Re: Add condition-case handler to ert test runner

[Richard H Lee]

On 06/05/2013 15:30, Davis Herring wrote:
> If that's what you want...

Sorry I probably didn't illustrate how ert works properly.

The way ert catches the signals it generates is by setting 
debug-on-error to t and debugger to ert--run-test-debugger, which is its 
own custom debugger. ert--run-test-debugger then records the result of 
the test according to the signal thrown.

So I'm not trying to invoke the debugger. ert itself invokes its 
debugger and always expects it to be invoked upon test failure. And this 
is where the problem lies. In the case where ert happens to be called 
from within a condition-case with matching handler, ert's debugger will 
not be invoked, so the result will not be recorded. ert interprets this 
as a test abort and so will not execute any more tests.


> ...you surely instead want
>
>            (condition-case nil
>              (funcall (ert-test-body (ert--test-execution-info-test
>                                       test-execution-info)))
>              (error (debug))))))

This invokes emacs' standard debugger not ert's ert--run-test-debugger . 
By having both error and debug in the handler clause ensures that the 
signal (error ert-test-failed) is caught and whatever is assigned to the 
variable debugger is called.

And the only way anyone could have figured out how to do this is from 
poking around eval.c . Hence my remark about the documentation.


> The real question is why ERT's signals are reaching IELM if the latter
> invoked the former (whose handlers should thus be innermost).

The answer to this is because ert does not set up a watertight handler 
around the execution of the test. My patch fixes this.


Richard