From owner-freebsd-current  Thu Apr 18 08:25:51 1996
Return-Path: owner-current
Received: (from root@localhost)
          by freefall.freebsd.org (8.7.3/8.7.3) id IAA10887
          for current-outgoing; Thu, 18 Apr 1996 08:25:51 -0700 (PDT)
Received: from godzilla.zeta.org.au (godzilla.zeta.org.au [203.2.228.19])
          by freefall.freebsd.org (8.7.3/8.7.3) with SMTP id IAA10878
          for <current@freebsd.org>; Thu, 18 Apr 1996 08:25:43 -0700 (PDT)
Received: (from bde@localhost) by godzilla.zeta.org.au (8.6.12/8.6.9) id BAA00627 for current@freebsd.org; Fri, 19 Apr 1996 01:25:09 +1000
Date: Fri, 19 Apr 1996 01:25:09 +1000
From: Bruce Evans <bde@zeta.org.au>
Message-Id: <199604181525.BAA00627@godzilla.zeta.org.au>
To: current@freebsd.org
Subject: #include bloat
Sender: owner-current@freebsd.org
X-Loop: FreeBSD.org
Precedence: bulk

Long ago, I introduced the #include of <machine/cpufunc.h> in <sys/systm.h>
to make all the machine-dependent inline functions visible without changing
many files.  The list of inlines has now grown so large that compiling them
now takes about 10% of the kernel build time:

Times to build a -current GENERIC from a freshly configured directory
immediately after booting:
      485.41 real       380.97 user        27.83 sys

Times to do the same thing with all the inlines in cpufunc.h replaced
by prototypes:
      446.99 real       346.64 user        26.44 sys

Times to build GENERICBT in 1.1.5 (on the same hardware with the same
compiler, but on a different file system and not immediately after
booting):
      213.92 real       163.58 user        16.50 sys
I guess that it would take 150-180 seconds on the same hardware under
1.1.5, mainly because gcc was faster.

Sigh.  I think only about 50% of the extra time is justified by features.

Bruce