Date:        Fri, 2 Sep 2005 19:35:01 +0100
    From:        David Laight <david@l8s.co.uk>
    Message-ID:  <20050902183501.GH996@snowdrop.l8s.co.uk>

  | I did a full build of current for i386 over wednesday and thursday
  | (interrupted by a thunderstorm and short power outage) and the build
  | I did fits fine.

It doesn't fit for me.

  | It is worth checking that you've not set any compilation options in
  | /etc/mk.conf.  They will cause the build to explode
  | (and since the ramdisk is built with the standard libc.a it is difficult
  | to avoid that - unless the relevant parts of libc are compiled again).

I am certain that I have no unusual config flags, I build with
MAKECONF=/dev/null (and env -i) to make sure that there's no
external influence.

My build command is (after various setup nonsense) ...

EMPTY=`mktemp -d /tmp/build-XXXXXXX`

env -i \
        PATH=/sbin:/bin:/usr/sbin:/usr/bin \
                "USER=$USER" \
                "HOME=$EMPTY" \
                "LOGNAME=$LOGNAME" \
                "DISPLAY=$DISPLAY" \
                "MAKECONF=/dev/null" \
        /bin/sh build.sh "$@" \
                -V "CHECKFLIST_FLAGS=-e"                \
                        -D "${TARGET}"                  \
                        -O "/usr/obj/$REL"              \
                        -R "$SNAP"                      \
                        -T "/usr/obj/$REL/tools"        \
                        -X "$TARGET/usr/src/xsrc"       \
                -u -x                                   \
                        $BUILDING

The HOME= new garbage empty directory may no longer be needed, that
was becanse onf oe the X program builds required HOME set in order
to build (I forget which now) - something required DISPLAY as well
I think (if it is there, it is because something demanded it...)

"$@" is empty in any normal build I do.

BUILDING=release (almost always).
REL=current (here, it can also be 3.0 2.0 ...)
TARGET=/release/current which was empty, aside from usr/src/*

I've been seeing the "6144 bytes too big" for at least the last couple
of days (i386 build of course).

The most recent build was done on a new system (running 3.0_BETA as of
a week ago or so) with completely empty target and obj directories
and with  "cvs co -P" (for src & xsrc) done into clean source directories
minutes earlier.  (That is, the -u cannot possibly be influencing anything,
as there was absolutely nothing previously compiled).

kre