sphinx_automodapi.automodsumm doesn't resolve autodoc_mock_imports

[NMudunuru]

I have a python package module that imports 3rd party modules which aren't included in the sphinx documentation. So I added these modules to mock imports in the conf.py

autodoc_mock_imports = ["abaqus", "abaqusConstants","material","section","part","assembly", "step", "interaction","mesh","load","job","optimization","sketch", "odbAccess", "connectorBehavior","pandas","numpy","matplotlib","scipy"]

While using the mock imports works with the sphinx automodule directive without throwing an error.

.. automodule:: mymodule

I get the an error when using the module with automodapi directive.

.. automodapi:: mymodule

Error:

Extension error (sphinx_automodapi.automodsumm):
Handler <function process_automodsumm_generation at 0x0000021867BC5510> for event 'builder-inited' threw an exception (exception: No module named 'abaqus')

[cpz-research]

I get a similar problem with automodsumm, but also no problem with automodule.

The error in more detail:

• import numba within a module leads to an error (exception: unsupported operand type(s) for *: 'itemsize' and 'int').
• I don't get this error when importing numba in python using the same conda environment
• Numpy statements doesn't seem to be resolved when numba code is called via sphinx. (e.g. when typing np.byte in the debugger, just np.byte is returned instead of numpy.int8 as expected.
• The error occurs using sphinx=5.1.1, but not using sphinx=5.1.0

Let me know if I a minimal example would help. Would be great if anyone has an idea how to fix it on my side or how to fit it in the package.

[pllim]

I don't think Astropy uses autodoc_mock_imports. We use all optional imports as local imports, so usually they don't even get triggered during doc build. And in narrative docs that need them for doctest, we use pytest-doctestplus to skip if it is not installed.

Unfortunately, we do not have enough resources to look at this in the near future. So, if you can propose a patch or find a way to work around it, it would be greatly appreciated. Thanks!

[HealthyPear]

@pllim not sure if my use case is the same, so I'll describe it here and if it's not, then I guess we can move it to a fresh issue

I am working on a python package for which 1 module uses PyTorch. This module is for supporting a specific set of operations that many users won't need, so I didn't want to install it everytime and basically created try/except clauses which in case of ModuleNotFoundError trigger an exception with a meaningful message.

Now for building the docs for my API it seems weird to me that PyTorch needs to be installed in the environment in which I do make html, even if I explicitly mention "torch" in the intersphinx mapping for example.

In fact, I get this,

Extension error (sphinx_automodapi.automodsumm):
Handler <function process_automodsumm_generation at 0x141b362a0> for event 'builder-inited' threw an exception (exception: PyTorch is required to use the API related to the `XXXX` module.
Please, see installation instructions about optional dependencies.)
make: *** [Makefile:20: html] Error 2

I also tried to do autodoc_mock_imports = ["torch"] without success

Is this the same problem?

If yes, what is the best course of action?

[pllim]

@HealthyPear , I am not able to answer this on top of my head and I don't have time to investigate, you could try ask the astropy-dev mailing list or on https://community.openastronomy.org/c/astropy/8 ? Sorry!