bug#63988: 30.0.50; Recent header line format changes cause spin/seg fault with format-mode-line

[Eshel Yaron via "Bug reports for GNU Emacs, the Swiss army knife of text editors" <bug-gnu-emacs@gnu.org>]

Eli Zaretskii <eliz@gnu.org> writes:

>> Cc: 63988@debbugs.gnu.org
>> Date: Sat, 10 Jun 2023 09:47:29 +0300
>> From: Eli Zaretskii <eliz@gnu.org>
>> 
>> > From: Aaron Jensen <aaronjensen@gmail.com>
>> > Date: Fri, 09 Jun 2023 21:09:45 -0400
>> > 
>> > 
>> > (setq header-line-format '(:eval (format-mode-line "")))
>> > 
>> > This causes Emacs to spin as of commit:
>> > 4f66cbbfe520ee31ef26676e09a926217d9736fe
>> > 
>> > After some time, it will segfault.
>> 
>> It causes infinite recursion, since format-mode-line also calls
>> window_wants_header_line (indirectly).
>> 
>> But what is the purpose of such a strange (to use a civilized word)
>> setting of header-line-format?  Why do you need :eval at all in this
>> case?
>> 
>> IOW, why not say "don't do that" and be done?
>
> For now, I installed a semi-kludgey fix.
>
> However, I wonder whether we should rethink this minor feature.
> Perhaps this minor convenience is not worth the complications?
> window_wants_header_line is called in many places, all of which can
> now evaluate arbitrary Lisp, and all of which can now GC.  I've
> audited the various callers, and didn't see anything obvious that
> could cause problems with calling Lisp or GC in those places, but I
> could have missed something.
>
> This is actually a general issue with Emacs: we keep piling one minor
> feature upon another, and don't always reflect on the hard-to-maintain
> monster that creates.  We have already a couple of areas in the code
> base where we are afraid of making changes, because we don't have a
> good understanding of the complicated state variables there.  Maybe we
> should reject such minor features instead of keeping doing what we
> have been doing?
>
> So maybe we should declare this feature a failed experiment and remove
> it?

FWIW I think that make sense, but IMO it'd be best to only remove the
treatment of `:eval` in `window_wants_header_line`, and keep the new
treatment of `header-line-format` being a cons cell with a void or
nil-valued variable car.  That's still useful because it works well with
minor mode variables, and it's less risky as it doesn't involve
evaluating arbitrary lisp, just inspecting a variable.