Re: [test] Mark tests with missing dependencies as "expected to fail"

[Eric Schulte <schulte.eric@gmail.com>]

David Maus <dmaus@ictsoc.de> writes:

> Hi all,
>
> Currently tests with missing dependency are silently skipped -- it
> might be worth changing this behavior to not skip them, but mark them
> as expected to fail. You can do this in ERT by placing the
> keyword :expected-result followed by either :passed or :failed before
> the test's body.
>
> Benefit of this is that the tests w/ missing dependencies will show up
> in the ERT result page (with a small letter f) but (obviously) don't
> count as failures.
>
> The following macro is a first shot at a convenient way to define
> tests with dependencies:
>
> #+begin_src emacs-lisp
>   (defmacro org-test-deftest-with-dependency (dependency &rest body)
>     (let* ((docstring (if (stringp (third body)) (list (third body))))
>            (deftest (nthcdr (if docstring 3 2) body))
>            (dependency-p (eval dependency)))
>       `(ert-deftest ,@(append (list (first body) (second body)) docstring)
>            ,@(if dependency-p
>                  '(:expected-result :passed)
>                '(:expected-result :failed (error "Missing dependency")))
>          ,@deftest)))
> #+end_src
>
> Here DEPENDENCY is a simple form that evaluates to non-nil if the
> dependency is met. If marking the tests this way is the agreed way to
> go we can extend the syntax of a dependency to an alist with a
> human-readable description of the dependency as key and a list of
> forms that all must eval to non-nil as value. E.g.
>
> #+begin_src emacs-lisp
>   (defvar test-ob-R-dependencies
>     '(("R executable" (org-test-for-executable "R"))
>       ("Package ESS" (featurep 'ess))))
> #+end_src
>
> And change the expander code to map through this alist and as soon one
> dependency evals to nil sets a variable `dependency-missing' to the
> respective dependency.
>
> Any comments on this?
>

Hi David,

I agree it would be preferable to note that not all tests are run when
dependencies are missing, although I don't think it is extremely
important.  I think some version of the above would be worthwhile if it
could be done in a file-wide manner (as are the current dependency
checks) and wouldn't require duplicating the dependency check or
changing every test form individually.  Perhaps a file-local-variable
could be used to expect failures for every form defined in the file?

Cheers -- Eric

-- 
Eric Schulte
http://cs.unm.edu/~eschulte/