Corinna Vinschen writes: >> Because of the openssl branch in Git. > > Oh, that was a bad idea. I won't remove the branch for historical > reasons, but we don't really want to link agaionst OpenSSL when we > already link against gcrypt.
OK, makes sense. >> > This was a drop-in replacement for Digest::MD5. Anything else I'll >> > let somebody do who better knows perl. >> >> THen the call $ctx->hexdigest() would need to be replaced by >> $ctx->b64digest(). > > No, it would need to support this additionally. Why? In any case it'd be easy enough to make it switchable. >> > Apart from this, the sah512.sum files are not only generated for the >> > Cygwin release dir, they are generated on the sourceware ftp area >> > system-wide. >> >> I'm only talking about SHA512 in the setup.ini files, though. The >> sha512.sum files have no bearing on that. > > So what exactly is the problem with the longer checksums? I don't > see any problem here, especially when the file is being compressed > anyway. I'm not concerned about the size (although I note that even if compression recognized it's looking at an SHA512 it would at best be able to compress it down to 64 bytes anyway). The difference between 128 and 86 (or 88 padded) characters is pretty noticeable on screen. It's probably a bit unusual that I look at setup files so often (since I generate my own), but it still disturbes me. The other thing (but that really is unrelated, as I could take care of it on other ways) is that I still plan to move to a repository format that allows multiple versions to be active at the same time (via different setup.ini files pointing into the same repo hierarchy) and I was thinking of using the SHA512 to do directory balancing like it's done for the object directory in Git. ~ (2001) du -sh /mnt/mirror/cygwin 52G /mnt/mirror/cygwin ~ (2002) find /mnt/mirror/cygwin -type f |wc 44778 44778 3340383 ~ (2003) find /mnt/mirror/cygwin -type d | wc 11079 11079 602401 ~ (2004) ls -d /mnt/mirror/cygwin/x86/release/* | wc 2071 2071 89831 Using a 6bit key into a two level tree would seem to provide a better balance for this structure than 8bit or I could go single level with a 12bit key. It needs to provide optimum performance over network, so the optimum likely is at a different point than a local file system. Regards, Achim. -- +<[Q+ Matrix-12 WAVE#46+305 Neuron microQkb Andromeda XTk Blofeld]>+ Factory and User Sound Singles for Waldorf rackAttack: http://Synth.Stromeko.net/Downloads.html#WaldorfSounds