bug#25326: 26.0.50; Testcover error when a defcustom value changed

[Gemini Lasswell <gazally@runbox.com>]

Testcover incorrectly marks symbols declared with defcustom as `1value',
which causes it to signal an error if a customizable value is changed
between uses by code under coverage by Testcover.

This is easy to reproduce using the tests for iCalendar. To reproduce,
starting in your Emacs source directory, with emacs -Q if you like:

M-x testcover-start RET lisp/calendar/icalendar.el RET
C-x C-f test/lisp/calendar/icalendar-tests.el RET
M-x eval-buffer RET
M-x ert RET RET

Result: Every test that varies a customization variable fails with the
message "Value of form marked with ‘1value’ does vary"

There was some discussion of this problem on emacs-devel in 2012 under
the title "testcover: setf-method and treatment of `defcustom'"
including a patch to make Testcover treat defcustom like defvar, which
was never applied, as well as mention of pros and cons of fixing it that
way.

I think that it is best to treat defcustom like defvar, which while it
will create more tan splotches on code where behavior under
customization doesn't get tested, it should also encourage people to
write tests that exercise their customization options.

Here's the patch:

  --- ../emacs/trunk/lisp/emacs-lisp/testcover.el       2012-04-20
19:50:27.000000000 +0200
  +++ /home/stefan/tmp/testcover.el     2012-09-10 08:58:01.000000000 +0200
  @@ -297,7 +297,7 @@
          (push (cadr form) testcover-module-1value-functions))
         (when (eq val 'maybe)
          (push (cadr form) testcover-module-potentially-1value-functions)))
  -     ((memq fun '(defconst defcustom))
  +     ((eq fun 'defconst)
         ;;Define this symbol as 1-valued
         (push (cadr form) testcover-module-constants)
         (testcover-reinstrument-list (cddr form)))

If this patch is applied, there is a workaround for this bug in
lisp/textmodes/rst.el which could be removed.