Crash on «import lightgbm» if «--python-flag=no_docstrings»

Let’s consider sample program using LightGBM and nuitka call for compilation on Linux:

https://github.com/belonesox/nuitka-light-gdm-troubles- branch «no_docstrings»

• ./reproduce-me.sh to (re)install virtual environment for install/build/run compiled

Running results we get crush on «import lightgbm»

Traceback (most recent call last):
  File "/home/stas/projects/nuitka-wtf/nuitka-light-gdm-troubles/test-lightgbm.dist/test-lightgbm.py", line 7, in <module>
  File "<frozen importlib._bootstrap>", line 1027, in _find_and_load
  File "<frozen importlib._bootstrap>", line 1006, in _find_and_load_unlocked
  File "<frozen importlib._bootstrap>", line 688, in _load_unlocked
  File "/home/stas/projects/nuitka-wtf/nuitka-light-gdm-troubles/test-lightgbm.dist/lightgbm/__init__.py", line 21, in <module lightgbm>
  File "<frozen importlib._bootstrap>", line 1027, in _find_and_load
  File "<frozen importlib._bootstrap>", line 1006, in _find_and_load_unlocked
  File "<frozen importlib._bootstrap>", line 688, in _load_unlocked
  File "/home/stas/projects/nuitka-wtf/nuitka-light-gdm-troubles/test-lightgbm.dist/lightgbm/dask.py", line 1088, in <module lightgbm.dask>
  File "/home/stas/projects/nuitka-wtf/nuitka-light-gdm-troubles/test-lightgbm.dist/lightgbm/dask.py", line 1143, in DaskLGBMClassifier
AttributeError: 'NoneType' object has no attribute 'partition'

The problem option is «–python-flag=no_docstrings», without it all works fine. But why, if anyone wants to strip docstring…

Same problem exists even with nuitka factory.

[kayhayen]

People are often controlling behaviour with doc strings. Like automatically setting function defaults to the documentation, etc. and this might be a case. I will look at it.

Ultimately what might have to happen is that we white list some things, that must retain some doc strings. Question of course is how this code runs with pure Python and -O

[kayhayen]

Well, Nuitka allows the second -O and the first -O to be separated, and gives no hint to what this needs, or when it will work or not. I am open however to anti-bloat measures for this. If you look at the code there, and believe the doc strings are only optional, you can patch it through user yaml files, and then of course we can integrate that in our standard config as well.

Eliminating doc strings seems very much worth it, not to speak of the obfuscucation parts of having them for your code.

As an enhancement, Nuitka probably needs to make it clear to the user, that it expects from the Python flags, that the original program already works with some options.