eval-buffer with debug-on-error t

eval-buffer with debug-on-error t

[Lennart Borgman]

I have debug-on-error set to t. When I do an eval-buffer and there is an 
undefined function:

   (this-is-no-fun)

I get the error "Wrong type argument: stringp, nul

If I instead let the region be the whole buffer and to eval-region I get 
a traceback with "Symbol's function definition is void: this-is-no-fun".

Is not the later much more useful? Can't eval-buffer return something 
similar?

When debug-on-error is nil eval-buffer returns the same as eval-region.

Re: eval-buffer with debug-on-error t

[Richard M. Stallman]

    I have debug-on-error set to t. When I do an eval-buffer and there is an 
    undefined function:

       (this-is-no-fun)

    I get the error "Wrong type argument: stringp, nul

Does this fix it?  (I installed a more elaborate patch.)

*** debug.el	05 Jul 2005 00:45:47 -0400	1.90
--- debug.el	09 Jul 2005 00:05:50 -0400	
***************
*** 315,322 ****
      (while (re-search-forward "^  eval-buffer(" nil t)
        (end-of-line)
        (insert (format "\n  ;;; Reading at buffer position %d"
! 		      (with-current-buffer (nth 2 (backtrace-frame (debugger-frame-number)))
! 			(point))))))
    (debugger-make-xrefs))
  
  (defun debugger-make-xrefs (&optional buffer)
--- 315,325 ----
      (while (re-search-forward "^  eval-buffer(" nil t)
        (end-of-line)
        (insert (format "\n  ;;; Reading at buffer position %d"
! 		      (with-current-buffer
! 			  (or (nth 2 (backtrace-frame (debugger-frame-number)))
! 			      debugger-old-buffer)
! 			(point))))
!       ))
    (debugger-make-xrefs))
  
  (defun debugger-make-xrefs (&optional buffer)

Re: eval-buffer with debug-on-error t

[Johan Bockgård]

"Richard M. Stallman" <rms@gnu.org> writes:

> Does this fix it?  (I installed a more elaborate patch.)

Well, it does make the error go away.

A problem is that `debugger-old-buffer' is not necessarily the same
buffer that was current when `eval-buffer' was called.


Example (with the patch):

Run emacs -Q

M-x toggle-debug-on-error RET

Insert this into *scratch*:

     (get-buffer-create "foo")
     (with-current-buffer "foo"
       (error "foo"))

M-x eval-buffer RET

=>

Debugger entered--Lisp error: (error "foo")
  signal(error ("foo"))
  error("foo")
  (save-current-buffer (set-buffer "foo") (error "foo"))
  (with-current-buffer "foo" (error "foo"))
  eval-buffer()  ; Reading at buffer position 1
                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
This is the position in the "foo" buffer.




`eval-region' has the same problem. In fact, regardless of that
problem the feature does not seem very useful at present when it comes
to eval-region.

E.g. if you do C-x h M-x eval-region RET and there is an error you
will get a note like this in the backtrace no matter where the error
occurred:

  eval-region(1 500)  ; Reading at buffer position 1

Since point doesn't move (eval-region apparently works differently
than eval-buffer in this respect) the value of point will always be at
the start or the end of the region (and these values are already
present in the backtrace).

-- 
Johan Bockgård

Re: eval-buffer with debug-on-error t

[Richard M. Stallman]

    A problem is that `debugger-old-buffer' is not necessarily the same
    buffer that was current when `eval-buffer' was called.

Strange, somehow I thought about that issue and concluded that it
would be the current buffer.  Now it's clear I was not thinking
straight at the time.

I fixed it another way.

Re: eval-buffer with debug-on-error t

[Lennart Borgman]

Richard M. Stallman wrote:

>    I have debug-on-error set to t. When I do an eval-buffer and there is an 
>    undefined function:
>
>       (this-is-no-fun)
>
>    I get the error "Wrong type argument: stringp, nul
>
>Does this fix it?  (I installed a more elaborate patch.)
>  
>
Yes, thanks, I think so. I tested with the latest CVS and emacs -Q.