[maemo-developers] Hibernation of applications during ptrace

[From: Eero Tamminen]

Hi,

ext Philip Van Hoof wrote:
> Hi there hackers!
> 
> While I'm debugging my application using gdb, I noticed that in both
> Scratchbox for intel, Scratchbox for ARM and on the device certain
> things in the Maemo platform decide that the application is hibernating,
> or ... something.
> 
> It seems to be the reason why while I'm debugging my application, I very
> often get an error from ptrace telling me that the process no longer
> exists.
> 
> Usually I guess this hibernation support for applications is most likely
> absolutely awesome and totally cool. However, while I'm developing and
> especially during a gdb session, I'd prefer that nothing touches my
> processes and threads.
> 
> As debugging info I usually get something like this:
> 
> ...
> hildon-desktop[25510]: GLIB DEBUG default - #### computing hibernation key ####
> hildon-desktop[25510]: GLIB DEBUG default - #### hibernation key: appname ####
> ...
> hildon-desktop[25510]: GLIB DEBUG default - removing child from AS ...
> hildon-desktop[25510]: GLIB DEBUG default - ... no more children, removing app.
> ...
> 
> Is there a way to completely disable all this stuff? I know about
> "run-standalone.sh": Even with this tool I am seeing various problems.
> 
> Especially when I run my application in gdb. I'd like to disable it
> "totally". For example: not starting it, patching it out and overwriting
> the library on my device and scratchbox environment, shooting it in the
> head before I start my debugging sessions, ... molesting it? Really
> hurting it? Anything!
> 
> I also know about the alias for gdb, that makes gdb start with
> run-standalone.sh. Even that seems to be insufficient as I'm still
> seeing threads (usually it's threads) being killed by something. Usually
> while I'm debugging (while I'm focusing on finding a bug or race cond,
> it's usually really not the moment to kill _my_ thread).
> 
> Even after using every possible imaginable runstandalonable run-
> standalone.sh script, I'm still seeing threads getting killed.
> 
> I have also recompiled GDB from CVS (yesterday's version) and with this
> version all problems related to GDB specifically seem vanished, the only
> remaining problem appears to be hildon-desktop killing things behind my
> back and gdb's ptrace really not enjoying that.

You're just not pressing the "Enter" one more time after you type
"cont"?  If your application crashes, the enter waiting in Gdb
readline buffer will then cause Gdb immediately to repeat the previous
command and continue the process (which then promptly exits)...

This is very annoying Gdb behaviour, it should empty readline
buffer before getting new input from the developer.


	- Eero