[Crash-catcher] improving handling of debuginfos by abrt

[Denys Vlasenko]

Hi Richard, folks,

Jiri said to me on IRC:

<jmoskovc> the debuginfo-install support in PackageKit seems to be finally working
<jmoskovc> so if you want, you can change the old debuginfo-install script for
pk-debuginfo-install

Here is a quick recap how we currently do it, what's wrong with it,
and how it can be improved.

When we process a crash, we have a core file. We just run gdb on it
in batch mode by running

gdb -batch -x FILE

where FILE contains:

file BINARY
code COREFILE
thread apply all backtrace full
q

This produces a backtrace. gdb tries to produce better backtrace
(with line numbers, variable names and so on) using debug info.

It tries to locate debuginfo by finding executable's build id,
and uses debuginfo if it is found.

example:
# ls -l /usr/lib/debug/.build-id/00/5af5b5e7d6ab560825b0747fcbe41112431b8c.debug 
lrwxrwxrwx 1 root root 28 2009-07-20 18:08
/usr/lib/debug/.build-id/00/5af5b5e7d6ab560825b0747fcbe41112431b8c.debug ->
../../usr/bin/makestrs.debug

However, we (abrt) do not know whether debuginfo is installed,
so currently we just run "debuginfo-install -y -- PACKAGE".

This is a simple approach, but it has several drawbacks.

* we install debuginfo packages even if debuginfo files
  are present. For example, some people in large installations
  mount a network filesystem on /usr/lib/debug/ with *all* debuginfos
  installed, in order to reduce unneeded downloads.
  Running debuginfo-install in this case is not only unproductive,
  it will likely fail.

* we install all debuginfos, even those not needed for this particular
  crash. There may be dozens of libraries linked in, yet the stacktrace
  processing may not need debuginfo for most of them.

* Jiri pointed out another problem: we may be installing debuginfos
  for a wrong version of the executable. It does not mess up stacktrace,
  it just does not help gdb at all.


So, Richard, what is this pk-debuginfo-install thing, and how it can
help us here? In what ways is it different from debuginfo-install?

-- 
vda

[Richard Hughes]

----- "Denys Vlasenko" <dvlasenk(a)redhat.com&gt; wrote:
>  So, Richard, what is this pk-debuginfo-install thing, and how it can
>  help us here? In what ways is it different from debuginfo-install? 
The main difference is that it uses PackageKit, so it:

* can be used by a non-root user
* doesn't take the yum lock
* integrates with other software using PackageKit
* error granularity, so you don't have to parse stdout to find out why the command
failed

This is the output from --help:

[hughsie@hughsie-laptop ~]$ pk-debuginfo-install --help
Usage:
  pk-debuginfo-install [OPTION...]

PackageKit Debuginfo Installer

Help Options:
  -h, --help           Show help options

Application Options:
  -v, --verbose        Show extra debugging information
  -s, --simulate       Don't actually install any packages, only simulate what would
be installed
  -n, --no-depends     Do not install dependencies of the core packages
  -q, --quiet          Do not display information or progress

And in the man page (man pk-debuginfo-install) more details are exposed:

RETURN VALUES
       0   Success, at least one package was installed
       1   Failed with miscellaneous internal error.
       3   Failed to get software sources list.
       4   Failed to enable software sources.
       5   Failed as nothing to do.
       6   Failed to find dependant packages.
       7   Failed to install.
       8   Failed to disable software sources.

But it's pretty much a drop in replacement for debuginfo-install from an end-user
perspective.

Richard.

[Denys Vlasenko]

On Thu, 2009-09-03 at 12:07 -0400, Richard Hughes wrote:
>  ----- "Denys Vlasenko" <dvlasenk(a)redhat.com&gt; wrote:
>  > So, Richard, what is this pk-debuginfo-install thing, and how it can
>  > help us here? In what ways is it different from debuginfo-install?
>  
>  The main difference is that it uses PackageKit, so it:
>  
>  * can be used by a non-root user
>  * doesn't take the yum lock
>  * integrates with other software using PackageKit
>  * error granularity, so you don't have to parse stdout to find out why the command
> failed
>  
>  This is the output from --help:
>  
>  [hughsie@hughsie-laptop ~]$ pk-debuginfo-install --help
>  Usage:
>    pk-debuginfo-install [OPTION...]
>  
>  PackageKit Debuginfo Installer
>  
>  Help Options:
>    -h, --help           Show help options
>  
>  Application Options:
>    -v, --verbose        Show extra debugging information
>    -s, --simulate       Don't actually install any packages, only simulate what would
> be installed
>    -n, --no-depends     Do not install dependencies of the core packages
>    -q, --quiet          Do not display information or progress 
Something is missing. You probably meant

pk-debuginfo-install [OPTION...] PACKAGE_NAME


>  And in the man page (man pk-debuginfo-install) more details are
> exposed:
>  
>  RETURN VALUES
>         0   Success, at least one package was installed
>         1   Failed with miscellaneous internal error.
>         3   Failed to get software sources list.
>         4   Failed to enable software sources.
>         5   Failed as nothing to do.
>         6   Failed to find dependant packages.
>         7   Failed to install.
>         8   Failed to disable software sources.
>  
>  But it's pretty much a drop in replacement for debuginfo-install from an end-user
> perspective. 
Can you implement an -y switch to be even more compatible?

Currently it is accepted but ignored, I still get confirmation prompt:

# pk-debuginfo-install -y -- unzip-5.52-10.fc11.x86_64
1. Getting sources list...OK. Found 10 enabled and 34 disabled sources.
2. Finding debugging sources...OK. Found 6 disabled debuginfo repos.
3. Enabling debugging sources...OK. Enabled 6 debugging sources.
4. Finding debugging packages...Failed to find the package : more than
one package found for 
Failed to find the package : more than one package found for 
OK. Found 1 packages:
5. Finding packages that depend on these packages...OK. Found 2 extra
packages.
Found 3 packages to install:
1	unzip-debuginfo-5.52-10.fc11(x86_64)	fedora-debuginfo
2	bash-debuginfo-4.0-8.fc11(x86_64)	updates-debuginfo
3	bash-debuginfo-4.0-8.fc11(x86_64)	updates-debuginfo
6. Installing packages...
Starting install
[=========================]         
The following packages have to be installed:
 unzip-debuginfo-5.52-10.fc11.x86_64	
 bash-debuginfo-4.0-8.fc11.x86_64	
Proceed with changes? [N/y] 



-- 
vda

[Richard Hughes]

----- "Denys Vlasenko" <dvlasenk(a)redhat.com&gt; wrote:
>  Something is missing. You probably meant
>  pk-debuginfo-install [OPTION...] PACKAGE_NAME 
Patch welcome :-)

>  > But it's pretty much a drop in replacement for
> debuginfo-install
>  from an end-user perspective.
>  
>  Can you implement an -y switch to be even more compatible? 
Sure, I'll stick it on my TODO. It should only take a few minutes.

Richard.

[Richard Hughes]

----- "Richard Hughes" <rhughes(a)redhat.com&gt; wrote:
>  Sure, I'll stick it on my TODO. It should only take a few
> minutes. 
commit 0c5c33dde299e018818a568f9756c0215239ba83
Author: Richard Hughes <richard(a)hughsie.com&gt;
Date:   Wed Sep 30 16:18:56 2009 +0100

    debuginfo-install: add a --noninteractive (short option -y) command line option to
supress the simulate-before-install behaviour

You can find the code in PackageKit git master. Yell if you want a F12 rpm.

Richard.