[gdal-dev] Python3 continuous testing

[Even Rouault]

Hi Tamas,

> 
> I'm always confused about the GDAL-python bindings what is the best
> approach to use. The SWIG generated output is also included in the
> bindings, but as far as I remember it wasn't always up to date.

As far as I'm concerned, I try to make sure to commit updated generated files 
(for Python). And making sure they are sync'ed with the .i is part of the 
release process when we cut a version.

> Also the
> windows makefile will generate the bindings by default and there's no
> option to bypass this step as far as I remember.
> 
> On the other hand, I'm not quite sure which one is the suggested SWIG
> version to generate the bindings,

Me neither. Your below error seems to suggest that another SWIG version as the 
one you use (2.0.4 apparently) should be used (provided recent SWIG versions 
have caught up with Python 3.4). Perhaps open a ticket if you don't have time 
to investigate on that.

> I also remember the cases when the recent
> version wasn't sufficient and we had to use an earlier SWIG version to
> compile. Not sure how it depends on the Python version either.

I somehow remember an issue (something like one language needed a particular 
SWIG version, but that didn't work with other languages), but I'm not sure if 
it was with Python or another language.

> 
> I think we should either remove the SWIG generated output from the
> repository or modify the makefiles not to generate the output every time.

Sounds reasonable. We've had such discussion in the past and I think the 
outcome was that requiring people to have swig available to build the Python 
bindings could be a pain for them. Beyond the pain, if we indeed need 
particular/very recent swig for Python, then having the generated files can 
make sense.

The Unix makefile has a 'generate' target that generates the bindings with SWIG 
and a 'build' target that compiles generated bindings . Perhaps similar 
approach could be taken for the Windows makefile (I don't anticipate myself 
trying that, so I let you experimenting if you wish)

Even

-- 
Spatialys - Geospatial professional services
http://www.spatialys.com