openoffice-1.1 build more broken than usual on 5.2-CURRENT

[Don Lewis]

On 12 Aug, Bartosz Fabianowski wrote:
>> gmake[3]: Entering directory `/usr/ports/editors/openoffice-1.1/work/mozilla/work/mozilla/nsprpub/config'
>> gcc32 -o now.o -c -O -pipe  now.c
>> now.c:45:2: #error "Architecture not supported"
>> now.c:123:2: #error "Architecture not supported"
> [snip]
> 
>> I think the trigger for the breakage was the recent changes to make's
>> handling of MAKEFLAGS in 5.2-CURRENT.
> 
> I have been wondering for a while what change in -CURRENT actually broke 
> the openoffice-1.1 port. When I last tried to compile the port on my 
> system, it died at the very same point as your build. I was able to get 
> mozilla to compile by simply cd'ing into the slave port directory and 
> running "make clean && make" in there. So, the problem must be with some 
> flag being passed from the main port's make to the mozilla slave port's 
> make.
> 
> Now, the curious thing here is that the ports actually use gmake. How 
> can that be affected by changes to FreeBSD's own make?

The change to FreeBSD's make is supposed to have made it conform to
POSIX.  When CFLAGS (or any other variable) is specified on the command
line, this variable assignment is also stashed in the MAKEFLAGS
environment variable, which gets passed to all the submakes, and keeps
CFLAGS (or whatever) from being modified by doing variable assignments
in the Makefiles.  Maybe gmake already behaved this way and the only
reason we didn't have this problem before was that FreeBSD's make didn't
pass MAKEFLAGS through the environment to the top level gmake.