Package: Emacs
Version: 30.0.50


`condition-case` has served us well for the usual error handling needs
of common code, but it does not let us capture information about
the dynamic context where the error takes place, such as
capturing the value of dynamically scoped vars or the backtrace.

For that reason, `signal_or_quit` has slowly grown as new needs have
come up to run various kinds of debugger-like operations in
special cases.

The attached patch adds the new special form `handler-bind` which
provides a functionality similar to `condition-case` except that
the handler is run *before* unwinding the stack.

The patch includes a change to `ert.el` to make use of it to capture the
backtrace of errors instead of messing with `debugger` and
`debug-on-error`, which is fiddly and comes with various problems (such
as the fact that it impacts `condition-case-unless-debug`).

It's not completely finished but I'm bumping into a question.
`ert-tests.el` includes the following test:

    (ert-deftest ert-test-error-debug ()
      (let ((test (make-ert-test :body (lambda () (error "Error message")))))
        (condition-case condition
            (progn
              (let ((ert-debug-on-error t))
                (ert-run-test test))
              (cl-assert nil))
          ((error)
           (cl-assert (equal condition '(error "Error message")) t)))))

Until now, this test passes just like that, i.e. without entering
the debugger.  With the new code, this test does enter the debugger.

Can anyone give me a hand figuring out why/how the debugger is not entered
with the current code?  AFAICT when running the inner `ert-run-test`,
we get to `ert--run-test-internal` which sets `debugger` to a function
that calls `ert--run-test-debugger` and `debug-on-error` to t, yet when
it calls `(error "Error message")` this debugger-function is not called
(or at least `ert--run-test-debugger` is not called), which I can't explain.


        Stefan