[gdal-dev] C# bindings compilation

[Paul Harwood]

I must admit that I needed to do it through conda.

I am no great fan of NuGet but this is not a "package war" thing. My target
platform is Unity and that is mono based and does not use NuGet - so if I
am going to have to do some handwork to integrate the package management I
would rather do it to conda which has vastly superior cross-platform
capabilities.

But, that aside, let me make a case why this approach is better. This
approach only builds the c# bindings and runs these on top of the factory
default gdal runtime package from conda on all platforms. This means that
the available gdal version is always up to date and the drivers are the
default set of drivers. There is a small amount of manual work on the c#
package when there is a new version of gdal - in most cases it will be of
the order of "run trial compile, look at the result, accept it" in which
case the c# package will only be a few days behind the latest gdal release.
And it runs across a wider range of platforms - including Mac and Windows.

On Mon, 12 Apr 2021 at 21:37, Björn Harrtell <bjorn.harrtell at gmail.com>
wrote:

> Interesting. I've as of yet only had success with using what is
> essentially a fork at https://github.com/MaxRev-Dev/gdal.netcore. I've in
> turn forked that to enable additional drivers and made my own nuget
> packages that are compatible with Debian 10 + .NET Core 3.1 which was my
> deployment target and it works like a charm.
>
> /Björn
>
> Den mån 12 apr. 2021 kl 21:48 skrev Paul Harwood <runette at gmail.com>:
>
>> Even et al.
>>
>> Just to update - your latest changes as per 3670 have done the trick. I
>> can now build C# bindings on top of the GDAL3.2.2 conda distribution on
>> Windows, Mac and Linux and have successfully run them in .NET / mono.
>>
>> I solved the compile time problems with a little judicious adjusting of
>> the compile-time environment and now have the basis for a package. I just
>> need a few more adjustments to make it more production-grade and then I
>> will submit it - particularly around tests and around run-time paths.
>>
>> I don't expect a fast turn around from conda-forge. This package does not
>> exactly fall into one of their normal teams :)
>>
>> Does anyone else want to volunteer to be on the maintainer list for the
>> package? Absolutely no work needed, you just the ability to deprecate if
>> the package gets very out of date.
>>
>> On Mon, 12 Apr 2021 at 08:59, Paul Harwood <runette at gmail.com> wrote:
>>
>>> Thanks Even
>>>
>>> On Sun, 11 Apr 2021 at 19:17, Even Rouault <even.rouault at spatialys.com>
>>> wrote:
>>>
>>>> Paul,
>>>>
>>>> RFCs are only required for substantial changes that affect the
>>>> OSGeo/GDAL repository. People who contribute to Conda Forge are free to do
>>>> so (hopefully in good faith, that is not defacing too much the GDAL
>>>> "brand"). If we decided to retire the C# bindings from OSGeo/GDAL this
>>>> should probably go through some motion
>>>>
>>>
>>> Cool. I will do that then.
>>>
>>> I would prefer if the result is at least semi-official and to have some
>>> other people from the community listed as maintainers/owners on the package
>>> in case something happens to me. There are too many zombie packages out
>>> there already!
>>>
>>> Would that be you?
>>>
>>>> Regarding build issues on Linux, I could reproduce that, but was like
>>>> "worked half time". I finally realized that you had to run "make" twice
>>>> because of oddities in the Makefile dependency rules. And there was the
>>>> base swig/SWIGmake.base file generating .cpp files in swig/csharp whereas
>>>> swig/csharp/GNUmakefile generated them in swig/csharp/gdal|const|ogr|osr.
>>>> I've hopefully addressed and rationalized that in
>>>> https://github.com/OSGeo/gdal/pull/3670. At least now on my local
>>>> machine (ubuntu 20.04), one a clean swig/csharp, "cd swig/csharp && make &&
>>>> make test" works. The error you got was similar to the one I got
>>>>
>>> It is good to able to see that it works and a working example. I will
>>> work with what you have in 3670
>>>
>>>> Regarding your compilation issues with your current recipee, I'd expect
>>>> you to have to rather copy most of the content of swig/include,
>>>> swig/include/csharp and swig/csharp in your own repository and adapt them
>>>> to a standalone build where you include and link against an installed
>>>> prefix of the native libgdal.
>>>>
>>>
>>>  That is sort of where I am heading. I want to avoid having anything
>>> between the conda feedstock and the root GDAL repo (unless and until there
>>> is any change to the way that the GDAL project manages bindings) for
>>> maintenance reasons and to avoid creating more zombies. I.e. there is only
>>> one definition of the SWIG objects but I have my own makefiles.
>>>
>>>> _______________________________________________
>> gdal-dev mailing list
>> gdal-dev at lists.osgeo.org
>> https://lists.osgeo.org/mailman/listinfo/gdal-dev
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/gdal-dev/attachments/20210412/1178b064/attachment-0001.html>