QA » openQA Project

Observation¶

We sometimes see things like

Reason: isotovideo died: Can't locate auto/NetAddr/IP/InetBase/AF_INET6.al in @INC

in the info panel of an openQA test.
See #124913 for example.

This message is distracting because it is actually not a fatal message, but from an eval:

https://metacpan.org/dist/NetAddr-IP/source/Lite/Util/lib/NetAddr/IP/InetBase.pm#L81

require Socket;

*AF_INET = \&Socket::AF_INET;

if (eval { AF_INET6() } ) {       # line 81
  *AF_INET6 = \&Socket::AF_INET6;
  $emulateAF_INET6 = -1;                        # have it, remind below
}

https://github.com/os-autoinst/os-autoinst/blob/master/isotovideo#L115

$SIG{__DIE__} = sub ($e) { $fatal_error = $e };

So when NetAddr::IP::InetBase calles the eval { AF_INET6() } (thanks to AutoLoader/AutoSplit this actually tries to load the subroutine AF_INET6 from a generated file AF_INET6.al) it is actually fine with it failing, but we still call our signal handler. We could check with $^S if we are in an eval and just discard the message if it's true - but see the notes section why that might not be nice. Then we will get rid of those distracting error messages in the Reason, which we have seen several times before.

Acceptance Criteria¶

AC1: We don't display non-fatal distracting error messages anymore

Notes¶

• $^S
• %SIG

The $SIG{__DIE__} hook is called even inside an eval(). It was never intended to happen this way, but an implementation glitch made this possible.
...
Having to even think about the $^S variable in your exception handlers is simply wrong. $SIG{__DIE__} as currently implemented invites grievous and difficult to track down errors. Avoid it and use an END{} or CORE::GLOBAL::die override instead.