bug#35230: char-displayable-p return code is conflictingly documented

[Robert Pluim <rpluim@gmail.com>]

>>>>> On Fri, 12 Apr 2019 15:51:48 +0300, Eli Zaretskii <eliz@gnu.org> said:

    >> From: Robert Pluim <rpluim@gmail.com> Date: Thu, 11 Apr 2019
    >> 16:49:29 +0200
    >> 
    >> (This comes as a result of the stackexchange question at
    >> https://emacs.stackexchange.com/questions/48810)
    >> 
    >> char-displayable-p docstring says
    >> 
    >> Return non-nil if we should be able to display CHAR.
    >> 
    >> The elisp manual says something subtly different:
    >> 
    >> This function returns ‘t’ if Emacs ought to be able to display
    >> CHAR.  More precisely, if the selected frame’s fontset has a
    >> font to display the character set that CHAR belongs to.
    >> 
    >> The function itself is more in line with the docstring:

    Eli> Yes, the manual is wrong and should be fixed.  Non-nil is
    Eli> exactly right, and callers should not depend on any finer
    Eli> definition of the return value, as it could be many different
    Eli> non-nil objects.

That I can do easily enough in emacs-26

    >> (char-displayable-p #xE01EF) => unicode
    >> 
    >> since I donʼt have a font with a glyph for that character, so
    >> it ends up displayed as a box with the unicode code point
    >> inside it. The code that results in 'unicode has the comment
    >> 
    >> ;; On a text terminal without glyph codes, CHAR is displayable
    >> ;; if the coding system for the terminal can encode it.
    >> 
    >> but Iʼm very much on a graphical terminal here, not a text
    >> terminal.

    Eli> This is a (known) deficiency in char-displayable-p, but one
    Eli> which is not easy to fix: no one says that every call to this
    Eli> function asks about the selected frame, so we cannot
    Eli> unconditionally disable the TTY branch when on GUI frames.
    Eli> Perhaps an optional argument could be added for that purpose,
    Eli> which is the frame for which to make the test.  At least
    Eli> some, if not most, of the calls will still need to omit that
    Eli> argument, though, because we currently need to know that up
    Eli> front to set up the quote-style, for example.

    Eli> Perhaps mentioning this caveat in the manual would be good.

Iʼve made an attempt below. Do we want a cross reference to 'Glyphless
Chars' as well?

diff --git a/doc/lispref/display.texi b/doc/lispref/display.texi
index b07999432c..fac883ae16 100644
--- a/doc/lispref/display.texi
+++ b/doc/lispref/display.texi
@@ -3507,12 +3507,16 @@ Fontsets
 @end defun
 
 @defun char-displayable-p char
-This function returns @code{t} if Emacs ought to be able to display
-@var{char}.  More precisely, if the selected frame's fontset has a
-font to display the character set that @var{char} belongs to.
+This function returns non-@code{nil} if Emacs ought to be able to
+display @var{char}.  More precisely, if the selected frame's fontset
+has a font to display the character set that @var{char} belongs to.
 
 Fontsets can specify a font on a per-character basis; when the fontset
 does that, this function's value may not be accurate.
+
+This function may return non-@code{nil} even when there is no font
+available, since it also checks if the coding system for the text
+terminal can encode the character.
 @end defun
 
 @node Low-Level Font