bug#66782: 29.1; ERT tests reports test redefined depending on loading sequence

[Eli Zaretskii <eliz@gnu.org>]

> From: Xiyue Deng <manphiz@gmail.com>
> Date: Fri, 27 Oct 2023 13:59:07 -0700
> 
> As you can see there's only one `first-test' defined.  The error message
> is misleading.
> 
> A real world example of this can be found in lsp-mode, where
> test/lsp-clangd-test.el[1] requires test/lsp-integration-test.el[2].
> See also the discussion on an Debian bug[3].

If test2 requires test1, why are both of them explicitly run from the
command line?  Isn't that redundant, since running test2 will also run
the tests defined by test1?

> However, I'd like to see whether upstream considers this type of usage
> well-formed and should be supported.  If not, upstream should give a
> warning on such usage, such as printing a warning when requiring other
> modules that has `ert-deftest'.  Meanwhile, an improved error message
> would also be great.

I could agree to improving the error message, but I don't see why we
should do anything beyond that, FWIW.

Adding Mattias, who added this check 2 years ago.