Maybe the debugger should not ignore read errors

Maybe the debugger should not ignore read errors

[Spencer Baugh]

If you read a file with a syntax error with debug-on-error=t, e.g.:

(let ((debug-on-error t)) (read ""))

Emacs will not drop to the debugger.  Instead, the error "End of file
during parsing" will just be printed.  This is because
debug-ignored-errors includes end-of-file, which is what read signals on
error.

This in turn means that --debug-init will not cause Emacs to drop to the
debugger if there's a read error in some file at load time, which can
happen in many ways.  For example, some state file such as created by
project.el or savehist.el might get corrupted one way or another.

This seems unnecessarily hostile to novice Emacs users who only know
that --debug-init seems to do nothing when faced with this kind of
error.  Certainly they have no idea that debug-ignored-errors even
exists.

Should we remove end-of-file from debug-ignored-errors?  end-of-file is
not like the other errors in debug-ignored-errors, which are all
user-errors.

That would make --debug-init and M-x toggle-debug-on-error drop to the
debugger when a read error is encountered, which seems like an
improvement.

Re: Maybe the debugger should not ignore read errors

[Eli Zaretskii]

> From: Spencer Baugh <sbaugh@janestreet.com>
> Date: Wed, 17 Jan 2024 13:51:03 -0500
> 
> 
> If you read a file with a syntax error with debug-on-error=t, e.g.:
> 
> (let ((debug-on-error t)) (read ""))
> 
> Emacs will not drop to the debugger.  Instead, the error "End of file
> during parsing" will just be printed.  This is because
> debug-ignored-errors includes end-of-file, which is what read signals on
> error.
> 
> This in turn means that --debug-init will not cause Emacs to drop to the
> debugger if there's a read error in some file at load time, which can
> happen in many ways.  For example, some state file such as created by
> project.el or savehist.el might get corrupted one way or another.
> 
> This seems unnecessarily hostile to novice Emacs users who only know
> that --debug-init seems to do nothing when faced with this kind of
> error.  Certainly they have no idea that debug-ignored-errors even
> exists.

No, when Emacs is invoked with --debug-init, we reset
debug-ignored-errors to nil while the init file is loaded.  This was
done as part of fixing bug#64163, so if you see this now on master, it
is some kind of regression, perhaps due to recent changes related to
handler-bind.

> Should we remove end-of-file from debug-ignored-errors?

No, I don't think so.  The problem with debugging the init files is
supposed to be solved (if that broke, we need to fix it), and other
than that I don't see a reason to change this age-old behavior.
Please keep in mind that end-of-file frequently happens when users
type expressions at the M-: prompt and make syntax mistakes; we don't
want to pop up the debugger in that case (and other similar ones).

> end-of-file is not like the other errors in debug-ignored-errors,
> which are all user-errors.

Actually, end-of-file is frequently a cockpit error, see above.

Re: Maybe the debugger should not ignore read errors

[Spencer Baugh]

Eli Zaretskii <eliz@gnu.org> writes:
>> From: Spencer Baugh <sbaugh@janestreet.com>
>> Date: Wed, 17 Jan 2024 13:51:03 -0500
>> 
>> 
>> If you read a file with a syntax error with debug-on-error=t, e.g.:
>> 
>> (let ((debug-on-error t)) (read ""))
>> 
>> Emacs will not drop to the debugger.  Instead, the error "End of file
>> during parsing" will just be printed.  This is because
>> debug-ignored-errors includes end-of-file, which is what read signals on
>> error.
>> 
>> This in turn means that --debug-init will not cause Emacs to drop to the
>> debugger if there's a read error in some file at load time, which can
>> happen in many ways.  For example, some state file such as created by
>> project.el or savehist.el might get corrupted one way or another.
>> 
>> This seems unnecessarily hostile to novice Emacs users who only know
>> that --debug-init seems to do nothing when faced with this kind of
>> error.  Certainly they have no idea that debug-ignored-errors even
>> exists.
>
> No, when Emacs is invoked with --debug-init, we reset
> debug-ignored-errors to nil while the init file is loaded.  This was
> done as part of fixing bug#64163, so if you see this now on master, it
> is some kind of regression, perhaps due to recent changes related to
> handler-bind.

Oh, that's great!  I didn't test the --debug-init part with master, so I
missed that this had changed.  That mostly resolves my concern, then.

>> Should we remove end-of-file from debug-ignored-errors?
>
> No, I don't think so.  The problem with debugging the init files is
> supposed to be solved (if that broke, we need to fix it), and other
> than that I don't see a reason to change this age-old behavior.
> Please keep in mind that end-of-file frequently happens when users
> type expressions at the M-: prompt and make syntax mistakes; we don't
> want to pop up the debugger in that case (and other similar ones).
>
>> end-of-file is not like the other errors in debug-ignored-errors,
>> which are all user-errors.
>
> Actually, end-of-file is frequently a cockpit error, see above.

That's a good point, it would be annoying to drop to the debugger in
that case.

(So then I'll just fix the related bug#68546 to improve the error
message, and I'll be happy)