bug#19054: 25.0.50; Wrong envelope-from in report-emacs-bug

[Peder O. Klingenberg]

(Creating a bug report from a mail thread on emacs-devel to keep it
alive.  I probably should have done this as a bug report originally.
Original thread on http://permalink.gmane.org/gmane.emacs.devel/175920)

I wrote:
> Earlier today I was bitten by something that has irritated be before as
> well.  When reporting bugs from an "emacs -Q", emacs doesn't know my
> email address.  No problem, I can edit the From: header.
>
> However, I like to use my local sendmail as transport, because I have
> already told that how and where to forward my mail, and it seems like a
> perfectly good option.  But message.el then decides to make up its own
> envelope-from, regardless of From: header.  That makes the mail bounce.
>
> Attached is a patch to emacsbug.el that checks if what message.el would
> do results in a bogus address, and if so, defaults to the From: header
> instead.

emacsbug.patch
Description: Text Data

Eli Zaretskii replied with a concern:
>> +;; It's the default mail mode, so it seems OK to use its features.
>> +(autoload 'message-bogus-recipient-p "message")
>> +(autoload 'message-make-address "message")
>> +(defvar message-send-mail-function)
>> +(defvar message-sendmail-envelope-from)
>
> What if the user overrides that default, and uses a different mail
> mode?

To which I replied:
> Then I guess that user will have some unused functions loaded, as well
> as some variables defined that she has no use for?  I was basically just
> cargo-culting the existing code.
>
> Existing code in emacsbug.el has 
>
>   ;; It's the default mail mode, so it seems OK to use its features.
>   (autoload 'message-bogus-recipient-p "message")
>   (defvar message-send-mail-function)
>
> I just added one more function and one more variable, and moved the
> whole bunch, including comment, earlier in the file, because I used
> those functions earlier in the file and thought it logical to place the
> definitions before the uses.
>
> Autoload of one function causes the entire file to be loaded, does it
> not?  So my additions are more for documentation purposes than any
> actual functional change.

Does anyone have any further thoughts?  I consider existing behaviour a
bug, and I believe my patch fixes it in a relatively unobtrusive manner.
message.el is already autoloaded today, and my patch does not change
behaviour if the user has already told message.el how to determine
envelope-from, or if the guessed address would be sane.