Using ccache for build{world, kernel} (was: can't compile -CURRENT since 2 days ago)

On Tue, Nov 15, 2005 at 06:11:20AM -0500, Nicolas Blais wrote:
# > Well, if ccache was clever enough to support kernel builds then it would
# > be ok. But since I've never managed to get it to work reliably I use:

Okay, getting the full effect of ccache for a buildworld and buildkernel
requires solving two problems:

1) 'make buildworld' switches from the compilers found in /usr/bin to
   the ones it built below /usr/obj/usr/src/tmp/usr/bin at some point in
   time (simply by prepending this directory to PATH). 'make buildkernel'
   uses the new compiler right from the start. Make ccache somehow do
   the same.

2) Have ccache ignore (i.e. not hash) the modification time of the compiler
   because this changes with every new 'make buildworld'. Otherwise you
   invalidate your cache with each buildworld.


#1 is done with

 make CC="/usr/local/bin/ccache cc" CXX="/usr/local/bin/ccache c++" buildworld
 make CC="/usr/local/bin/ccache cc" CXX="/usr/local/bin/ccache c++" buildkernel

and NOT using any CCACHE_PATH, so ccache scans PATH for the real
compiler. This automatically uses /usr/bin/{cc,c++} in the early
buildworld stages, and the new compilers the moment PATH is prepended
with the correct directories by 'make buildworld'.


#2 is done with this patch (a week ago I sent it to the ccache developer
mailing list bugs_at_ccache.samba.org in order to hopefully get it
committed to the release version; so far no reply... maybe you guys join
me in bugging them about it :-)

Put this in /usr/ports/devel/ccache/files/patch-nohash-size-mtime and
reinstall the port. Then export CCACHE_NOHASH_SIZE_MTIME=y during your
makes for buildworld and buildkernel. This melts my buildworld time
from 2700s (real_at_~100%CPU) to under 1000s and buildkernel from 800s
to 250s, so you gain about a factor 3 for ab-initio builds. Sweet!


--- ccache.c.orig	Mon Oct 31 17:38:21 2005
+++ ccache.c	Mon Oct 31 17:52:31 2005
_at__at_ -331,8 +331,10 _at__at_
 		hash_string(str_basename(args->argv[0]));
 	}
 
-	hash_int(st.st_size);
-	hash_int(st.st_mtime);
+	if (!getenv("CCACHE_NOHASH_SIZE_MTIME")) {
+		hash_int(st.st_size);
+		hash_int(st.st_mtime);
+	}
 
 	/* possibly hash the current working directory */
 	if (getenv("CCACHE_HASHDIR")) {
--- ccache.1.orig	Mon Oct 31 18:08:59 2005
+++ ccache.1	Mon Oct 31 18:09:02 2005
_at__at_ -234,6 +234,16 _at__at_
 incorrect setting of this debug info rarely causes problems\&. If you
 strike problems with gdb not using the correct directory then enable
 this option\&.
+.IP
+.IP "\fBCCACHE_NOHASH_SIZE_MTIME\fP"
+This tells ccache to not hash the real compiler's size and modification
+time. Normally this is the mechanism to detect compiler upgrades.
+There are situations however, where even though the compiler's size or
+modification time has changed you can safely use the cached objects
+(e.g. if as part of your build system the compiler is built as well
+and the compiler's source has not changed; or if the compiler has only
+changes that do not affect code generation). Use this feature only if
+you know what you are doing.
 .IP 
 .IP "\fBCCACHE_UNIFY\fP" 
 If you set the environment variable CCACHE_UNIFY

Regards,

	Jens
-- 
Jens Schweikhardt http://www.schweikhardt.net/
SIGSIG -- signature too long (core dumped)