bug#33654: 27.0.50; No record of init file errors printed in *Messages*

[N. Jackson]

Hello Eli,

At 09:13 +0200 on Friday 2018-12-07, Eli Zaretskii wrote:

>> From: "N. Jackson" <nljlistbox2@gmail.com>
>> Date: Thu, 06 Dec 2018 19:42:05 -0500
>> 
>> When there is an error in the init file, so that Emacs fails to
>> start normally, the user is not notified of the problem in the
>> *Messages* buffer.
>
> Generally, that happens when we cannot safely display anything,
> nor even insert text into *Messages*, if problems occur early
> enough into the startup process, because the necessary
> infrastructure is not yet available.  But we do try our best, see
> below.  So your report, which sounds as if there's a general
> non-display of any error or warning messages in this case, is
> inaccurate.

Yes, you are quite right. Sorry for the rather sweeping statement, I
was trying to characterise the problem I saw. I've typically been
happy with the information I've had from Emacs when I've had errors
in my init files, and I think that was why I was surprised when I
was (in my opinion) inadequately informed in this case.

I though something might have changed, but perhaps the specific
circumstances were different from the errors I've had in the past.

> Please be more specific regarding the particular problems you saw
> that didn't get logged.

Please see below.

> Such errors are generally displayed right away, after interrupting
> the initialization process.  So I wonder what specific example do
> you have in mind.

Please see below.

> Are you talking about errors or warnings?  And did you include the
> *Warnings* buffer in your review?  That's where we put delayed
> warnings from the initialization process.  More generally, did you
> read up on the "delayed warnings" machinery?

No, I've not read that documentation recently -- I'll add it to my
reading list. Yes there was a *Warnings* buffer, but it was in a
window on a frame I didn't visit. I did not know that I needed to
check both *Messages* and *Warnings*.

I think, perhaps, that if something were written to *Messages* to inform
the user that there was a warning/error, and to direct them to the
*Warnings* buffer for details, it would make it less likely that the
user would miss the information altogether. Something like

  An initialization error occurred -- see *Warnings* buffer for details.

Also, the frame that was split to display the *Warnings* buffer was not
the one that had the focus when Emacs finished starting up. (Indeed, I
think it might have been on the most deeply buried frame.) There is a
very high probability I would never see a warning/error displayed so
obscurely.

It would seem that it would be better, if it were feasible, if the
*Warnings* buffer could be displayed in a window on the frame that
has the focus after Emacs finishes starting up.


DETAILS

Here is a detailed example of the problem. I think this might have
really happened to me in real life (resulting in Bug##33536 [1]).

Scenario: I have a notion that it would be a good idea to remove and
then re-install an Emacs package. I decide that to be sure that none of
it still hanging around in memory, I should exit an restart Emacs
between removing the package and reinstalling it. The package in this
case happens to be bbdb from Gnu Elpa. I have

  (bbdb-initialize 'gnus 'message)

in my init file.

When Emacs starts after bbdb has been uninstalled, the above line
results in an error. A *Warnings* buffer is displayed as follows:

  Warning (initialization): An error occurred while loading `/home/nlj/.emacs':

  Symbol's function definition is void: bbdb-initialize

  To ensure normal operation, you should investigate and remove the
  cause of the error in your initialization file.  Start Emacs with
  the `--debug-init' option to view a complete error backtrace.

But I don't see this error/warning message because I don't visit the
frame on which it is displayed. After all, it's just going to be a short
Emacs session during which all I'm going to do is install bbdb.

In my *Messages* buffer is displayed:

  Loading cua-base...done
  Loading delsel...done
  Loading desktop...done
  Loading battery...done
  Loading time...done
  Loading elec-pair...done
  Loading saveplace...done
  Loading savehist...done
  Loading paren...done
  Loading /home/nlj/.recentf...done
  Cleaning up the recentf list...done (0 removed)
  Starting new Ispell process /usr/bin/aspell with default dictionary...
  Loading dired-x...done
  Setting up indent for shell type sh
  Indentation variables are now local.
  Indentation setup for shell type sh
  Ispell process killed
  Starting new Ispell process /usr/bin/aspell with default dictionary...
  Setting up indent for shell type sh
  Indentation variables are now local.
  Indentation setup for shell type sh
  Omitting...
  Omitted 384 lines.
  uncompressing de.map.gz...done
  Wrote /data/projects/vc/emacs/git/emacs/.emacs.desktop.lock
  Desktop: 7 frames, 54 buffers restored.
  For information about GNU Emacs and the GNU system, type C-h C-a.

This shows no indication of the initialization error.


Now it might be a user error to look for an indication of start up
problems only in *Messages*, and it might a user error to miss the
*Warnings* buffer, but I think it would improve Emacs, were it feasible,
if it helped the user to avoid these mistakes by mentioning the
error/warning in *Messages*, and by displaying the *Warnings* buffer in
a frame that isn't buried.


[1] https://debbugs.gnu.org/cgi/bugreport.cgi?bug=33536