if external hash command fails for any reason, fall back to internal hashing

This way, if a system's sha1sum etc is broken, it will be tried if
git-annex was built to use it, but at least it will fall back to using
internal hashing when it fails.

A side benefit of this is that hashFile consistently throws an IOError if
the file is unable to be read. In particular, if the disk is failing with
IO errors, and external hash command is used, it used to throw a user error
with the error message from externalSHA. Now, the external hash command
will fail, that message will be printed as a warning, and it'll fall back
to the internal hash command. If the disk IO error is not intermittent, it
will re-occur, and so an IOError will be thrown.

Of course, this can mean it reads a file twice, but only in edge cases.

0 files changed, 0 insertions, 0 deletions