Python 3 recipe #1412

inclement · 2018-10-16T21:28:28Z

I haven't had much time to work on this since my initial proof of concept, but I've got back to it now. This PR is a work in progress, I wouldn't usually make it anywhere near this early but I know people are following this closely so I thought I might as well be clear what is going on with it.

Right now this is the most minimal possible implementation, I literally just got it working and there are many hacks to be resolved. However, it should work, the whole build process is now included although it's only tested on my machine so far there could be issues in other environments (especially macOS?). The main difference to the proof of concept is that I began moving the build into p4a's system rather than the shell/Makefile system used in Python bpo-30386. There's nothing wrong with that system, but since it isn't maintained it seems safer to make full use of p4a's tools instead.

Non-exhaustive non-ordered list of things to do (note: even if checked off it doesn't mean the implementation is final, just that the bulk of the work is done on that feature):

Minor note: as it stands this recipe doesn't support API lower than 21 due to Android limitations. There are some patches that exist to fix that, but I'm probably not going to look into that any time soon, so API 21 may effectively be the lowest supported API for this recipe.

pmp-p · 2018-10-17T03:50:59Z

Good job !
Waiting for it and ready to offer on the side some hacky patches to help on API 19 for all those devices who still requires kitkat ;)
btw 3.7.1rc1 is ok with the patches you know so i guess it's finally getting better for Android.

inclement · 2018-10-17T20:00:51Z

Waiting for it and ready to offer on the side some hacky patches to help on API 19 for all those devices who still requires kitkat ;)

Sounds good to me :)

btw 3.7.1rc1 is ok with the patches you know so i guess it's finally getting better for Android.

Glad to hear it! I was worried there could be changes that break things.

opacam · 2018-10-18T09:50:22Z

pythonforandroid/bootstraps/sdl2/build/jni/Application.mk

@@ -5,3 +5,4 @@

 # APP_ABI := armeabi armeabi-v7a x86
 APP_ABI := $(ARCH)
+APP_PLATFORM := android-21


This causes sdl2 recipe to fail. We can comment this line to bypass the error...as a temporary hack in order to make further tests, NOT AS A FINAL SOLUTION.

# APP_PLATFORM := android-21

Note: This already has been informed by @inclement into the "to do" section : "add proper APP_PLATFORM setting functionality"

Here are the log of the failure:

[DEBUG]: -> running ndk-build V=1 [DEBUG]: Android NDK: APP_PLATFORM set to unknown platform: 21. [DEBUG]: /media/DEVEL/Android/android-ndk-r17b/build/core/setup-app-platform.mk:135: *** Android NDK: Aborting . Stop. Exception in thread background thread for pid 31274: Traceback (most recent call last): File "/usr/lib/python2.7/threading.py", line 801, in __bootstrap_inner self.run() File "/usr/lib/python2.7/threading.py", line 754, in run self.__target(*self.__args, **self.__kwargs) File "/home/opacam/.local/lib/python2.7/site-packages/sh.py", line 1540, in wrap fn(*args, **kwargs) File "/home/opacam/.local/lib/python2.7/site-packages/sh.py", line 2459, in background_thread handle_exit_code(exit_code) File "/home/opacam/.local/lib/python2.7/site-packages/sh.py", line 2157, in fn return self.command.handle_command_exit_code(exit_code) File "/home/opacam/.local/lib/python2.7/site-packages/sh.py", line 815, in handle_command_exit_code raise exc ErrorReturnCode_2: RAN: /media/DEVEL/Android/android-ndk-r17b/ndk-build V=1 STDOUT: Android NDK: APP_PLATFORM set to unknown platform: 21. /media/DEVEL/Android/android-ndk-r17b/build/core/setup-app-platform.mk:135: *** Android NDK: Aborting . Stop. STDERR: Traceback (most recent call last): File "/home/opacam/.local/bin/p4a", line 11, in <module> load_entry_point('python-for-android', 'console_scripts', 'p4a')() File "/home/opacam/Devel/python-for-android/pythonforandroid/toolchain.py", line 981, in main ToolchainCL() File "/home/opacam/Devel/python-for-android/pythonforandroid/toolchain.py", line 518, in __init__ getattr(self, args.subparser_name.replace('-', '_'))(args) File "/home/opacam/Devel/python-for-android/pythonforandroid/toolchain.py", line 148, in wrapper_func build_dist_from_args(ctx, dist, args) File "/home/opacam/Devel/python-for-android/pythonforandroid/toolchain.py", line 192, in build_dist_from_args build_recipes(build_order, python_modules, ctx) File "/home/opacam/Devel/python-for-android/pythonforandroid/build.py", line 579, in build_recipes recipe.build_arch(arch) File "/home/opacam/Devel/python-for-android/pythonforandroid/recipes/sdl2/__init__.py", line 41, in build_arch shprint(sh.ndk_build, "V=1", _env=env) File "/home/opacam/Devel/python-for-android/pythonforandroid/logger.py", line 176, in shprint for line in output: File "/home/opacam/.local/lib/python2.7/site-packages/sh.py", line 863, in next self.wait() File "/home/opacam/.local/lib/python2.7/site-packages/sh.py", line 792, in wait self.handle_command_exit_code(exit_code) File "/home/opacam/.local/lib/python2.7/site-packages/sh.py", line 815, in handle_command_exit_code raise exc sh.ErrorReturnCode_2: RAN: /media/DEVEL/Android/android-ndk-r17b/ndk-build V=1 STDOUT: Android NDK: APP_PLATFORM set to unknown platform: 21. /media/DEVEL/Android/android-ndk-r17b/build/core/setup-app-platform.mk:135: *** Android NDK: Aborting . Stop.```

I don't know what caused this for you, does your NDK not have platform-21?

opacam · 2018-10-18T10:00:57Z

pythonforandroid/bootstraps/sdl2/__init__.py

+                with current_directory(
+                        join(self.ctx.python_recipe.get_build_dir(arch.arch),
+                             'Lib')):
+                    shprint(sh.zip, '-r', stdlib_zip, *os.listdir())


Fix missing argument for command os.listdir, we can fix it like this:

stdlib_dir = join( self.ctx.python_recipe.get_build_dir(arch.arch), 'Lib') with current_directory(stdlib_dir): shprint(sh.zip, '-r', stdlib_zip, *os.listdir(stdlib_dir))```

Alternative fix: drop python2 support :<

If you maintain the affected code as it is you will also drop support for python 3 prior to version 3.2 and it would be great that was documented in somewhere: https://docs.python.org/3/library/os.html?highlight=listdir#os.listdir

References: kivy#1412

opacam · 2018-10-19T01:20:27Z

Good job @inclement, this is going well and fast!

I applied some changes to your python3 recipe in order to grant support for some libs: sqlite3, openssl, libffi and libexpat.

I pushed this work into a branch of my p4a fork (with your commits until 9b6436b, squashed in one commit) . It would complete the last two points of the "to do". Be aware that the libs aren't fully tested in a real application, but successfully builds and the apk initializes fine. Feel free to use that if you want, no problem 😉

Here is the commit that do the job: opacam/python-for-android@b38c6ae

@inclement

Includes commits from @inclement's pr "Python 3 recipe" (squashed into this commit): - Moved from bpo-30386 branch to python 3.7.0 official distribution - Made all version references in python url dynamic Plus, update the libs patches in order to work with python 3.7.0. References: kivy#1412, inclement/python-for-android@588f736 and inclement/python-for-android@a7f9348

In order to keep the recipe more maintainable and readable, all the environment variables has been moved into method "get_recipe_env", overwriting the base class. To achieve this goal some other changes has been made: - Remove unused variables (platform_dir) - Move some "toolchain bin variables" directly into the env dictionary (because they are only used to set the corresponding env variables) - Add new env variable HOSTARCH (we use that in build_arch, cause this commit moves the needed variable that was used before this commit...and moreover...this new variable targets 'android_host' which still is not set as it should, so introducing this variable we avoid future errors when the target variable 'android_host' will get fixed) - Shorted long lines to avoid pep8 errors and to be more readable - Add some todo comments References: kivy#1412

- F401 (imported but unused) - W293 blank line contains whitespace - E305 expected 2 blank lines after class or function definition, found 1 - Also add/remove some line breaks to make code more readable Note: The imports for "pythonforandroid.patching" has been commented because we probably will need those imports when patches applied References: kivy#1412

- F401 XXX imported but unused - E303 too many blank lines (2) - Shortened long lines to improve reading References: kivy#1412

@inclement

The squashed commits in order to update the branch are: - Made bootstraps check the ndk api against the requested minsdk (inclement/python-for-android@cf1491a) - Fixed APP_PLATFORM setting (inclement/python-for-android@5acace3) So, right now, the branch should be up to date with @inclement's branch until date 2018-10-18, except the recipes (hostpython and python which has some additions and the commit 5db62ac which grants build compatibility with all supported python version supported by p4a) References: kivy#1412

inclement · 2018-10-19T22:24:02Z

Thanks @opacam, it's good to hear that these libraries will apparently work without issue.
I'd like to try to make these work without patching python if possible, looking at the setup.py I think this is supposed to work. I'll probably come to this only after some of the other tasks though, so might not look at it quite immediately.

tito · 2018-10-20T15:27:22Z

So you wrote you tested with API 24, but which NDK are you using?
Also could you give a line for compilation that works for you?

inclement · 2018-10-20T18:02:29Z

I spoke with @tito on discord about the question above, but for the benefit of others:

So far I've tested with NDK 14 and API 24 (and ndk_api 21). This is because these are the specific versions that bpo-30386 was designed for. I will move to NDK 17c and API 26+ soon, but others have confirmed this seems to work.
My compilation line is python3 setup_testapp_python3.py apk --ndk-dir=/home/sandy/android/android-ndk-r14b --android-api=24, where setup_testapp_python3.py is in python-for-android/testapps. This is the main build I test during development, it is expected to work.

inclement · 2018-10-20T23:48:31Z

For reference, @tito found some more places where we need to be careful to use the right ndk api target, see https://gist.github.com/tito/73a2d35c236bb34f6441b64be44c5683 . These will be fixed in this PR soon.

opacam · 2018-10-21T12:40:22Z

OOoo, this is great, I'm testing the @tito's patch over the last published @inclement's work, we are getting closer but unfortunately the openssl libs causes troubles, linkages error when generating python 3 lib, but this can bypassed adding a new argument into the perl's configure (if we don't do that, the openssl build system will use the higher api we have available and if it don't match with the one we use with python 3 we have the mentioned linkage errors), so here are a patch to apply after @tito's patch:

https://github.com/opacam/python-for-android/commit/944e47da87eed8c6fd379f78daf989025dbfc997.patch

And another patch to add testapp for python3 with openssl and sqlite3

https://github.com/opacam/python-for-android/commit/351bcc946741d89b6394f65586afa9a3fe73377c.patch

This will solve the linkage problem, however no_ssl generated. The good news are the sqlite3 module is generated.

Note: Right now I'm testing with ndk-r17b, python 3.6 interpreter and targeting api 24 and with @inclement's method but with the testapp i posted above with openssl and sqlite added.

opacam · 2018-10-21T18:48:05Z

The following commit solves the _ssl problem:

opacam@675e2fb

And here another commit that we grant support fo libffi and libexpat (with testapp included), following the method introduced in the openssl fix and tested with both api 24 and 27 and using ndk-r17b:

opacam@bec21c7

Fix python3crystax:

opacam@3b26a57

Fix python2 (no working openssl with newest libraries):

opacam@77f4b60

By the way, so many thanks for your testing method @inclement!!...I was using some kind of hacky method which I prefer not to mention 😬

inclement · 2018-10-23T20:30:36Z

General update: Currently seems working fine with API 27 and newest SDK, I'm moving on to cleaning up the code and fixing some of the other things I broke along the way (like the other python recipes as noted by @opacam). I appreciate the testing and feedback.

I won't let the openssl and sqlite bullet points hold up this main PR, since others have confirmed so well that they work fine, but I expect to check through everything else in my original post before merging.

The amount of progress will depend on how much time I have to work on it, but hopefully I can get a lot of the remaining work done by next week.

inclement · 2018-10-27T17:15:40Z

I think this is essentially ready now. I'm still working to fix some code style issues and I'll try to make the tests pass, so don't merge just yet, but it should be good to review.

I've focused on making the new NDK API arguments work properly, as this is the most important thing for API 26+ support. When this is merged, I'll make a new issue with a list of the things that still need doing for the python3 recipe. I've already fixed all the TODOs I left in the code except within the python3 recipe itself, which can be considered experimental for now.

I also haven't tested python3crystax yet because I don't have python3.6 installed. Has anyone checked that this works? I'll do it if necessary.

^^ for @tito @AndreMiras in particular, but comments from others are welcome.

AndreMiras · 2018-10-27T17:37:32Z

Awesome work! The overall looks good.
I can take a look/play with the python3crystax to check for regression.
Simply double check you're rebased with upstream and I'll hook to your fork to build/run some apps on python3crystax.

AndreMiras

I looked through the code and didn't see any red flag so far.
I'll know more after checking out and building for python3crystax to look for regressions to begin with.
In the meantime fixing the linter will also help us checking for regression since Travis will go further in the integration testing process.
Some unit tests also broke in tests/recipes/, but this is probably because of the refactoring. I may pick this up later.

AndreMiras · 2018-10-27T17:43:22Z

pythonforandroid/bootstraps/sdl2/build/build.py

@@ -243,6 +244,8 @@ def make_package(args):
        tar_dirs.append('private')
    if exists('crystax_python'):
        tar_dirs.append('crystax_python')
+    if exists('_python_bundle'):
+        tar_dirs.append('_python_bundle')


Minor suggestion for this and the lines above:

for tar_dir in {'private', 'crystax_python', '_python_bundle'}: if exists(tar_dir): tar_dirs.append(tar_dir)

AndreMiras · 2018-10-27T17:44:59Z

pythonforandroid/bootstraps/sdl2/build/build.py

@@ -409,7 +412,17 @@ def make_package(args):

 def parse_args(args=None):
    global BLACKLIST_PATTERNS, WHITELIST_PATTERNS, PYTHON
-    default_android_api = 12
+
+    # Get the default minsdk, equal to the NDK API that this dist it built against


Minor typo:
s/it built/is built/

AndreMiras · 2018-10-27T17:47:45Z

pythonforandroid/bootstraps/sdl2/build/build.py

+        info = json.load(fileh)
+        if 'ndk_api' not in info:
+            print('Failed to read ndk_api from dist info')
+            default_android_api = 12  # The old default before ndk_api was introduced


should we add the actual Android API that will be used in the message?

print('Failed to read ndk_api from dist info, defaulting to android API {}'.format(default_android_api))

And of course invert both lines so default_android_api is well defined

pythonforandroid/bootstraps/sdl2/build/jni/src/start.c

AndreMiras · 2018-10-27T18:04:22Z

pythonforandroid/bootstraps/sdl2/build/jni/src/start.c

  } else {
-    LOGP("crystax_python does not exist");
+      LOGP("crystax_python does not exist");


Minor, should not this be?

LOGP("crystax_python or _python_bundle does not exist");

AndreMiras · 2018-10-27T18:48:54Z

pythonforandroid/recipes/hostpython3/__init__.py

+                # Create the Setup file. This copying from Setup.dist
+                # seems to be the normal and expected procedure.
+                assert exists(join(build_dir, 'Modules')), (
+                    'Expected dir {} does not exist'.format(join(build_dir, 'Modules')))


Minor: maybe we want to go for an explicit raise rather than assert. Who knows perhaps some people run p4a with python -O 😄

AndreMiras · 2018-10-27T18:51:54Z

pythonforandroid/recipes/hostpython3/__init__.py

+            info('Skipping hostpython3 build as it has already been completed')
+
+        self.ctx.hostpython = join(build_dir, 'python')
+        self.ctx.hostpgen = '/usr/bin/false'  # is this actually used for anything?


Good question, should we try to remove it and if it leads to issues we can just fix it and add a comprehensive comment?

AndreMiras · 2018-10-27T19:04:47Z

pythonforandroid/recipes/python2/__init__.py

+                    for suffix in EXCLUDE_EXTS:
+                        if filename.endswith(suffix):
+                            removes.append(filename)
+            shprint(sh.rm, '-f', *removes)


Out of scope since it was already in our code base.
But it hurts a little bit knowing this entire block could be a single line command using find:

find . -name '*.ext`' -o -name '*.ext2' -o -name '*.ext3' -delete

But yes that's not portable. However I'm wondering if there's not a way to do it in Python that would look nicer

I'm thinking more and more about making some better python abstractions for this. Something for the near future, perhaps.

There is fnmatch in the standard lib that can be used.

Actually I realize endswith can take tuple 😄
So we could actually go one for less:

if filename.endswith(EXCLUDE_EXTS):

But I think it's OK to tackle it outside this pull request, since we have enough to check

AndreMiras · 2018-10-27T19:11:40Z

pythonforandroid/recipes/python2/__init__.py

+            shprint(sh.rm, '-rf', 'lib2to3')
+            shprint(sh.rm, '-rf', 'idlelib')
+            for filename in glob.glob('config/libpython*.a'):
+                shprint(sh.rm, '-f', filename)


Minor, maybe this would work directly?

sh.rm('-f', glob.glob('config/libpython*.a'))

pythonforandroid/recipes/python3/__init__.py

tito · 2018-10-28T08:33:18Z

pythonforandroid/bootstraps/sdl2/build/jni/src/start.c

-     configurable */
-  /* AND: P4A uses env vars...not sure what's best */
-  LOGP("Initialize Python for Android");
+  setenv("P4A_BOOTSTRAP", "SDL2", 1);  // env var to identify p4a to applications


This is a great addition!

tito · 2018-10-28T08:39:46Z

pythonforandroid/bootstraps/sdl2/build/src/main/java/org/kivy/android/PythonUtil.java

@@ -45,6 +45,8 @@ protected static void addLibraryIfExists(ArrayList<String> libsList, String patt
        addLibraryIfExists(libsList, "crypto.*", libsDir);
        libsList.add("python2.7");
        libsList.add("python3.5m");
+        libsList.add("python3.6m");
+        libsList.add("python3.7m");


Shouldn't theses changes added in all others bootstraps?

The other bootstraps don't actually support python3 yet, I think it's best to leave that as a separate task rather than delay merging.

tito · 2018-10-28T08:43:48Z

testapps/setup_testapp_python3crystax.py

+options = {'apk': {'debug': None,
+                   'requirements': 'sdl2,pyjnius,kivy,python3crystax',
+                   'android-api': 19,
+                   'ndk-dir': '/home/asandy/android/crystax-ndk-10.3.2',


os.environ["ANDROIDNDK"] ?

AndreMiras · 2018-10-28T22:45:00Z

pythonforandroid/recipes/python3crystax/__init__.py

+
+    def link_root(self, arch_name):
+        return join(self.ctx.ndk_dir, 'sources', 'python', self.major_minor_version_string(),
+                    'libs', arch_name)

 recipe = Python3Recipe()


Shouldn't we rename this lass Python3CrystaxRecipe to avoid having it clashing with the new Python3Recipe one?

AndreMiras · 2018-10-28T23:13:04Z

pythonforandroid/recipes/sdl2/__init__.py

        py2 = self.get_recipe('python2', arch.ctx)
        env['PYTHON2_NAME'] = py2.get_dir_name()
+        py3 = self.get_recipe('python3', arch.ctx)


py3 is unused here, right?

AndreMiras · 2018-10-28T23:19:12Z

pythonforandroid/build.py

@@ -615,7 +626,9 @@ def run_pymodules_install(ctx, modules):

    venv = sh.Command(ctx.virtualenv)
    with current_directory(join(ctx.build_dir)):
-        shprint(venv, '--python=python2.7', 'venv')
+        shprint(venv,
+                '--python=python{}'.format(ctx.python_recipe.major_minor_version_string()),


major_minor_version_string is a property, not a method, so we should remove the brackets

inclement · 2018-11-08T23:39:25Z

The python3 testapp is failing because something tries to call python3.7. I will investigate tomorrow.
The other tests revealed a bug with ndk_api being set to 0 if a dist already exists. I will maybe improve how this is handled, but I'd rather leave a larger refactoring to a separate PR.

AndreMiras

Thanks for that PR, amazing work! I'm in for the merge even though two builds out of five are broken.
Find below few of my personal feelings regarding things I wish we could improve:

I think it's a shame we broke E129 because it's very easy to fix and the warning made sense
organise the git log history a bit more or squash
go for a "more mainstream" Python version by default i.e. 3.6 rather than 3.7

As for the broken tests, for records we have:

configure: error: python3.7 interpreter not found
Having python3.7 installed on the Docker host would help. Or simply putting python3.6 in the requirements rather than 3.7.
Didn't find any valid dependency graphs.
That's a known conditional build limitation, I can fix later.
https://github.com/kivy/python-for-android/blob/3e561b0/ci/rebuild_updated_recipes.py#L17

inclement · 2018-11-09T22:09:16Z

go for a "more mainstream" Python version by default i.e. 3.6 rather than 3.7

I haven't actually tried, but I wouldn't be surprised if this doesn't work with Python 3.6.

configure: error: python3.7 interpreter not found

What concerns me about this is that we have a hostpython3 recipe, there should be no need for the user to have their own python3.7 installed, so I'll look into it before merging. It's a nice thing for a test to catch because of course I didn't test locally without python3.7!

AndreMiras

Looking good.
Conditional build is failing because of #1454
But we're good for a merge

RobertFlatt · 2018-11-27T01:13:46Z

Really fine work, thank you.
FYI a user experience with actual app using p4a master, Python3, and android-28

Below my full Python3/Crystax build recipe - builds, installs, and runs great.
(Though it would be nice if "pip install python-for-android" didn't require sudo)

EDIT

These are the changes to my Crystax build environment for r17c
.p4a:

--requirements=python3,kivy
--ndk_dir <WHERE_EVER>/android-ndk-r17c
--ndk_version 17c

sudo apt-get install zip

This builds, installs, and runs great.

Sorry for earlier mis-infomation

END EDIT

# Recipe for building a python3 app (written using Kivy) for Android/Crystax.
#
# ASSUMES Ubantu 18.04 CLEAN INSTALL
# Tested with Windows Subystem for Unix Ubantu 18.04 and Android Pie
#

sudo apt-get update
sudo apt-get upgrade

### Workaround Issue #1327. Java heap issue - thanks JonasT
### Also add this line to .bashrc
export GRADLE_OPTS="-Xms1724m -Xmx5048m -Dorg.gradle.jvmargs='-Xms1724m -Xmx5048m'"

### Install modified P4A Python 2.7 dependencies
# changes to documentation:
# python3-all-dev replaces python2.7-dev   
# openjdk-8-jdk   replaces openjdk-7-jdk
# apparently not required : libncurses5:i386 libstdc++6:i386 zlib1g:i386 

sudo apt-get install -y build-essential ccache git zlib1g-dev python3-all-dev openjdk-8-jdk unzip autoconf libtool

### Install more P4A dependencies
### P4A requires Python 3.6, the default Python version on Ubantu 18 is 3.6
### gradle, virtualenv, pip, pkg, libgl1
### Gradle MUST be installed AFTER openjdk-8-jdk has been installed

sudo apt-get install -y gradle virtualenv python3-pip  pkg-config libgl1-mesa-dev

### Install Python packages
### Versions I get:  Cython 0.29.1 , Kivy 1.10.1,  P4A '0.6.0'
### The P4A version is mostly meaningless it is really a newer 'master'(?) build
### P4A must be installed with sudo else p4a stub is not created
### P4A requires Cython to be installed with sudo, else it will not find cython
### In some environment with more than one Python version, these two sudo
### might be an issue?

sudo -H pip3 install cython
pip3 install kivy
sudo -H pip3 install git+https://github.com/kivy/python-for-android.git

# Download Android and Crystax Tools (ndk and sdk) to <WHERE_EVER>
# https://www.crystax.net/en/download    crystax-ndk-10.3.2-linux-x86_64.tar.xz
# https://developer.android.com/studio/  sdk-tools-linux-4333796.zip
mkdir <WHERE_EVER>
cd    <WHERE_EVER>
unzip  -d sdk  sdk-tools-linux-4333796.zip 
tar xf crystax-ndk-10.3.2-linux-x86_64.tar.xz

# Build the Android Tools
cd <WHERE_EVER>/sdk/tools/bin
./sdkmanager "platforms;android-28"
./sdkmanager "build-tools;28.0.3"

### Install Done
### Build app-version-debug.apk

cd <location of app main.py>

# Create a ".p4a" containing (replace text between and including <>)
# An app may need more options than these:
--dist_name=<XXX>
--private .
--package=com.<YYY>.<XYZ>
--name <ZZZ>
--bootstrap=sdl2
--requirements=python3crystax,kivy
--arch=armeabi-v7a
--icon <ICON>.png
--sdk_dir <WHERE_EVER>/sdk
--ndk_dir <WHERE_EVER>/crystax-ndk-10.3.2
--android_api 28
--ndk_version 10.3.2
--ndk-api 21
--version <0.0.0>

```
`

ApplySci · 2018-12-03T13:45:55Z

FYI a user experience with actual app using p4a master, Python3, and android-28
Below my full Python3/Crystax build recipe - builds, installs, and runs great.

@Ham-Merhead does this produce an apk that the Google Play Store accepts? And have you tried running the apk on an Android 8.1 platform? (I've hit a wall with an apk that worked on 8.0, but fails on 8.1, failing to import _sqlite3.so)

OptimusGREEN · 2018-12-03T15:32:19Z

I thought crystax only supported up to 21?

RobertFlatt · 2018-12-03T17:11:47Z

@ApplySci My .apk works with Android 9 (I dont know about 8.1), but my app does not use sql which perhaps yours does. I have not tried submitting to the store.

RobertFlatt · 2018-12-03T17:24:33Z

@OptimusGREEN My understanding is the NDK (here Crystax) defines the minimum, and the (Android) SDK the maximum. But perhaps I don't understand?!
I also have an Android-NDK (in place of Crystax) build environment that is simpler
https://groups.google.com/forum/#!topic/kivy-users/Lf4zlYmLVPo

inclement added the work in progress label Oct 16, 2018

inclement mentioned this pull request Oct 16, 2018

Proof of concept for Python 3 without CrystaX, plus API 27 support #1375

Closed

opacam reviewed Oct 18, 2018

View reviewed changes

opacam added a commit to opacam/python-for-android that referenced this pull request Oct 19, 2018

Merge @inclement's python 3 recipe

9efead4

References: kivy#1412

opacam added a commit to opacam/python-for-android that referenced this pull request Oct 19, 2018

Pep8 stylish corrections for hostpython:

cd3bbf9

- F401 XXX imported but unused - E303 too many blank lines (2) - Shortened long lines to improve reading References: kivy#1412

AndreMiras mentioned this pull request Oct 24, 2018

[PEP8] fixed PEP 8 #1432

Closed

AndreMiras reviewed Oct 27, 2018

View reviewed changes

tito reviewed Oct 28, 2018

View reviewed changes

AndreMiras reviewed Oct 28, 2018

View reviewed changes

Updated travis.yml to target python3crystax testapp

e48d82e

inclement force-pushed the python3_recipe branch from d9e14ef to e48d82e Compare November 8, 2018 22:00

inclement added 5 commits November 8, 2018 22:06

Updated documentation to talk about Python 3 support

47a8daf

Improved handling of --android-api and --ndk-api

54c5b38

Restructured discovery of android-api and ndk-api inputs

c817ad0

Updated CI scripts to run a python3 testapp

f74ce04

Added ndk-api for sqlite,openssl testapp

12346d2

AndreMiras previously approved these changes Nov 9, 2018

View reviewed changes

AndreMiras mentioned this pull request Nov 9, 2018

Show actual values in info messages #1453

Closed

Added path to hostpython in python3 build env

f150014

inclement dismissed AndreMiras’s stale review via f150014 November 10, 2018 00:12

inclement added 2 commits November 10, 2018 00:43

Updated python3 testapp ndk-api to 21

82ad9e5

Removed Python3 testapp from CI for now

b076cd9

inclement removed the work in progress label Nov 10, 2018

inclement mentioned this pull request Nov 10, 2018

Python 3 recipe follow up issues #1455

Closed

6 tasks

AndreMiras approved these changes Nov 10, 2018

View reviewed changes

inclement merged commit 146bab2 into kivy:master Nov 10, 2018

AndreMiras mentioned this pull request Nov 11, 2018

Add support for minimum Android API. #1197

Closed

opacam mentioned this pull request Nov 12, 2018

[WIP] Refactor python recipes and update python2's to version 2.7.15 #1460

Closed

17 tasks

AndreMiras mentioned this pull request Nov 12, 2018

Google requiring API Target 26 in Aug/Nov 2018 #1219

Closed

SomberNight mentioned this pull request Nov 23, 2018

android: build apk using new python3 p4a toolchain spesmilo/electrum#4864

Merged

opacam mentioned this pull request Dec 17, 2018

Refactor python recipes and update python2's to version 2.7.15 (fixed history) #1534

Closed

15 tasks

AndreMiras mentioned this pull request Apr 16, 2019

Flaky continuous integration #1306

Closed

Python 3 recipe #1412

Python 3 recipe #1412

Conversation

inclement commented Oct 16, 2018 • edited Loading

pmp-p commented Oct 17, 2018

inclement commented Oct 17, 2018

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

opacam commented Oct 19, 2018 • edited Loading

inclement commented Oct 19, 2018

tito commented Oct 20, 2018 • edited Loading

inclement commented Oct 20, 2018

inclement commented Oct 20, 2018

opacam commented Oct 21, 2018

opacam commented Oct 21, 2018 • edited Loading

inclement commented Oct 23, 2018

inclement commented Oct 27, 2018 • edited Loading

AndreMiras commented Oct 27, 2018

AndreMiras left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

inclement commented Nov 8, 2018

AndreMiras left a comment

Choose a reason for hiding this comment

inclement commented Nov 9, 2018 • edited Loading

AndreMiras left a comment

Choose a reason for hiding this comment

RobertFlatt commented Nov 27, 2018 • edited Loading

ApplySci commented Dec 3, 2018

OptimusGREEN commented Dec 3, 2018

RobertFlatt commented Dec 3, 2018

RobertFlatt commented Dec 3, 2018

inclement commented Oct 16, 2018 •

edited

Loading

opacam commented Oct 19, 2018 •

edited

Loading

tito commented Oct 20, 2018 •

edited

Loading

opacam commented Oct 21, 2018 •

edited

Loading

inclement commented Oct 27, 2018 •

edited

Loading

inclement commented Nov 9, 2018 •

edited

Loading

RobertFlatt commented Nov 27, 2018 •

edited

Loading