Embedded Python¶
Kit comes with embedded Python. Regular CPython 3.7 is used with no modifications.
Kit initializes Python interpreter before any extension is started. Then each extension can add own folder (or subfolder) to sys.path
(using [[python.module]]
definitions). By defining IExt
subclass 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 Python interpreter initialized system env vars (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 system 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 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 standard library, site-packages
and everything else will be used from specified python path.
Add extra search paths¶
To add search path (to sys.path
) /app/python/extraPaths
setting can be used. For example:
> kit.exe --/app/python/extraPaths/0="C:/temp"
or in 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(...)
/app/python/extraPaths
setting
Other Python Configuration Tweaks¶
Most python ‘configuration variables <https://docs.python.org/3/c-api/init.html#global-configuration-variables>’_ can be changed using following settings:
/plugins/carb.scripting-python.plugin/Py_VerboseFlag
Py_VerboseFlag/plugins/carb.scripting-python.plugin/Py_QuietFlag
Py_QuietFlag/plugins/carb.scripting-python.plugin/Py_NoSiteFlag
Py_NoSiteFlag/plugins/carb.scripting-python.plugin/Py_IgnoreEnvironmentFlag
Py_IgnoreEnvironmentFlag/plugins/carb.scripting-python.plugin/Py_NoUserSiteDirectory
Py_NoUserSiteDirectory/plugins/carb.scripting-python.plugin/Py_UnbufferedStdioFlag
Py_UnbufferedStdioFlag/plugins/carb.scripting-python.plugin/Py_IsolatedFlag
Py_IsolatedFlag
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 freely do import numpy
. Declare dependency on this extension in your extension or enable it by 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 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 search path (sys.path
). That makes them discoverable by import system. Any of methods described above can be used for that.
Alternatively, Kit has omni.kit.pipapi
extension to install modules from 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 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 installation or kit Python. This is recommended way, so that when extension is downloaded and installed it is ready to use. There is no requirement on connectivity to public registries and no runtime cost during installation.
Why some native Python modules don’t work in Kit?¶
It is often that something that works out of the box installed from pip or Anaconda doesn’t work in Kit. Or vice versa 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 not really helpful.
Native Python modules are just regular OS shared libraries, with special C API that Python looks for. They also 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 issues, specific to 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 issue, by loading in a simpler or more similar environment.
Kit doesn’t do anything special in that regard and can be treated as just another instance of Python with potentially different set of loaded modules.
Running Kit from Python¶
Normally kit.exe
process starts and loads embedded Python library. Kit provides Python bindings to it’s core runtime components. That allows to start Python and then start Kit from Python.
It is an experimental feature and not used often. Example of that can be found within Kit package: example.pythonapp.bat
.
Difference from regular run:
Actually another Python library us used (different
python.dll
).That may have some GIL implications, because call stack looks differently.
Allows explicit control over update loop.