Fix LDSHARED and missing CFLAGS include path for python3crystax sources

[ghost]

As proposed in #1360 this change fixes env["LDSHARED"] not being set which can confuse distutils / python setup.py build_ext to use the incorrect host compiler, and it adds the -I folder to the python3 crystax sources when using python 3 crystax. This should allow a lot of recipes to be cleaned up later which currently have the same repeated hacks to work around this. (Uncleaned recipes should just duplicate include paths which should do no harm, so this shouldn't require any recipes to be changed to be safely merged)

Testing state:

• tested with python3crystax without kivy (but with Pillow recipe)
• not tested with python 2
• not tested with other recipes

[AndreMiras]

Thanks, you're a machine! 😂
I'd like to test this change myself, hence killing this flag from the 17 recipes before and after your change and see if the compilation first fails and then passes.
Of course if we had a 100% integration coverage that would be way easier, but as discussed we don't have that yet for various reasons 😕

[ghost]

@AndreMiras I really need python-for-android for my app, so it is worth it for me to sink some time into getting things fixed. 👍

I just tested this with my actual real world app which is python3crystax and uses the recipes Pillow and pysdl2, and some other pure python dependencies. For this app, this pull request builds & runs fine.

[AndreMiras]

OK good stuff. However it seems to break the Python2 builds 😕 yes I know it's a nightmare.
3 builds are failing with the same error https://travis-ci.org/kivy/python-for-android/builds/425450653

/usr/bin/ccache arm-linux-androideabi-gcc -DANDROID -mandroid -fomit-frame-pointer -D__ANDROID_API__=19 -isystem /opt/android/android-ndk-r16b/sysroot/usr/include/arm-linux-androideabi -isysroot /opt/android/android-ndk-r16b/sysroot -I/root/.local/share/python-for-android/build/python-installs/bdisttest_python2/include/python2.7 --sysroot /opt/android/android-ndk-r16b/platforms/android-19/arch-arm -lm -L/root/.local/share/python-for-android/build/libs_collections/bdisttest_python2/armeabi -Xlinker -export-dynamic -o python \
		Modules/python.o \
		-L. -lpython2.7 -ldl  -L/root/.local/share/python-for-android/build/other_builds/python2/armeabi/python2/python-install/lib -lz    -lm  
Creating directory /root/.local/share/python-for-android/build/other_builds/python2/armeabi/python2/python-install/bin
/bin/sh: 2: --sysroot: not found
Makefile:429: recipe for target 'sharedmods' failed
make: *** [sharedmods] Error 127
make: *** Waiting for unfinished jobs....
/usr/bin/install -c python /root/.local/share/python-for-android/build/other_builds/python2/armeabi/python2/python-install/bin/python2.7
if test -f libpython2.7.so; then \
	if test -n "" ; then \
		/usr/bin/install -c -m 555  /root/.local/share/python-for-android/build/other_builds/python2/armeabi/python2/python-install/bin; \
	else \
		/usr/bin/install -c -m 555 libpython2.7.so /root/.local/share/python-for-android/build/other_builds/python2/armeabi/python2/python-install/lib/libpython2.7.so; \
		if test libpython2.7.so != libpython2.7.so; then \
			(cd /root/.local/share/python-for-android/build/other_builds/python2/armeabi/python2/python-install/lib; ln -sf libpython2.7.so libpython2.7.so) \
		fi \
	fi; \
else	true; \
fi


  STDERR:

You can view the complete error clicking on the "Raw logs" button, e.g. https://api.travis-ci.org/v3/job/425450654/log.txt

Yes sometimes simple change break unsuspected things. I also don't see yet the relation between your changes and the way the build broke, but that's quite usual in p4a 😛
The way I usually debug this is by comparing the ENV before and after my changes, trying to understand what would break.

Edit: oh that might be that trailing quote after -Bsymbolic-functions there https://github.com/kivy/python-for-android/pull/1361/files#diff-569e13021e33ced8b54385f55b49cbe6R116
So now I'm curious why this didn't fail in Python3Crystax build for instance. Maybe this get overridden...

[ghost]

Yeah, interesting, it definitely didn't fail for me for some reason. Anyway, I fixed the trailing comma now. Let's see if that improves things

[ghost]

Is there something you have in mind I could test? Or how do we go about making a test branch with the recipes adapted accordingly so we can find out if this works for them?

[AndreMiras]

No, it's looking good, I do think it works.
I actually addressed the python include part in that PR https://github.com/kivy/python-for-android/pull/1250/files#diff-569e13021e33ced8b54385f55b49cbe6

I had more concern regarding if it should be done that way.
For example why don't we use the actual linker env['LD'] rather than the compiler env['CC']. I know we're doing it in the recipes and I know gcc can link, but well, was just wondering.
Also the env["CFLAGS"] python include part, I'm wondering if it should not sit in PythoRecipe.get_recipe_env() instead or if there's some rework we should do about it.

I'll discuss briefly with other devs on IRC and if it's OK we'll merge.