Funny one that caused e.g. unability to properly use %ifs for kernel
rpms.  I don't know if this is reported, but I see this bug for very
long time already, and query for 'rpm' and 'arch' didn't find any
refs for this.

Here it is:

 %define var 1
 %ifarch i386
 %if %{var}
 echo var
 %endif
 %endif

For this, rpm will complain "parseExpressionBoolean returns -1" to the
'%if %{var}' line.  If the order is changed (i.e. %ifarch inside %ifvar),
all works fine.

This bug caused funny bug in kernel rpm.  Redhat places kernel-sources
and kernel-headers to one package by default, but there is a define in
kernel.spec (%define headersinsource 1) that can be changed.  Ok,
I see this thing in spec:

  %files source
  ...
  %{!?headersinsource:%files headers}
  ...

This wont work (no kernel-headers produced even if headersinsource is
set to 0).  If I change this to

  %if ! %{headersinsource}
  %files headers
  %endif

rpm will build for i386 and -headers will be created, but it will
complain to the line `%if ! %{headersinsource}' for e.g. i686
(with the same message).  So, there is no ability currently exists
to make kernel-headers package, while at least one of this bugs not
fixed.

I don't know why it won't build -headers in the first place -- line
  %{!?headersinsource:%files headers}
looks strange to me.  But I can't say that this is an rpm bug also,
as this at least not documented anywhere. So there is may be one bug
(nested %ifs) or two bugs.