Resulting python3.3 binary segfaults.

[shizmob]

Currently, the resulting python3.3 binary in $BUILD/bin crashes when launched. It seems to trigger a segfault in Bionic:

F/libc    (19156): Fatal signal 11 (SIGSEGV) at 0x00000000 (code=1), thread 19156 (python3.3)
I/DEBUG   (  273): *** *** *** *** *** *** *** *** *** *** *** *** *** *** *** ***
...
I/DEBUG   (  273): pid: 19156, tid: 19156, name: python3.3  >>> bin/python3.3 <<<
I/DEBUG   (  273): signal 11 (SIGSEGV), code 1 (SEGV_MAPERR), fault addr 00000000
W/NativeCrashListener( 1032): Couldn't find ProcessRecord for pid 19156
I/DEBUG   (  273):     r0 00000000  r1 00000000  r2 00000000  r3 00000000
I/DEBUG   (  273): AM write failure (32 / Broken pipe)
I/DEBUG   (  273):     r4 00000000  r5 bef22a4c  r6 00000000  r7 b6f4dfd8
I/DEBUG   (  273):     r8 00000001  r9 00d3a030  sl 00000000  fp 00d3a020
I/DEBUG   (  273):     ip 0018dc2c  sp bef229d0  lr b6f2f4a1  pc b6f270b0  cpsr 600e0030
I/DEBUG   (  273):     d0  0000000000000000  d1  0000000000000000
I/DEBUG   (  273):     d2  0000000000000000  d3  0000000000000000
I/DEBUG   (  273):     d4  0000000000000000  d5  0000000000000000
I/DEBUG   (  273):     d6  0000000000000000  d7  ac5327c000000000
I/DEBUG   (  273):     d8  0000000000000000  d9  0000000000000000
I/DEBUG   (  273):     d10 0000000000000000  d11 0000000000000000
I/DEBUG   (  273):     d12 0000000000000000  d13 0000000000000000
I/DEBUG   (  273):     d14 0000000000000000  d15 0000000000000000
I/DEBUG   (  273):     d16 c1d4ea5d87124dd3  d17 3f50624dd2f1a9fc
I/DEBUG   (  273):     d18 41aa6eb704000000  d19 0000000000000000
I/DEBUG   (  273):     d20 0000000000000000  d21 0000000000000000
I/DEBUG   (  273):     d22 0000000000000000  d23 0000000000000000
I/DEBUG   (  273):     d24 0000000000000000  d25 0000000000000000
I/DEBUG   (  273):     d26 0000000000000000  d27 0000000000000000
I/DEBUG   (  273):     d28 0000000000000000  d29 0000000000000000
I/DEBUG   (  273):     d30 0000000000000000  d31 0000000000000000
I/DEBUG   (  273):     scr 00000010
I/DEBUG   (  273): 
I/DEBUG   (  273): backtrace:
I/DEBUG   (  273):     #00  pc 000230b0  /system/lib/libc.so (strlen+83)
I/DEBUG   (  273):     #01  pc 0002b49d  /system/lib/libc.so (strdup+4)
I/DEBUG   (  273):     #02  pc 00012f48  /data/py3/bin/python3.3

Seeing as the fault address is 0, it's very likely a NULL pointer being passed around somewhere.

[bootchk]

How are you getting that debug info?

As I recall, in other x-compiles of Python to Android there are several patches to Python to account for differences in libc. (Bionic is Android's libc?.) Maybe look at prior patch sets?

[shizmob]

adb logcat was responsible for that. Turns out Android logs quite a bit when a segfault occurs. I'm trying to get gdb working, unsuccessful so far.

I'll take a look but I fear there's nothing much of help in there; I've skimmed them before.

[bootchk]

Seems to me you need to build Python with debug symbols (or not stripped?). So that:

I/DEBUG ( 273): #1 pc 0002b49d /system/lib/libc.so (strdup+4)
I/DEBUG ( 273): #2 pc 00012f48 /data/py3/bin/python3.3

tells you what (function+offset) python3.3 was in when it called strdup.

Just thinking off the top of my head, its been a while since I did such low level debugging.

[bootchk]

See: http://bugs.python.org/issue21390 which might be related, and which includes a patch.

It seems like it might be related because: 1) it is a call to strdup 2) it seems to say it only happens when you invoke Python without a script (a parameter to Python naming a file to execute) which is exactly what I tested (but I haven't tested it in the other case, passing a script.)

Again, I took a a very shallow look at the problem, maybe some of your other patches address this problem.

[shizmob]

That is very useful indeed! Thanks for finding that. Android's locale implementation is pretty damn broken, so this is not surprising at all. I'll try that patch and see what comes out of it. Python also seems to crash if you call it with a script argument, though.

As for debugging symbols, that's what I'm currently trying, but remote gdb debugging is kind of a different beast than local debugging. I'll figure it out, though.

[shizmob]

Yep, it definitely seems to be a similar issue:

#2 0x0001b5e0 in main (argc=1, argv=0xbe907a34) at ./Modules/python.c:45

Not the exact same issue as in the given bug report though, but it should be trivial to patch against.

[bootchk]

Yes.

That's interesting, that locale is critical to wide character/unicode coding, and that Google differs on whether setlocale() can return NULL.

[shizmob]

I've currently managed to patch Python to get that issue out of the way, but other locale issues keep popping up in other places. I'll push the modified .patch file when I've managed to get Python in at least a bootable state.

[shizmob]

This latest commit should allow you to start an interactive Python REPL at least. There's still numerous issues (like time not importing), but it's something! We'll fix issues as they appear.

[shizmob]

These latest commits will allow you to run regrtest.py. Closing this issue for now as it's fixed, let's move all remaining things to their own issues as they are encountered.