Hello Valentin,
On Thu, Nov 06, 2025 at 02:56:59PM +0100, Valentin Haudiquet wrote:
>
> Running autopkgtests on the latest inetutils version fails on Ubuntu.
> (this is inetutils 2.6-4 from Debian).
>
> Full log available:
> https://autopkgtest.ubuntu.com/results/autopkgtest-resolute-vhaudiquet-inetutils-lp2130124/resolute/amd64/i/inetutils/20251105_145814_d8bd2@/log.gz
>
> The error seems to originate from the counting method of successes in
> the script.
I think that this is a deliberate error introduced to propagate a test
failure, according to the following comment inside the test script:
```
# Provoke a later total miscount, if message counts do not match now.
```
> 284s + test 36 -eq 33
> 284s + echo Registered 36 successes out of 33.
> 284s Registered 36 successes out of 33.
> ...
> 284s + [ 36 -eq 33 ]
> 284s + echo Failing some tests.
> 284s Failing some tests.
>
> Maybe that is a real bug, and a message is really received twice where
> it should not be, but maybe this is just a counting error in the test
> script (IMHO that counting formula seems really complicated, and I'm
> not sure I understand what is going on).
> I tried to investigate more but am not familiar with the intrinsics of this.
I see several instances of the following two log lines in the autopktest
logs (only difference is in the second of the timestamp):
```
284s Nov 5 14:53:25 autopkgtest syslogd: /dev/sshd: No such file or directory
284s Nov 5 14:53:26 autopkgtest syslogd: /dev/sshd: No such file or directory
```
That is not expected. I do not see these log messages in my test runs.
The configuration used for the test should not reference "/dev/sshd".
> [...]
> It seems that this specific test is skipped on Debian, which is why
> they don't see the issue. If you have a fix, I can send it to Debian
> too.
On Debian 11.11, the syslogd test PASSes for me both with and without
VERBOSE=yes, when run from the release tar-ball.
> If you want to reproduce, running syslogd.sh test should be enough,
> [...]
> Tell me if you need more details or logs.
> I can try to make a reproducer on a direct upstream build if needed,
> and not inside autopkgtests.
The test creates two configurations, the errors only occur with the
first one. I do not know if the generated configuration is faulty, but
the logs look as if syslog were configured to send output to "/dev/sshd".
Using NOCLEAN would not help to get this first configuration, since the
second generated configuration overwrites the first one, and only the
first one results in problems. The attached patch adds the generated
configurations to the test log file when the test is run in verbose mode.
It would be great if you could run the Syslog test in verbose mode
with this patch applied, and provide the resulting logs (e.g., the file
"tests/syslogd.sh.log").
Thanks,
Erik
diff --git a/tests/syslogd.sh b/tests/syslogd.sh
index 4d765b49..6099ab83 100755
--- a/tests/syslogd.sh
+++ b/tests/syslogd.sh
@@ -355,6 +355,14 @@ if [ -n "$REMOTE_LOGHOST" ]; then
EOT
fi
+if [ -n "${VERBOSE+yes}" ]; then
+ cat <<-EOT
+ -------------- First Syslog configuration. ----------
+ `cat "$CONF"`
+ -----------------------------------------------------
+ EOT
+fi
+
# Attempt to start the server after first
# building the desired option list.
#
@@ -503,6 +511,19 @@ cat > "$CONFD/local0" <<-EOT
local0.=notice $OUT_LOCAL0
EOT
+if [ -n "${VERBOSE+yes}" ]; then
+ cat <<-EOT
+ ------------- Second Syslog configuration. ----------
+ ---------------------- syslog.conf ------------------
+ `cat "$CONF"`
+ -------------------- syslog.d/debug -----------------
+ `cat "$CONFD/debug"`
+ -------------------- syslog.d/debug -----------------
+ `cat "$CONFD/local0"`
+ -----------------------------------------------------
+ EOT
+fi
+
# Use another tag for better discrimination.
TAG2="syslogd-reload-test"
