Spurious test failure in 21.4

[Jerry James]

I requested a hold on my recent monster 21.4 patch because I thought it
introduced a test failure.  It didn't.  (I also thought it was an md5
test that had failed, but that's because my eyes didn't line things up
right.  It was actually a mule test failure.)

In tests/automated/mule-tests.el, the next to last block tests
file-system character conversion.  The last part of that test creates a
directory, tries to cd to it, and then checks the value of pwd.  The
first time I ran the 21.4 tests, that part failed.  It failed, because
the directory already existed.  I don't know why; perhaps it was left
over from a failing run with 21.5.  In any case, make-directory signals
a file-error when the directory already exists, thus prematurely
terminating the test.  Later, when I figured that out and nuked
everything in my temp directory, a rerun of the tests with the same
binary passed all of the mule tests.

I would propose that we delete-directory right before trying the
make-directory, except that delete-directory signals a file-error when
the directory does not exist.  We could insert this line right before
the make-directory line:

(condition-case nil
    (delete-directory (concat (temp-directory) "/\260\354"))
  (file-error nil))

We could also wrap the make-directory in a condition-case like this, but
then we don't find out about failures of make-directory for other
reasons.
-- 
Jerry James, Assistant Professor        james at xemacs.org
Computer Science Department             http://www.cs.usu.edu/~jerry/
Utah State University