bug#36145: 26.2; Unable to "trigger another backtrace"

bug#36145: 26.2; Unable to "trigger another backtrace"

[Xu Chunyang]

(info "(elisp) Using Debugger") says

> When the debugger has been entered, the ‘debug-on-error’ variable is
> temporarily set according to ‘eval-expression-debug-on-error’.  If the
> latter variable is non-‘nil’, ‘debug-on-error’ will temporarily be set
> to ‘t’.  This means that any further errors that occur while doing a
> debugging session will (by default) trigger another backtrace.

Text before "This means.." is correct, but the conclusion seems
incorrect, to reproduce

- emacs -Q
- M-: (/ 1 0)     ;; enter the debugger
- M-: (/ 2 0)     ;; just message the error, no another backtrace

Am I misunderstanding "trigger another backtrace"?

bug#36145: 26.2; Unable to "trigger another backtrace"

[Noam Postavsky]

Xu Chunyang <mail@xuchunyang.me> writes:

> (info "(elisp) Using Debugger") says
>
>> When the debugger has been entered, the ‘debug-on-error’ variable is
>> temporarily set according to ‘eval-expression-debug-on-error’.  If the
>> latter variable is non-‘nil’, ‘debug-on-error’ will temporarily be set
>> to ‘t’.  This means that any further errors that occur while doing a
>> debugging session will (by default) trigger another backtrace.
>
> Text before "This means.." is correct, but the conclusion seems
> incorrect, to reproduce
>
> - emacs -Q
> - M-: (/ 1 0)     ;; enter the debugger
> - M-: (/ 2 0)     ;; just message the error, no another backtrace
>
> Am I misunderstanding "trigger another backtrace"?

I think this is a documentation bug.  You don't get another backtrace
because inhibit-debugger is bound to t before invoking the debugger.
This has been the case since Emacs 24.3 when inhibit-debugger was
introduced [1: 45b82ad0eb].

But even before that, debug-on-error was in fact bound to nil when
entering the debugger (and that has been the case since the initial
revision of debug.el), so the documentation was always wrong (although
it looks like you could trigger another backtrace with M-: because
eval-expression would rebind debug-on-error according to
eval-expression-debug-on-error, so the conclusion used to be correct,
even though the explanation was always wrong).

[1: 45b82ad0eb]: 2012-09-11 20:14:50 -0400
  * src/eval.c: Add `inhibit-debugger'.
  https://git.savannah.gnu.org/cgit/emacs.git/commit/?id=45b82ad0ebedaa1b7094912e218bea1510c33feb

bug#36145: 26.2; Unable to "trigger another backtrace"

[Noam Postavsky]

Noam Postavsky <npostavs@gmail.com> writes:

> Xu Chunyang <mail@xuchunyang.me> writes:
>
>> (info "(elisp) Using Debugger") says
>>
>>> When the debugger has been entered, the ‘debug-on-error’ variable is
>>> temporarily set according to ‘eval-expression-debug-on-error’.  If the
>>> latter variable is non-‘nil’, ‘debug-on-error’ will temporarily be set
>>> to ‘t’.  This means that any further errors that occur while doing a
>>> debugging session will (by default) trigger another backtrace.
>>
>> Text before "This means.." is correct, but the conclusion seems
>> incorrect, to reproduce
>>
>> - emacs -Q
>> - M-: (/ 1 0)     ;; enter the debugger
>> - M-: (/ 2 0)     ;; just message the error, no another backtrace
>>
>> Am I misunderstanding "trigger another backtrace"?
>
> I think this is a documentation bug.  You don't get another backtrace
> because inhibit-debugger is bound to t before invoking the debugger.

Although it is possible to trigger another backtrace if the first one
was from calling `debug' directly, e.g., M-: (debug) RET.

bug#36145: 26.2; Unable to "trigger another backtrace"

[Noam Postavsky]

[-- Attachment #1: Type: text/plain, Size: 491 bytes --]

severity 36145 minor
tags 36145 + patch
quit

> Although it is possible to trigger another backtrace if the first one
> was from calling `debug' directly, e.g., M-: (debug) RET.

And it's also possible to trigger another backtrace with 'e', if using
it from a frame above the debugger's.  I bump into this every so often
by accident, and it's quite annoying.  Of course, I'm always in the
middle of debugging something else so I forget about it, but I'd like to
change it.  Patch attached.


[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: patch --]
[-- Type: text/x-diff, Size: 8408 bytes --]

From c6d52846598eeaf200c553ff0ac9f16db612b64e Mon Sep 17 00:00:00 2001
From: Noam Postavsky <npostavs@gmail.com>
Date: Fri, 21 Jun 2019 08:14:20 -0400
Subject: [PATCH] Suppress recursive debugger for 'e' command (Bug#36145)

* lisp/emacs-lisp/debug.el (debugger-eval-expression):
* lisp/emacs-lisp/edebug.el (edebug-eval-expression)
(edebug-eval-last-sexp, edebug-eval-print-last-sexp): Suppress
debugger with inhibit-debugger, unless given a prefix argument.
* doc/lispref/edebug.texi (Edebug Eval):
* doc/lispref/debugging.texi (Debugger Commands, Backtraces): Document
new behavior.
(Using Debugger): Remove incorrect paragraph about debug-on-error.
* etc/NEWS: Announce new behavior.
---
 doc/lispref/debugging.texi | 15 ++++-----------
 doc/lispref/edebug.texi    | 10 +++++++---
 etc/NEWS                   | 11 +++++++++++
 lisp/emacs-lisp/debug.el   | 12 ++++++++----
 lisp/emacs-lisp/edebug.el  | 33 +++++++++++++++++++++------------
 5 files changed, 51 insertions(+), 30 deletions(-)

diff --git a/doc/lispref/debugging.texi b/doc/lispref/debugging.texi
index 9e43343310..7538e70ab5 100644
--- a/doc/lispref/debugging.texi
+++ b/doc/lispref/debugging.texi
@@ -383,16 +383,6 @@ Using Debugger
 rather than bury it.  Consult the variable's documentation for more
 possibilities.)
 
-  When the debugger has been entered, the @code{debug-on-error}
-variable is temporarily set according to
-@code{eval-expression-debug-on-error}.  If the latter variable is
-non-@code{nil}, @code{debug-on-error} will temporarily be set to
-@code{t}.  This means that any further errors that occur while doing a
-debugging session will (by default) trigger another backtrace.  If
-this is not what you want, you can either set
-@code{eval-expression-debug-on-error} to @code{nil}, or set
-@code{debug-on-error} to @code{nil} in @code{debugger-mode-hook}.
-
   The debugger itself must be run byte-compiled, since it makes
 assumptions about the state of the Lisp interpreter.  These
 assumptions are false if the debugger is running interpreted.
@@ -527,7 +517,10 @@ Debugger Commands
 temporarily restores their values from outside the debugger, so you can
 examine and change them.  This makes the debugger more transparent.  By
 contrast, @kbd{M-:} does nothing special in the debugger; it shows you
-the variable values within the debugger.
+the variable values within the debugger.  By default, this command
+suppresses the debugger during evaluation, so that an error in the
+evaluated expression won't add a new error on top of the existing one.
+Use the prefix argument to override this.
 
 @item R
 Like @kbd{e}, but also save the result of evaluation in the
diff --git a/doc/lispref/edebug.texi b/doc/lispref/edebug.texi
index 2c0ee3969b..823a5009e8 100644
--- a/doc/lispref/edebug.texi
+++ b/doc/lispref/edebug.texi
@@ -681,8 +681,11 @@ Edebug Eval
 @table @kbd
 @item e @var{exp} @key{RET}
 Evaluate expression @var{exp} in the context outside of Edebug
-(@code{edebug-eval-expression}).  That is, Edebug tries to minimize its
-interference with the evaluation.
+(@code{edebug-eval-expression}).  That is, Edebug tries to minimize
+its interference with the evaluation.  By default, this command
+suppresses the debugger during evaluation, so that an error in the
+evaluated expression won't add a new error on top of the existing one.
+Use a prefix argument to override this.
 
 @item M-: @var{exp} @key{RET}
 Evaluate expression @var{exp} in the context of Edebug itself
@@ -690,7 +693,8 @@ Edebug Eval
 
 @item C-x C-e
 Evaluate the expression before point, in the context outside of Edebug
-(@code{edebug-eval-last-sexp}).
+(@code{edebug-eval-last-sexp}).  Similar to @key{e}, this command
+suppresses the debugger unless a prefix argument is used.
 @end table
 
 @cindex lexical binding (Edebug)
diff --git a/etc/NEWS b/etc/NEWS
index 4a168d5678..43dc6ea969 100644
--- a/etc/NEWS
+++ b/etc/NEWS
@@ -1074,6 +1074,11 @@ Backtrace mode adds fontification and commands for changing the
 appearance of backtrace frames.  See the node "(elisp) Backtraces" in
 the Elisp manual for documentation of the new mode and its commands.
 
++++
+*** Recursive debugging is suppressed by default for 'e'.
+That is bound to the command 'debug-eval-expression'.  Use a prefix
+argument to override this.
+
 ** Edebug
 
 +++
@@ -1098,6 +1103,12 @@ The new 'backtrace-goto-source' command, bound to 's', works in
 Edebug's backtraces on backtrace frames whose source code has
 been instrumented by Edebug.
 
++++
+*** Backtrace debugging is suppressed by default for 'e' and 'C-x C-e'.
+Those are bound to the commands 'edebug-eval-expression' and
+'edebug-eval-last-sexp', respectively.  Use a prefix argument to
+override this.
+
 ** Enhanced xterm support
 
 *** New variable 'xterm-set-window-title' controls whether Emacs sets
diff --git a/lisp/emacs-lisp/debug.el b/lisp/emacs-lisp/debug.el
index 8989aa0719..20aceccb68 100644
--- a/lisp/emacs-lisp/debug.el
+++ b/lisp/emacs-lisp/debug.el
@@ -516,17 +516,21 @@ (defun debugger--backtrace-base ()
 	 'debug--implement-debug-on-entry)
 	(t 'debug)))
 
-(defun debugger-eval-expression (exp &optional nframe)
+(defun debugger-eval-expression (exp &optional nframe allow-recursive-debug)
   "Eval an expression, in an environment like that outside the debugger.
-The environment used is the one when entering the activation frame at point."
+The environment used is the one when entering the activation frame at point.
+With a prefix argument, errors signaled while evaluating may
+trigger the debugger again."
   (interactive
-   (list (read--expression "Eval in stack frame: ")))
+   (list (read--expression "Eval in stack frame: ")
+         nil current-prefix-arg))
   (let ((nframe (or nframe
                     (condition-case nil (1+ (debugger-frame-number 'skip-base))
                       (error 0)))) ;; If on first line.
 	(base (debugger--backtrace-base)))
     (debugger-env-macro
-      (let ((val (backtrace-eval exp nframe base)))
+      (let ((val (let ((inhibit-debugger (not allow-recursive-debug)))
+                   (backtrace-eval exp nframe base))))
         (prog1
             (debugger--print val t)
           (let ((str (eval-expression-print-format val)))
diff --git a/lisp/emacs-lisp/edebug.el b/lisp/emacs-lisp/edebug.el
index ab5553b4e8..1f9d994fed 100644
--- a/lisp/emacs-lisp/edebug.el
+++ b/lisp/emacs-lisp/edebug.el
@@ -3580,28 +3580,37 @@ (defalias 'edebug-prin1-to-string #'cl-prin1-to-string)
 (defalias 'edebug-format #'format-message)
 (defalias 'edebug-message #'message)
 
-(defun edebug-eval-expression (expr)
+(defun edebug-eval-expression (expr &optional allow-debugger)
   "Evaluate an expression in the outside environment.
 If interactive, prompt for the expression.
-Print result in minibuffer."
-  (interactive (list (read--expression "Eval: ")))
+Print result in minibuffer.
+With a prefix argument, errors signaled while evaluating may
+trigger the debugger."
+  (interactive (list (read--expression "Eval: ")
+                     current-prefix-arg))
   (princ
    (edebug-outside-excursion
-    (setq values (cons (edebug-eval expr) values))
+    (push (let ((inhibit-debugger (not allow-debugger)))
+            (edebug-eval expr))
+          values)
     (concat (edebug-safe-prin1-to-string (car values))
             (eval-expression-print-format (car values))))))
 
-(defun edebug-eval-last-sexp ()
+(defun edebug-eval-last-sexp (&optional allow-debugger)
   "Evaluate sexp before point in the outside environment.
-Print value in minibuffer."
-  (interactive)
-  (edebug-eval-expression (edebug-last-sexp)))
+Print value in minibuffer.
+With a prefix argument, errors signaled while evaluating may
+trigger the debugger."
+  (interactive "P")
+  (edebug-eval-expression (edebug-last-sexp allow-debugger)))
 
-(defun edebug-eval-print-last-sexp ()
+(defun edebug-eval-print-last-sexp (&optional allow-debugger)
   "Evaluate sexp before point in outside environment; insert value.
-This prints the value into current buffer."
-  (interactive)
-  (let* ((form (edebug-last-sexp))
+This prints the value into current buffer.
+With a prefix argument, errors signaled while evaluating may
+trigger the debugger."
+  (interactive "P")
+  (let* ((form (edebug-last-sexp allow-debugger))
 	 (result-string
 	  (edebug-outside-excursion
 	   (edebug-safe-prin1-to-string (edebug-safe-eval form))))
-- 
2.11.0

bug#36145: 26.2; Unable to "trigger another backtrace"

[Noam Postavsky]

tags 36145 - patch
quit

> And it's also possible to trigger another backtrace with 'e', if using
> it from a frame above the debugger's.  I bump into this every so often
> by accident, and it's quite annoying.  Of course, I'm always in the
> middle of debugging something else so I forget about it, but I'd like to
> change it.  Patch attached.
>
> Subject: [PATCH] Suppress recursive debugger for 'e' command (Bug#36145)

Uh, in my haste to put something down before I forgot about it again, I
neglected to test this properly; it doesn't actually work.  So I guess
consider just the doc part of the patch as the proposal, something
working to follow later.

bug#36145: 26.2; Unable to "trigger another backtrace"

[Lars Ingebrigtsen]

Noam Postavsky <npostavs@gmail.com> writes:

>> And it's also possible to trigger another backtrace with 'e', if using
>> it from a frame above the debugger's.  I bump into this every so often
>> by accident, and it's quite annoying.  Of course, I'm always in the
>> middle of debugging something else so I forget about it, but I'd like to
>> change it.  Patch attached.

Yes, that's super duper annoying, and it'd be really great to get that
fixed.

>> Subject: [PATCH] Suppress recursive debugger for 'e' command (Bug#36145)
>
> Uh, in my haste to put something down before I forgot about it again, I
> neglected to test this properly; it doesn't actually work.  So I guess
> consider just the doc part of the patch as the proposal, something
> working to follow later.

This was two years ago -- did you get any further with this?

-- 
(domestic pets only, the antidote for overdose, milk.)
   bloggy blog: http://lars.ingebrigtsen.no

bug#36145: 26.2; Unable to "trigger another backtrace"

[Lars Ingebrigtsen]

Lars Ingebrigtsen <larsi@gnus.org> writes:

>>> And it's also possible to trigger another backtrace with 'e', if using
>>> it from a frame above the debugger's.  I bump into this every so often
>>> by accident, and it's quite annoying.  Of course, I'm always in the
>>> middle of debugging something else so I forget about it, but I'd like to
>>> change it.  Patch attached.
>
> Yes, that's super duper annoying, and it'd be really great to get that
> fixed.

I respun the patch, and observed that it did, indeed, not work, so I've
reworked it (instead of inhibiting the debugger, I just use
condition-case in these commands), and it now seems to work fine.  Oh,
and instead of having a prefix, I added a user option (both because I
think some people want to have the old action always, and because the
prefix was already taken in edebug now).

-- 
(domestic pets only, the antidote for overdose, milk.)
   bloggy blog: http://lars.ingebrigtsen.no

bug#36145: 26.2; Unable to "trigger another backtrace"

[Lars Ingebrigtsen]

Xu Chunyang <mail@xuchunyang.me> writes:

>> temporarily set according to ‘eval-expression-debug-on-error’.  If the
>> latter variable is non-‘nil’, ‘debug-on-error’ will temporarily be set
>> to ‘t’.  This means that any further errors that occur while doing a
>> debugging session will (by default) trigger another backtrace.
>
> Text before "This means.." is correct, but the conclusion seems
> incorrect, to reproduce
>
> - emacs -Q
> - M-: (/ 1 0)     ;; enter the debugger
> - M-: (/ 2 0)     ;; just message the error, no another backtrace

I've now fixed the documentation here in Emacs 29.

-- 
(domestic pets only, the antidote for overdose, milk.)
   bloggy blog: http://lars.ingebrigtsen.no