re: lossage with successive builds on netbsd-6 branch

[matthew green <mrg%eterna.com.au@localhost>]

> When I do two (or more) build.sh runs in a row switching MACHINE with -m
> only the first one runs to completion. E.g.:
> 
> ./build.sh -j8 -U -O obj -m i386 tools release >& build.i386.log
> ./build.sh -j8 -U -O obj -m sparc tools release >& build.sparc.log

isn't this telling it to use the same objdir for both builds?

i can't imagine this can be safe.  "cleandir" for the second one
won't clean objects from the prior build it doesn't know about,
and it wouldn't surprise me if some how this causes the build
to break.  it will depend upon what make and the compiler end up
doing, i think.

what actually fails?

> The same sequence works for netbsd-5.
> 
> The value of -O doesn't seem to matter.  Building without -O doesn't make
> a difference either.
> 
> I'm doing this on a netbsd-6/amd64 machine.
> 
> Is this known lossage?  If so, is this expected lossage?

i strongly recommend not using the same objdir for two different
builds.

things you could try to figure out what's wrong:

- does 'nbmake-i386 cleandir' remove all the objdir files?
(ie, is it currently correct?  it's usually missing something.)

- after the sparc 'cleandir' phase, stop the build and go see
what is still there.


.mrg.