[ccache] Why not using MD5?

[Egmont Koblinger]

On Mon, Nov 15, 2004 at 06:25:49PM +1100, Martin Pool wrote:

> The other factor is whether the cache will grow to include a million
> files.

We're building a distribution using a build system completely created by us
and it intensively uses ccache. We have approx 1000 packages, including, of
course, all the standard linux applications.

Through about 1.5 years we've had one huge ccache pool for all our packages
and sometimes manually dropped old files after a change that caused old
files not to match anymore (gcc, glibc etc. upgrade). The ccache size was
around 5-10 GB.

Recently we changed to use separate pool for each package, but for other
reasons, not because we've met any collision. Fortunately we didn't meet
any, but I didn't even realize that we've been not far from it...

(If you're curious I can easily count the number of entries in our ccache
pools later this day.)

So there's one more vote here for md5. It doesn't hurt if ccache is prepared
for "extreme use". :-)) And it's perfect time to do it now since the
2.4->2.5 version number change would reflect the md4->md5 change :-))))))))


PS. AFAIK no-one has ever found two different files with the same md5. Am I
right?


bye,

Egmont