[PyQt] Trouble using pyqtdeploy wuth SciPy's cythonized modules

[Robert Kent]

Hi Phil,

Is that rev. 654 'Fixed object leaks in the importing of side-by-side 
packages.' with changes to 'pyqtdeploy/builder/lib/pdytools_module.cpp'?

I've got a feeling this is something to do with me physically displacing 
the .so but I can pull this version if you think its worth a go.

Cheers, Rob



On 04/02/2016 12:50, Phil Thompson wrote:
> On 4 Feb 2016, at 12:36 pm, Robert Kent <rob at gulon.co.uk> wrote:
>> Hi All,
>>
>> I'm using pyqtdeploy 1.2 on OSX to package my application and so far have succeeded in getting NumPy and matplotlib to build. I (and Google!) came up with a method for including modules built as static objects (.so) by adding them to the Resources directory of the app bundle and then adding a hook via sys.meta_path to allow them to be imported outwith the resource system. To clarify, this means that anything that is a .py file is built into the application via the Qt resource system and anything that is an .so file is packaged separately. This has worked fine until I tried it with SciPy and it's cythonized .so files which throw a SystemError on import as follows:
>>
>> ...
>> File ":/scipy/optimize/__init__.py", line 233, in <module>
>>    File "bootstrap.py", line 2237, in _find_and_load
>>    File "bootstrap.py", line 2226, in _find_and_load_unlocked
>>    File "bootstrap.py", line 1191, in _load_unlocked
>>    File "bootstrap.py", line 1161, in _load_backward_compatible
>>    File ":/scipy/optimize/_minimize.py", line 26, in <module>
>>    File "bootstrap.py", line 2237, in _find_and_load
>>    File "bootstrap.py", line 2226, in _find_and_load_unlocked
>>    File "bootstrap.py", line 1191, in _load_unlocked
>>    File "bootstrap.py", line 1161, in _load_backward_compatible
>>    File ":/scipy/optimize/_trustregion_dogleg.py", line 5, in <module>
>>    File "bootstrap.py", line 2237, in _find_and_load
>>    File "bootstrap.py", line 2226, in _find_and_load_unlocked
>>    File "bootstrap.py", line 1191, in _load_unlocked
>>    File "bootstrap.py", line 1161, in _load_backward_compatible
>>    File ":/scipy/linalg/__init__.py", line 175, in <module>
>>    File "bootstrap.py", line 2237, in _find_and_load
>>    File "bootstrap.py", line 2226, in _find_and_load_unlocked
>>    File "bootstrap.py", line 1191, in _load_unlocked
>>    File "bootstrap.py", line 1161, in _load_backward_compatible
>>    File ":/scipy/linalg/basic.py", line 21, in <module>
>>    File "bootstrap.py", line 2237, in _find_and_load
>>    File "bootstrap.py", line 2226, in _find_and_load_unlocked
>>    File "bootstrap.py", line 1191, in _load_unlocked
>>    File "bootstrap.py", line 1161, in _load_backward_compatible
>>    File "__main__.py", line 30, in load_module
>>    File ":/imp.py", line 243, in load_module
>>    File "scipy/linalg/_solve_toeplitz.pyx", line 4, in init scipy.linalg._solve_toeplitz (scipy/linalg/_solve_toeplitz.c:19797)
>> SystemError: Parent module '' not loaded, cannot perform relative import
>>
>> The whole 'relative import' bit seems to be a red herring as what its actually trying to do at that point is import bits of numpy. I'm guessing this is to do with me moving these .so files outside of the normal import tree so it cant find things...
>>
>> I'm just wondering if Phil or anybody else has had success in including either SciPy or any other cythonized modules in a pyqtdeploy project and, if so, how?
> There is a fix in the 1.2-maint branch in hg for a problem triggered by cython. It doesn't look like the same problem, but it might be worth a try.
>
> Phil

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://www.riverbankcomputing.com/pipermail/pyqt/attachments/20160204/9031eaae/attachment.html>