rpmbuild sometimes is killed by SIGPIPE with bad dependency generator sciptlets

[mlschroe]

This is a bit of a corner case, but I just received a bug report that builds sometimes terminate for a specific package. It turned out the package contained this little gem:

%define __perllib_provides /bin/true

Anyway, I don't think rpm should be killed by this. Once upon a time getOutputFrom() had code that made it ignore SIGPIPE, but it seems it was removed in commit 375a6b5

This is probably not urgent to fix, just filing so that you know about this race.

[mlschroe]

For reference: https://bugzilla.suse.com/show_bug.cgi?id=1220213

[pmatilai]

Oh, that sounds a whole lot like #2470

[pmatilai]

The flaky test-case is:

#!/bin/sh
echo 'bad = *'
EOF
chmod a+x "${RPMTEST}"/tmp/bad.req

runroot rpmbuild -bb --quiet \
                --define "__script_requires /tmp/bad.req" \
                /data/SPECS/shebang.spec

The common theme being a tiny generator that outputs little to nothing.

[mlschroe]

You probably already know this, but rpmbuild is killed by the write() system that writes the filelist to the generator being done after the child has exited.
I think the only way to prevent this is to ignore SIGPIPE before calling write(). Everything else like the select() thing we currently do is not race free.