[Mobile-sig] Progress update on Mobile Python work

Sun Jan 3 08:39:08 EST 2016

(I sent this reply originally to just Russell, but meant it for the
whole list - the information about Python with CrystaX in particular may
be interesting. Sorry for the double reply Russell.)

Hi Russell (just Russ?),

I'm one of the Kivy developers and have been working mainly on our own
python-for-android build tools recently, so I'm very interested in your
updates there - thanks for the post. I haven't posted here before (I
feel too inexpert!) but have read some of the previous discussion, and
certainly agree that it would be great to have python build for Android
without difficulty.

Regarding Cyd Haselton's patch, is the issue at
http://bugs.python.org/issue23496 the right place to follow it, or is
there discussion elsewhere? I've recently been trying to add python3
support to Kivy's python-for-android, but had no luck with any patches,
though I didn't know about Cyd's effort and mainly used SL4A's old
experimental python3 support which (although allowing compilation to
complete) I couldn't get running on the device. That problem could
easily be my own mistake elsewhere, but I'll give this method a try
instead, as it's much preferable anyway.

More recently, I was actually able to sidestep the build entirely with
the new support for python by the CrystaX NDK, which as of their recent
10.3.0 release includes precombiled python distributions for 2.7 and 3.5
on Android (all supported architectures). For anyone not aware, this is
a drop-in replacement for google's own NDK with many features added,
like newer versions of gcc/clang and better C++ standard support. This
means that the CrystaX NDK can compile upstream Python with no patches
at all. I gave it a try and it was extremely easy to get working, so
I'll soon be able to (finally...) add python3 support to
python-for-android. Using this method might be very practical for other
projects as well, if anyone is interested, though of course it's still
very valuable to get Python building for Android with the offical NDK.

I'm sad to hear that manipulating a native-widget Android app from
Python via JNI seems impractical, but good luck with VOC, I would love
to see this become doable.

Thanks,

Alexander

On 02/01/16 15:12, Russell Keith-Magee wrote:
> Hi all,
> 
> This SIG has been a bit quiet of late, but over the last few months (and
> especially over the Christmas/New Year break) I’ve made some significant
> progress on support for Python on mobile platforms (especially Android),
> so I thought I’d give an update of where things (at least, from where I
> stand).
> 
> iOS Support
> ===========
> 
> iOS support is getting quite strong. There is a patch against CPython
> 3.4.2 sources:
> 
> http://bugs.python.org/issue23670
> 
> This patch provides an XCode project that runs the Python test suite.
> There are still a handful of bugs; these are mostly related to edge
> cases in ctypes. 
> 
> The issues that previously existed with libffi have now been resolved.
> The biggest question with libffi support now is whether there is an
> opportunity to merge libffi support on OS X with the new libffi support
> on iOS. The generated code for libffi_osx is several years old at this
> point, and it would be easy to replace the current libffi_osx generated
> code with an updated version sharing sources with iOS. The only catch is
> that libffi no longer supports PowerPC architectures in their OS X
> builds (because Apple no longer supports those architectures).
> 
> There are two impediments currently preventing this patch from being
> merged into trunk.
> 
> The first is a bug, reported during the 3.5 alpha, that is still lingering:
> 
> http://bugs.python.org/issue22625
> 
> This bug was introduced in the process of fixing support for parallel
> builds in the root Makefile. This problem prevents *any* cross platform
> build from working. I don’t have enough Makefile mojo to sort this one out.
> 
> The second is the absence of a feedback from someone on the core team.
> The patch really needs a review from someone to indicate which of the
> design decisions I’ve made are acceptable to the CPython core team.
> 
> In the meantime, I’ve been maintaining an external project to make it
> easy to build a Python framework for use in an iPhone (or iPad, or
> AppleTV, or Apple Watch) app:
> 
> https://github.com/pybee/Python-iOS-support
> 
> You can download a pre-compiled version of the libraries from the
> project page. Add an Objective C bridging library (like Rubicon), and
> you can write your iOS (or tvOS) app in Python. 
> 
> Android Support
> ===============
> 
> Cyd Haselton has been working on an analogous patch for CPython to add
> support for Android. From what I can see, he’s having some difficulties;
> the code is segfaulting, and debugging the source of the segfault is
> proving difficult.
> 
> I experimented with CPython on Android briefly at the end of 2014; I
> came to the conclusion that it wasn’t really viable, because once you
> have CPython, you need to access Java libraries to actually write an
> application. That means you have to use JNI, and while JNI *exists* on
> Android, it is constrained to the point that it makes writing apps nigh
> on impractical.
> 
> Instead, I’ve been working on an alternate approach: transpiling CPython
> 3.4 bytecode directly to Java class files. The result is a project
> called VOC:
> 
> https://github.com/pybee/voc
> 
> This is a different approach to that used by Jython. There’s no Python
> interpreter at runtime - what 
> you push to the phone is a pure Java class file implementing Python
> language semantics. The price you pay is that there’s no exec(), eval()
> or REPL - but if you’re writing an app for distribution on a phone, that
> probably doesn’t matter.
> 
> Performance is passable, but not great - PyStone indicates that it’s
> roughly an order of magnitude slower than CPython on the same machine.
> However I haven’t been focussing on performance to date, and there’s a
> lot of low hanging opportunities for optimisation. It’s also worth
> noting that mobile apps (other than games) generally won’t be CPU bound
> - they’re network bound - so this performance issue isn’t as critical as
> it may seem.
> 
> Here’s a sample project:
> 
> https://gist.github.com/freakboy3742/3c6b74e8506d47d9bd97
> 
> When you compile that code with VOC, you get 3 class files (one for each
> class, and one for the module itself). If you add those class files to
> an Android project and upload to a phone, you get this:
> 
> https://twitter.com/PyBeeWare/status/683258762816192513/video/1
> 
> There’s still a lot of work to do on VOC; there are some language
> constructs that it doesn’t handle yet (most notably generators and
> context managers), and I need to port the parts of the standard library
> that are written in C. However, as the video shows, even without those
> parts, you can write simple Android applications in Python.
> 
> Packaging
> =========
> 
> One last thing - I’ve also created project called Briefcase to tie all
> these pieces together. 
> 
> https://github.com/pybee/briefcase
> 
> Briefcase is an extension to distutils that adds setup.py targets to
> package a project for distribution as an app. iOS and OS/X support is
> quite stable. A stub command exists for Android, but it isn’t fully
> operational yet. Apple TV support would be trivial to add - most of the
> pieces are already there, they just need to be tied together.
> 
> Summary
> =======
> 
> So - that’s where, IMHO, Python on Mobile is at right now. 
> 
> If you’ve got any questions, comments, or feedback - let me know.
> 
> Yours,
> Russ Magee %-)
> 
> 
> _______________________________________________
> Mobile-sig mailing list
> Mobile-sig at python.org
> https://mail.python.org/mailman/listinfo/mobile-sig
> 

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: OpenPGP digital signature
URL: <http://mail.python.org/pipermail/mobile-sig/attachments/20160103/109fbbc6/attachment.sig>