[ccache] [PATCH] Potentially produce better relative pathnames when symlinks are

Mike Crowe mac at mcrowe.com
Mon Mar 14 05:03:27 MDT 2011


On 2011-02-28 13:30, Mike Crowe wrote:
>> When making relative pathnames try and do so both before and after
>> calling realpath and pick the version that is the shortest. This helps
>> if the build starts from a path that contains a symlink and a Makefile
>> somewhere decides to use realpath to generate absolute paths.
 
On Sun, Mar 13, 2011 at 11:20:07PM +0100, Joel Rosdahl wrote:
> First I thought that your use case is this: There's a file named
> /dir1/src/file.c, /dir2 is a symlink to /dir1 and CCACHE_BASEDIR=/dir2.
> The Makefile in /dir1/src runs "gcc -c $(realdir file.c) -o file.o", so
> when invoked in /dir2/src, make will execute "gcc -c /dir1/src/file.c -o
> file.o", and ccache will thus not detect that /dir1/src/file.c should be
> rewritten since it doesn't match CCACHE_BASEDIR.

Well, that certainly is a problem that I would also like to resolve
but you are correct in saying that the patch does not address it. I
ended up deciding that the only way to solve it would be to invent
CCACHE_BASEDIRS to support multiple base directories. I have a patch
in development that does this.

> But your patch doesn't address that problem, so that's not the use case.
> Let's try again. There's a file named /dir/1/src/file.c, /dir/2 is a
> symlink to /dir/1 and CCACHE_BASEDIR=/dir. The Makefile in /dir/1/src
> runs "gcc -c $(realdir file.c) -o file.o", so when invoked in
> /dir/2/src, make will execute "gcc -c /dir/1/src/file.c -o file.o" and
> ccache will rewrite /dir/1/src/file.c to ../../2/src/file.c. With your
> patch, /dir/1/src/file.c will instead be rewritten to file.c, thus in a
> sense normalizing the path so that there will be a cache hit if you
> later build in /dir/1/src. Is that the use case?

Yes, this is the use case that is addressed by the patch. The reason
it is a problem for me is that we share a ccache among multiple users
and the rewritten path ends up being unique for each user (without the
patch.)

I'm open to suggestions if you think there is a better way to fix
either or both problems.

Thanks.

Mike.


More information about the ccache mailing list