bug#9531: md5sum: confusing documentation for file type output

[Eric Blake]

On 09/17/2011 07:10 AM, Rüdiger Meier wrote:
> On Saturday 17 September 2011, Reuben Thomas wrote:
> > Secondly, looking at the code, this isn't really what md5sum does.
> > Rather, it prints '*' if in binary mode (--binary, or if O_BINARY is
> > non-zero), and ' ' if in text mode (--text, or if O_BINARY is zero),
>
> BTW I was always confused about the --binary thing.
> If you're going to update man pages anyway it would be nice to point out
> that --binary makes a difference only if O_BINARY exists on the target
> system (which is usually not the case for most users I guess).

I hate the way md5sum currently works.  What we _should_ have done is 
make binary normal, and do a new symbol for O_TEXT being non-zero.  That 
way, the only platforms where it matters (such as cygwin) would default 
to binary, and make it explicit when carriage returns were not being 
considered due to text mode differing from binary mode.

Also, presence of a literal carriage return in a file name must be 
escaped as \r (just as we escape \n and \\ for newlines and unambiguous 
escapes), so that md5sum output on a platform with text files does not 
cause ambiguities when literal carriage returns are eaten or added.

I've had ideas of how to change things for several years now, but never 
enough of an itch to actually code it up.  The biggest problem is the 
fact that it would be an incompatible format to what existing md5sum 
parses, so you'd have to phase it out with options on whether to 
generate new or old style.

--
Eric Blake   address@hidden    +1-801-349-2682
Libvirt virtualization library http://libvirt.org