FC-4 Extras buildsystem breakage details

[Dan Williams]

Hi,

Some of you have noticed and pointed out the fact that the FC-4 Extras
target wasn't successfully building any packages and wasn't giving much
failure information.  So after a bit of detective work this morning
we've figured out that:

1) boa, which provides "webserver", is the immediate culprit of the
breakage because it does not hardcode its UID for /usr/sbin/useradd.
This leads useradd to select its first default of 100 for the UID, which
is a direct conflict with mock, which hardcodes 100.  It's unclear if
mock should use a different UID or not.

2) dap-server, which appeared on 20-Feb-2006 around when the problems
started, is the real culprit.  dap-server requires webserver, which
pulls in 'boa' because it has the shortest package name of those
packages which provide 'webserver'.  dap-server is getting pulled into
all buildroots because it mistakenly provides built-in versions of
perl-HTML-Parser modules

3) plague-builder appears to be dropping output from mock that's printed
to stderr, causing the real error message not to appear in the logs

To correct the problem, dap-server has been temporarily removed from
Extras repositories, since it even blocks rebuilds of a fixed dap-server
itself.  When the buildsystem local repository rsyncs the new repo data
down, dap-server will be rebuilt and builds on the fc-4 target should be
able to continue as normal.

We should also start up a discussion about how to prevent problems of
this nature in the future.  We would not have noticed the problem had
boa not been conflicting with mock, but the real issue of dap-server
providing perl-HTML-Parser modules itself would still have been a
problem, albeit a silent one.  If there are scripts, rules, and/or
pre-submit checks which could be performed at various stages, it might
go a long way towards prevent this sort of breakage in the future.

Dan