/pythonHome
instead of PYTHONHOME
Note
You can find default values for this setting in kit-core.json
file.
To use a system-level Python installation, override PYTHONHOME
, e.g.: --/plugins/carb.scripting-python.plugin/pythonHome="C:\Users\bob\AppData\Local\Programs\Python\Python37"
.
Changing PYTHONHOME
won’t change the loaded Python library. This is platform specific, but for instance on Windows, Kit is linked with python.dll
and loads the one that is in the package using standard dll search rules. However, the standard library, site-packages
, and everything else will be used from the specified python path.
Add extra search paths
To add search paths (to sys.path
), the /app/python/extraPaths
setting can be used. For example:
> kit.exe --/app/python/extraPaths/0="C:/temp"
or in a kit file:
[settings]
app.python.extraPaths = ["C:/temp"]
To summarize, those are all the methods to extend sys.path
:
Create new extension with
[python.module]
definitions (recommended).Explicitly in python code:
sys.path.append(...)
The
/app/python/extraPaths
setting.
Other Python Configuration Tweaks
Most python configuration variables can be changed using following settings:
config variable |
python flag documentation |
---|---|
|
|
|
|
|
|
|
|
|
|
|
|
|
Using numpy
, Pillow
etc.
Kit comes with omni.kit.pip_archive
extension which has few popular Python modules bundled into it. Have a look inside of it on filesystem.
After this extension is started you can freely do import numpy
. Declare a dependency on this extension in your extension, or enable it by any other means to use any of them.
E.g.: run > kit.exe --enable omni.kit.pip_archive --exec use_numpy.py
to run your script that can import and use numpy
.
Using Anaconda environment
As a starting point change PYTHONHOME
setting described above to point to Anaconda environment: --/plugins/carb.scripting-python.plugin/pythonHome="C:/Users/bob/anaconda3/envs/py37"
. It is known to work for some packages and fail for others, on a case by case basis.
Using other packages from pip
For most Python packages (installed with any package manager or locally developed) it is enough to add them to the search path (sys.path
). That makes them discoverable by the python import system. Any of the methods described above can be used for that.
Alternatively, Kit has the omni.kit.pipapi
extension to install modules from the pip
package manager at runtime. It will check if the package is not available, and will try to pip install it and cache it. Example of usage: omni.kit.pipapi.install("some_package")
. After that call, import the installed package.
Enabling the omni.kit.pipapi
extension will allow specification of pip dependencies by extensions loaded after it. Refer to omni.kit.pipapi
doc.
At build-time, any Python module can be packaged into any extension, including packages from pip. That can be done using other Python installations or kit Python. This is the recommended way, so that when an extension is downloaded and installed, it is ready to use. There is also no requirement for connectivity to public registries, and no runtime cost during installation.
Why do some native Python modules not work in Kit?
It is common for something that works out of the box as-installed from pip or Anaconda not to work in Kit. Or vice versa, the Kit Python module doesn’t load outside of Kit.
For pure Python modules (only *.py
files), finding the root cause might be a matter of following import errors. However, when it involves loading native Python modules (*.pyd
files on Windows and *.so
files on Linux), errors are often not really helpful.
Native Python modules are just regular OS shared libraries, with a special C API that Python looks for. They also are often implicitly linked with other libraries. When loaded, they might not be able to find other libraries, or be in conflict with already loaded libraries. Those issues can be debugged as any other library loading issue, specific to the OS. Some examples are:
Exploring
PATH
/LD_LIBRARY_PATH
env vars.Exploring libraries that are already loaded by the process.
Using tools like Dependency Walker.
Trying to isolate the issue, by loading in a simpler or more similar environment.
Kit doesn’t do anything special in this regard, and can be treated as just another instance of Python, with a potentially different set of loaded modules.
Running Kit from Python
Normally the kit.exe
process starts and loads an embedded Python library. Kit provides Python bindings to its core runtime components. This allows you to start Python, and then start Kit from that Python.
It is an experimental feature, and not used often. An example can be found within the Kit package: example.pythonapp.bat
.
Differences from running normally:
A different Python library file is used (different
python.dll
).There may be some GIL implications, because the call stack is different.
Allows explicit control over the update loop.