New Test for file modification

[Stephen J. Turnbull]

I don't have time to actually generate patches right now, but this can
wait a week to ten days, right?  If somebody's in a big hurry, here
are some hints.

Also, as long as we're here, it has long been my opinion that there
should be a single test suite, with individual tests conditionalized
on version as appropriate.  This could be packaged, which would also
make test-harness.el available to packages for their own automated
tests.

However Martin was quite strongly opposed, on the grounds that tests
are very often implementation specific.  My response to that (at the
risk of understating his reasons) is that where code can be shared, it
should be, and where it can't be, it should be explicitly marked as
such.  The one test suite, version-conditional tests satisfies both
criteria IMO.

Comments?

Vin Shelton writes:

 > So, I think there are 3 things wrong here:
 > 
 > (1) [Minor] the "Wrote /home/acs/test-revert-buffer-resets-modiff"
 > should be suppressed.

Wrap the call to `save-buffer' in a call to `Silence-Message'.  This
may have been intentional to ensure a reminder to fix:

 > BTW, why is it writing this file to my home
 > directory?  Shouldn't the test use some temporary directory  or the
 > build or test directory?

Yes.  I couldn't remember whether `make-temp-name' was the sanctioned
(robust and secure) way to do this, so I punted.

 > (2) The actual unexpected errors: "Unsupported Unicode code point"...

If Aidan wants to work on this, he probably knows more about it than I
do offhand.  If he doesn't, I'll get to it.

 > (3) The fact that the summary does not report the unexpected errors.

This may be hard to do well.  I'll take a look at it.

 > Unexpected error (error "Can't activate input method
 > `german-postfix'") while executing interpreted code.

This is due to the fact that the Makefile suppresses adding packages
to the load-path.  Probably the best thing to do is have a
package-dependent-tests.el, which is run as a separate command.