[Bug default/26941] abidiff assert in build_or_get_type_decl when comparing a library to itself

["woodard at redhat dot com" <sourceware-bugzilla@sourceware.org>]

https://sourceware.org/bugzilla/show_bug.cgi?id=26941

--- Comment #4 from Ben Woodard <woodard at redhat dot com> ---
Just to illuminate how I test this:

I have debuginfod running and have downloaded all the debuginfo packages for
fedora33. My configuration file is included. The basic idea is that I have a
directory called /data/debug which I added to the default directories that I
tell debuginfod to look into. To keep this up to date I also have a script
called 
freshen-debuginfo.sh which does a reposync of the directories to keep them up
to date. I generally run this before a test run.

I have also downloaded flux-core https://github.com/flux-framework/flux-core
and built that in my environment. I setup my path to my installation of flux
and a current build of libabigail and then run "flux start". I will include
these in the reproducer package as well. But you will need to adjust your paths
as well.

flux start launches a nested shell from which I can run many flux jobs. I have
a script that launches all of those jobs called "flux-abidw.sh". Running that
launches all the jobs and queues them up to be run sequentially as processors
are available.

You can see how many jobs failed by simply running "flux jobs -f failed". On my
system a 6 core i7-8700 running at 3.2GHz it takes about an hour for all the
jobs to complete. However, all but a couple of them are done within about 10
min.

To get the backtraces I have a bunch of other scripts which gather the
backtraces for me. To kick off gathering the backtraces there is a script which
is ./flux-gen-bts-abidw.sh which uses the other scripts ./flux-pull-fails-2.sh
gather-bt-abidw.sh to do the work. What you are left with is a list of files
ending in .bt

Then I pick one of them and look at the line causing the failure and then see
how often it is amongst the backtraces. In this case I looked at 
$ cat libTKBool.so.7.4.0.bt
$ grep -l abg-reader.cc:1287 *.bt | wc -l
105

Then I ran a another program called "all-with.sh" which generated the list I
attached to the commebnt 3.

-- 
You are receiving this mail because:
You are on the CC list for the bug.