ERT: should-error doesn't catch failed assertions

[Sean McAfee <eefacm@gmail.com>]

I've just begin trying to use ERT to write unit tests for my code.  I
got hung up almost immediately when trying to confirm that a particular
function call fails an assertion.  A simple example:

(should-error (cl-assert nil))

The assertion failure escapes the should-error form and causes a stack
trace.

Is there a way to catch assertion failures like this?

The code I'm testing implements a checksum of sorts that can only work
with strings of length exactly 5, ie:

(def my-checksum (str)
  (cl-assert (= 5 (length str)))
  ; ...
  )