summaryrefslogtreecommitdiff
path: root/README.Android
blob: 9d022a719f80f87155643fb374885bd9128f03b2 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
Android-specific notes

Note that this document has not necessarily been updated to match
reality...

For instructions on how to build for Android, see README.cross.

* Getting something running on an emulated device

	Create an AVD in the android UI, don't even try to get
the data partition size right in the GUI, that is doomed to producing
and AVD that doesn't work. Instead start it from the console:

	LD_LIBRARY_PATH=$(pwd)/lib emulator-arm -avd <Name> -partition-size 500

In order to have proper acceleration, you need the 32-bit libGL.so:

        sudo zypper in Mesa-libGL-devel-32bit

	Where <Name> is the literal name of the AVD that you entered.

	Then:

	make cmd cmd=bash
	cd android/qa/sc
	make clean all install
	make run ; adb shell logcat

	And if all goes well - you should have some nice unit test output to
enjoy. After a while of this loop you might find that you have lost a lot of
space on your emulator's or device's /data volume. If using the emulator, you
can do:

	adb shell stop; adb shell start

but on a (non-rooted) device you probably just need to reboot it. On the other
hand, this phenomenon might not happen on actual devices.

	and continue onwards & upwards.

* What about using a real device?

	That works fine, too. You won't be able to use the "adb shell
stop" and "adb shell start" commands to do anything, as far as I
know. But don't seem to be necessary on a real device anyway?

* Debugging

	Some versions of the NDK had a broken gdb in the way that it can see
symbols only for shlibs that were already loaded when the debuggee was
attached, so you need to carefully guess where to put:

	fprintf(stderr, "Sleeping NOW!\n"); ::sleep(20);

	into the code; and when you see that in logcat, you have time
to run: ndk-gdb and it will attach the process.

	thread 12 # or perhaps 13
	backtrace

	may show you the native code trace.

	In r8b the ndk-gdb seems to work a bit better, and I think it isn't
necessary to use the mingw-and-ndk ndb-gdb any longer.

* Getting the symbols

In order to be able to debug, you also need the symbols.  Currently they are
stripped using a $(STRIP) call in android/Bootstrap/Makefile.shared ; make
sure you change it only to 'cp'.

But then you need to limit the size of the resulting binary by other means,
that is strip most of the symbols (but the interesting ones) already during
the build.  For that, use something like

--enable-dbgutil
--enable-selective-debuginfo="sal/"

in your autogen.input (but of course limit the --enable-selective-debuginfo
only to directories / libraries that are interesting to you).

* Common Errors / Gotchas

lo_dlneeds: Could not read ELF header of /data/data/org.libreoffice...libfoo.so
	This (most likely) means that the install quietly failed, and that
the file is truncated; check it out with adb shell ls -l /data/data/....


* Detailed explanation

Note: the below talk about unit tests is obsolete; we no longer have
any makefilery etc to build unit tests for Android.

Unit tests are the first thing we want to run on Android, to get some
idea how well, if at all, the basic LO libraries work. We want to
build even unit tests as normal Android apps, i.e. packaged as .apk
files, so that they run in a sandboxed environment like that of
whatever eventual end-user Android apps there will be that use LO
code.

Sure, we could quite easily build unit tests as plain Linux
executables (built against the Android libraries, of course, not
GNU/Linux ones), push them to the device or emulator with adb and run
them from adb shell, but that would not be a good test as the
environment such processs run in is completely different from that in
which real end-user apps with GUI etc run. We have no intent to
require LibreOffice code to be used only on "rooted" devices etc.

All Android apps are basically Java programs. They run "in" a Dalvik
virtual machine. Yes, you can also have apps where all *your* code is
native code, written in a compiled language like C or C++. But also
also such apps are actually started by system-provided Java
bootstrapping code (NativeActivity) running in a Dalvik VM.

Such a native app (or actually, "activity") is not built as a
executable program, but as a shared object. The Java NativeActivity
bootstrapper loads that shared object with dlopen.

Anyway, our current "experimental" apps (DocumentLoader,
LibreOffice4Android and LibreOfficeDesktop) are not based on
NativeActivity any more. They have normal Java code for the activity,
and just call out to a single, app-specific native library (called
liblo-native-code.so) to do all the heavy lifting.