idea: include/exclude files by mime type (use magic like file(1) does)

[Shawn Rutledge]

I have been thinking this would be a nice feature: exclude files by
type rather than just regular expressions.  Mostly because binaries
don't have any extension on Linux, so they aren't recognizable just by
name.  At work I rsync a source directory for doing builds in a chroot
environment for the target system, and I'd like to exclude binaries
that happen to be in the tree, on the host system.  (It's easy enough
to exclude the .o files but not the final linked executables.)
Likewise at home, I was thinking of rsyncing my home directory between
a couple of machines, one of which is 32-bit, so it doesn't make sense
to sync any 64-bit binaries from other machines to that one.  So I
would like to exclude "ELF executables" or "application/x-executable"
from the rsync.

I think for speed, if the --exclude-mime option is given, rsync would
need to process the other include/exclude patterns first, then as a
final pass, basically look up the magic numbers the same way that
file(1) does for each file that it plans to transfer, and exclude
files on that basis.  For --include-mime, it wouldn't be as efficient
(would have to get the mime type for every candidate file).