With help from pkgsrc-users and a colleague, I've partially figured out
a ccache mystery. All is with HEAD pkgsrc. On a Mac, I started getting
constant cache misses. I tested on NetBSD/i386 5.1ish, or so I thought,
and seemed to have the same problem.
Note that CCACHE_HASHCC was in 2.4 and has been withdrawn from 3.1.4.
On the mac, putting CCACHE_COMPILERCHECK in the environment resolves the
issue, and caching works as expected (but the old cache contents won't
hit, since they used a different compiler version hash).
On NetBSD, the following doesn't seem to have any effect.
I don't understand why this was ever needed, because work/.ccache/bin
seems to be a symlink to ccache, and the real compiler doesn't change.
But I don't fully understand the ccache/gcc layering.
So, can anyone explain why CCACHE_HASHCC was necessary? And if so,
whether they think CCACHE_COMPILERCHECK should be set?
--- ccache.mk.~1.32.~ 2009-10-05 08:36:46.000000000 -0400
+++ ccache.mk 2011-03-30 16:38:04.000000000 -0400
@@ -119,9 +119,12 @@ BUILD_DEPENDS+= ccache-[0-9]*:../../deve
. endif
# Override the compiler-specific hash with the version string for the
-# compiler.
-#
+# compiler (ccache 2.4).
PKGSRC_MAKE_ENV+= CCACHE_HASHCC=${CC_VERSION_STRING:Q}
+# Similar for modern ccache (ccache 3.1.4). We ask ccache to run the
+# compiler with -v, and treat matching output as a matching compiler.
+PKGSRC_MAKE_ENV+= CCACHE_COMPILERCHECK="%compiler% -v"
+
PKGSRC_MAKE_ENV+= CCACHE_DIR=${CCACHE_DIR:Q}
# Create symlinks for the compiler into ${WRKDIR}.
Attachment:
pgpmpD4ghvG9v.pgp
Description: PGP signature