bug#42383: 28.0.50; Two bugs with M-x compile

[Eli Zaretskii <eliz@gnu.org>]

> Date: Thu, 16 Jul 2020 09:48:56 +0000
> From: Gregory Heytings via "Bug reports for GNU Emacs,
>  the Swiss army knife of text editors" <bug-gnu-emacs@gnu.org>
> 
> 
> > 1. create a Makefile with a few targets
> > 2. start emacs -Q
> > 3. M-x compile
> > 4. press TAB: the list of targets is displayed
> > 5. type the first character of an existing target, and press TAB
> > 6. the result is a "[No match]"
> >
> > This has been working correctly for years (tested with Emacs 24, 25, 
> > 26), and does not work anymore with Emacs 28.
> >
> > It seems that at step 5 above the list of completion candidates that is 
> > considered are subdirectories.
> >
> 
> Upon further investigation, this bug has been introduced by the patch for 
> bug#34330, which moved pcomplete-completions-at-point last in 
> shell-dynamic-complete-functions in shell.el:
> 
> (defcustom shell-dynamic-complete-functions
>     '(comint-c-a-p-replace-by-expanded-history
>       shell-environment-variable-completion
>       shell-command-completion
>       shell-c-a-p-replace-by-expanded-directory
>       shell-filename-completion
>       comint-filename-completion
>       ;; Put `pcomplete-completions-at-point' last so that other
>       ;; functions can run before it does, see bug#34330.
>       pcomplete-completions-at-point)
> 
> See https://git.savannah.gnu.org/cgit/emacs.git/commit/?id=e4d17d8cb479ffeeb7dfb7320a1432722ac8df75 .
> 
> The obvious fix is to revert the situation (that is, to move 
> pcomplete-completions-at-point again before shell-filename-completion), 
> but I'm not sure, because bug#34330 would again exist.  That being said, 
> I'm not sure that handling the exceptional case of filenames with '&' 
> characters correctly when completing in M-x shell (which is what bug#34330 
> is about) justifies to break normal behavior in other situations.

Indeed.

Stefan, Alex: any ideas for how to fix this?  This problem exists on
the emacs-27 branch, so we must find a safe solution for that branch,
and we must do that quickly.  If we don't have any better ideas, I
think we should revert that fix on emacs-27 at least.

Thanks.