Question about test failure on Hydra

[Stephen Berman]

Part of the Dired tests for bug#27243 continues to fail intermittently
on Hydra and I see in the logfiles why it fails, but I don't understand
how the trigger for the failure arises.  Here's the test code (for the
first of the bug cases; the second case has almost the same code and
fails in the same way) up to the point of failure (the should sexp):

(ert-deftest dired-test-bug27243-01 ()
  "Test for https://debbugs.gnu.org/cgi/bugreport.cgi?bug=27243#5 ."
  (let ((test-dir (make-temp-file "test-dir-" t))
        (dired-auto-revert-buffer t) buffers)
    (with-current-buffer (find-file-noselect test-dir)
      (make-directory "test-subdir"))
    (push (dired test-dir) buffers)
    (unwind-protect
        (let ((buf (current-buffer))
              (pt1 (point))
              (test-file (concat (file-name-as-directory "test-subdir")
                                 "test-file")))
          (write-region "Test" nil test-file nil 'silent nil 'excl)
          ;; Sanity check: point should now be on the subdirectory.
          (should (equal (dired-file-name-at-point)
                         (concat (file-name-as-directory test-dir)
                                 (file-name-as-directory "test-subdir"))))

When it fails, Hydra show dired-file-name-at-point returning e.g. this
(it's a temporary directory so the name is different on each test run):

"/build/test-dir-26691mMc/../"

while the concat sexp it's being compared with returns (e.g.) this:

"/build/test-dir-26691mMc/test-subdir/"

IIUC dired-file-name-at-point can only return the above result if point
is on the ".." entry in Dired.  But the sexp (dired test-dir) should
(and in all my test runs does) put point on the first "nontrivial" file
(i.e. it skips "." and "..").  I don't see how the above code can lead
to the failure Hydra reports.  And again, it doesn't always fail in
Hydra.  Does anyone have any idea what's going on here and what to do
about it?

Steve Berman