https://bugs.kde.org/show_bug.cgi?id=511717

--- Comment #21 from Philippe Waroquiers <[email protected]> ---
I downloaded and build knot.
when I start the ctl/concurrent test, I see a set of memcheck processes
running.

After some time, the test stops with an error message telling:
  ... * ctl/concurrent   FIALED (Can't start all servers)

Looking in the knottest-last..../knot1, I see a set of output files, but none
of them contains a segv error like the below.
There are a bunch of stdout/stderr/valgrindXXXXXXX   output files (not clear
what is the XXXXXX number).
So, I could not reproduce.

However, the valgrind output files look incomplete.
It looks like several valgrind processes are started in parallel (and you use
--trace-children=yes), and likely they all have the option
     --log-file=...../valgrind

Having a single log file for several processes is giving problem.
I have changed in test.py the log file argument to use a suffix _%p  (see
valgrind user manual for --log-file
This has then shown that the --trace-children-skip=/usr/*sh,/bin/*sh  is not
matching /usr/bin/bash
I tried to add /usr/bin/*sh  and /usr/bin/bash  but still I find many
valgrind_%p files with bash being traced.

I added some more debug log in valgrind to see what happened there.
Then it looks like several valgrind processes are mixing their standard error
into the same file (stderr)
as the msg I added was only partial and this stderr contained a mix of
different pid output.

So, at this point:
* I cannot reproduce the problem on Ubuntu 25.10 (while knot-dns seems to at
least do something with ctl-concurrent)
* likely the way parallel valgrind processes are launched could be changed to
have their stdout/stderr and --log-files being unique
  (if my assumption that these are mixed is correct).
 * would be nice to see if the problem reproduces with the --tool=none
    and/or  with  memcheck + --vgdb=full (this will run slower but ensures that
all registers are up to date at each instructions
   in case the problem is due to some register optimisation done by valgrind
interacting badly with gdb)

-- 
You are receiving this mail because:
You are watching all bug changes.

Reply via email to