StacktraceSource contains errors instead of source code

[Brian Murray]

The StacktraceSource information being added by apport-retrace contains Error messages see the attachment in bug 1459716.
Here's part of it:

#0 0xb7751be0 in __kernel_vsyscall ()
#1 0xb6813057 in __GI_raise (sig=6) at ../sysdeps/unix/sysv/linux/raise.c:55
  [Error: ../sysdeps/unix/sysv/linux/raise.c was not found in source tree]
#2 0xb6814699 in __GI_abort () at abort.c:89
  [Error: abort.c was not found in source tree]
#3 0xb685119e in __libc_message (do_abort=1, fmt=0xb694b444 "*** Error in `%s': %s: 0x%s ***\n") at ../sysdeps/posix/libc_fatal.c:175
  [Error: ../sysdeps/posix/libc_fatal.c was not found in source tree]

I think part of the issue is that apport-retrace only ever searches the srcdir returned by get_source_tree (which is a temporary directory) and doesn't search the sandbox for the filename. Another problem is that find_file_dir is being passed a filename like "../sysdeps/unix/sysv/linux/raise.c" which won't be found even if that were in the srcdir.

[Brian Murray (brian-murray)]

I fixed the filename being searched for via the following.

@@ -178,7 +185,8 @@
         for frame in report['Stacktrace'].splitlines():
             m = src_frame.match(frame)
             if m:
- result += frame + '\n' + get_code(srcdir, m.group(1), int(m.group(2)))
+ filename = m.group(1).split("/")[-1]
+ result += frame + '\n' + get_code(srcdir, cache, filename, int(m.group(2)))

Then a StacktraceSource looked like this:

 #3 0x00007ffff776889d in XtAppMainLoop (app=0x6090e0) at ../../src/Event.c:1554
   [Error: Event.c was not found in source tree]
 #4 0x000000000040198f in main (argc=1, argv=<optimized out>) at ../xeyes/xeyes.c:137
   132:
   133: (void) XtCreateManagedWidget ("eyes", eyesWidgetClass, toplevel, NULL, 0);
   134: XtRealizeWidget (toplevel);
   135: (void) XSetWMProtocols (XtDisplay(toplevel), XtWindow(toplevel),
   136: &wm_delete_window, 1);
   137: XtAppMainLoop(app_context);
   138:
   139: return 0;
   140: }

[Brian Murray (brian-murray)]

Examples of bugs with good StacktraceSources include:

bug 1297212
bug 1061026
bug 1225365
bug 1415515
bug 1304003
bug 1351558

So it seems to work some of the time.

[Brian Murray (brian-murray)]

Looking in my stash of bug report attachments for foundations-bugs it seems that about only 10% don't contain an error in the StacktraceSource.

[Brian Murray (brian-murray)]

Also when working with the xeyes crash I found the missing NextEvent.c file in the source for libx11-1.6.3.

[Martin Pitt (pitti)]

Fixed in trunk: http://bazaar.launchpad.net/~apport-hackers/apport/trunk/revision/2992

Thanks Brian, nicely spotted! I tested this with the usual bash -c 'kill -SEGV $$' and StacktraceSource went from "useless" to "OMG real code!".

[Launchpad Janitor (janitor)]

This bug was fixed in the package apport - 2.18-0ubuntu9

---------------
apport (2.18-0ubuntu9) wily; urgency=medium

  * Revert changes to data/package_hook to include the package version. This
    just hides the problem that somewhere during whoopsie add_package_info()
    is not called. (See LP #1485787)
  * packaging-apt-dpkg.py, is_distro_package(): If there is no origin and
    /etc/system-image/channel.ini exists, assume the package is from a
    read-only system image and accept it as distro package. With this we don't
    need /var/lib/apt/lists/ indexes any more just to confirm the origin.
    (LP: #1489410)
  * Merge fixes from trunk:
    - whoopsie-upload-all: Intercept OSError too (e. g. "No space left on
      device"). (LP: #1476258)
    - apport-retrace: Only consider the file name of a source file, not its
      path; the latter often contains parts like "../" or directories which are
      specific to a build machine. This fixes most broken StacktraceSource
      results. (LP: #1462491)

 -- Martin Pitt <email address hidden> Mon, 31 Aug 2015 11:35:54 +0200