bug#70000: 29.2; Grapheme handling incorrect

bug#70000: 29.2; Grapheme handling incorrect

[Phillip Susi]

I had some terminal breakage the other day when browsing email with
notmuch.  Now a ways down the rabbit hole, it seems this is because
emacs does not correctly handle graphemes.  I found this article here:

https://mitchellh.com/writing/grapheme-clusters-in-terminals

If I paste that gramehe into GUI emacs, it is displayed as two separate
characters, each two columns wide, instead of the correct way: as a
single double wide character.  C-f and C-b move over the character as if
it were one, however, backspace deletes only the second, leaving both
the first and the zero width joiner.  If C-f and C-b treat it as one,
then so should backspace.

Under recent versions of the foot terminal emulator, this character is
displayed as a single, double wide character, but emacs assumes it still
is 4 colums wide, leading to terminal breakage.  Emacs needs to not
assume the width of graphemes are what wcwidth() reports, but instead
need to query the cursor position after printing one to find out how
wide the terminal actually dispalyed it as.



In GNU Emacs 29.2 (build 1, x86_64-pc-linux-gnu, GTK+ Version 3.24.39,
 cairo version 1.18.0) of 2024-02-26 built on localhost
System Description: Gentoo Linux

Configured using:
 'configure --prefix=/usr --build=x86_64-pc-linux-gnu
 --host=x86_64-pc-linux-gnu --mandir=/usr/share/man
 --infodir=/usr/share/info --datadir=/usr/share --sysconfdir=/etc
 --localstatedir=/var/lib --datarootdir=/usr/share
 --disable-silent-rules --docdir=/usr/share/doc/emacs-29.2-r1
 --htmldir=/usr/share/doc/emacs-29.2-r1/html --libdir=/usr/lib64
 --program-suffix=-emacs-29 --includedir=/usr/include/emacs-29
 --infodir=/usr/share/info/emacs-29 --localstatedir=/var
 --enable-locallisppath=/etc/emacs:/usr/share/emacs/site-lisp
 --without-compress-install --without-hesiod --without-pop
 --with-file-notification=inotify --with-pdumper --enable-acl
 --with-dbus --with-modules --without-gameuser --with-libgmp --with-gpm
 --with-native-compilation=aot --without-json --without-kerberos
 --without-kerberos5 --with-lcms2 --without-xml2 --without-mailutils
 --without-selinux --without-sqlite3 --with-gnutls --with-libsystemd
 --with-threads --with-tree-sitter --without-wide-int --with-sound=alsa
 --with-zlib --with-pgtk --without-x --without-ns
 --with-toolkit-scroll-bars --without-gconf --without-gsettings
 --without-harfbuzz --without-libotf --without-m17n-flt
 --without-xwidgets --with-gif --with-jpeg --with-png --with-rsvg
 --with-tiff --without-webp --without-imagemagick --with-dumping=pdumper
 'CFLAGS=-march=native -O2 -pipe' 'LDFLAGS=-Wl,-O1 -Wl,--as-needed''

Configured features:
ACL CAIRO DBUS FREETYPE GIF GLIB GMP GNUTLS GPM JPEG LCMS2 LIBSYSTEMD
MODULES NATIVE_COMP NOTIFY INOTIFY PDUMPER PGTK PNG RSVG SECCOMP SOUND
THREADS TIFF TOOLKIT_SCROLL_BARS TREE_SITTER XIM GTK3 ZLIB

Important settings:
  value of $LANG: en_US.UTF-8
  locale-coding-system: utf-8-unix

Major mode: Lisp Interaction

Minor modes in effect:
  tooltip-mode: t
  global-eldoc-mode: t
  eldoc-mode: t
  show-paren-mode: t
  electric-indent-mode: t
  mouse-wheel-mode: t
  menu-bar-mode: t
  file-name-shadow-mode: t
  global-font-lock-mode: t
  font-lock-mode: t
  blink-cursor-mode: t
  column-number-mode: t
  line-number-mode: t
  indent-tabs-mode: t
  transient-mark-mode: t
  auto-composition-mode: t
  auto-encryption-mode: t
  auto-compression-mode: t

Load-path shadows:
None found.

Features:
(shadow sort mail-extr emacsbug message yank-media puny dired
dired-loaddefs rfc822 mml mml-sec epa derived epg rfc6068 epg-config
gnus-util text-property-search time-date mm-decode mm-bodies mm-encode
mail-parse rfc2231 mailabbrev gmm-utils mailheader sendmail rfc2047
rfc2045 ietf-drums mm-util mail-prsvr mail-utils cus-start cus-load
wid-edit descr-text enriched disp-table facemenu comp comp-cstr warnings
icons rx cl-extra help-mode manoj-dark-theme site-gentoo
ranger-autoloads scopeline-autoloads package browse-url url url-proxy
url-privacy url-expand url-methods url-history url-cookie
generate-lisp-file url-domsuf url-util mailcap url-handlers url-parse
auth-source cl-seq eieio eieio-core cl-macs password-cache json subr-x
map byte-opt gv bytecomp byte-compile url-vars cl-loaddefs cl-lib rmc
iso-transl tooltip cconv eldoc paren electric uniquify ediff-hook
vc-hooks lisp-float-type elisp-mode mwheel term/pgtk-win pgtk-win
term/common-win pgtk-dnd tool-bar dnd fontset image regexp-opt fringe
tabulated-list replace newcomment text-mode lisp-mode prog-mode register
page tab-bar menu-bar rfn-eshadow isearch easymenu timer select
scroll-bar mouse jit-lock font-lock syntax font-core term/tty-colors
frame minibuffer nadvice seq simple cl-generic indonesian philippine
cham georgian utf-8-lang misc-lang vietnamese tibetan thai tai-viet lao
korean japanese eucjp-ms cp51932 hebrew greek romanian slovak czech
european ethiopic indian cyrillic chinese composite emoji-zwj charscript
charprop case-table epa-hook jka-cmpr-hook help abbrev obarray oclosure
cl-preloaded button loaddefs theme-loaddefs faces cus-face macroexp
files window text-properties overlay sha1 md5 base64 format env
code-pages mule custom widget keymap hashtable-print-readable backquote
threads dbusbind inotify dynamic-setting font-render-setting cairo gtk
pgtk lcms2 multi-tty make-network-process native-compile emacs)

Memory information:
((conses 16 121243 14450)
 (symbols 48 22924 0)
 (strings 32 87992 2869)
 (string-bytes 1 2065634)
 (vectors 16 27491)
 (vector-slots 8 1623278 223666)
 (floats 8 58 48)
 (intervals 56 908 0)
 (buffers 984 13))

bug#70000: 29.2; Grapheme handling incorrect

[Eli Zaretskii]

tags 70000 notabug
thanks

> From: Phillip Susi <phill@thesusis.net>
> Date: Mon, 25 Mar 2024 14:45:48 -0400
> 
> I had some terminal breakage the other day when browsing email with
> notmuch.  Now a ways down the rabbit hole, it seems this is because
> emacs does not correctly handle graphemes.  I found this article here:
> 
> https://mitchellh.com/writing/grapheme-clusters-in-terminals
> 
> If I paste that gramehe into GUI emacs, it is displayed as two separate
> characters, each two columns wide, instead of the correct way: as a
> single double wide character.

First, the above blog talks about text-mode terminals (a.k.a. "TTYs"),
so it is not relevant to GUI Emacs session.

And second, how that particular sequence of codepoints is displayed on
GUI frames depends on how your Emacs was built.  According to the list
of features included in your report, viz.:

  Configured features:
  ACL CAIRO DBUS FREETYPE GIF GLIB GMP GNUTLS GPM JPEG LCMS2 LIBSYSTEMD
  MODULES NATIVE_COMP NOTIFY INOTIFY PDUMPER PGTK PNG RSVG SECCOMP SOUND
  THREADS TIFF TOOLKIT_SCROLL_BARS TREE_SITTER XIM GTK3 ZLIB

your Emacs is built without HarfBuzz, which I think explains why your
Emacs displays the above sequences as 2 separate characters.
Furthermore, the appearance depends on the fonts you have installed;
specifically, Emoji sequences need a font that has a good support of
the Emoji Unicode blocks.  In my Emacs, which does use HarfBuzz, I see
a single grapheme cluster.

> C-f and C-b move over the character as if
> it were one, however, backspace deletes only the second, leaving both
> the first and the zero width joiner.  If C-f and C-b treat it as one,
> then so should backspace.

That Backspace deletes a single codepoint is a feature: it allows
easier editing of composable character sequences, such as Emoji.
E.g., imagine you want to make a slight change to the Emoji by
modifying just the second of the two characters composed into a
grapheme cluster.  Emacs supports deletion of the entire grapheme
cluster with the command delete-forward-char, by default bound to the
<Delete> function key.

> Under recent versions of the foot terminal emulator, this character is
> displayed as a single, double wide character, but emacs assumes it still
> is 4 colums wide, leading to terminal breakage.

Emacs cannot know what the terminal does with these characters,
because there's no widely-accepted protocol for accessing that
information.  Different terminal emulators behave differently, and
some even have options to modify their behavior via the various
settings.

> Emacs needs to not assume the width of graphemes are what wcwidth()
> reports, but instead need to query the cursor position after
> printing one to find out how wide the terminal actually dispalyed it
> as.

Querying the cursor position won't help in this case because it is
Emacs that moves the cursor when you type C-f, not the terminal.

I see no Emacs bug here.  Until we have standard ways of querying
text-mode terminals about their processing of composable character
sequences into grapheme clusters, there's no way for Emacs to behave
correctly with all such terminal emulators.  Sorry.

bug#70000: 29.2; Grapheme handling incorrect

[Phillip Susi]

Eli Zaretskii <eliz@gnu.org> writes:

> Querying the cursor position won't help in this case because it is
> Emacs that moves the cursor when you type C-f, not the terminal.

I'm not talking about C-f, but simply displaying the characters on the
screen.  Emacs assumes the width is 4 when it prints this character, and
so it thinks that the cursor moved over 4 places.  When the terminal
actually only moves the cursor over 2 spaces, emacs gets out of sync
with the terminal, and massive breakage occurs.

By reading back the cursor position from the terminal after displaying a
grapheme cluster, it would learn how the terminal displayed it and
update its idea of where the cursor is correctly.

I originally ran into this problem not with a ZWJ, but with an emoji
followed by alternate selector 16 that someone used in a subject line of
an email, and when browsing my inbox with notmuch, the terminal went
FUBAR.

bug#70000: 29.2; Grapheme handling incorrect

[Eli Zaretskii]

> From: Phillip Susi <phill@thesusis.net>
> Cc: 70000@debbugs.gnu.org
> Date: Wed, 27 Mar 2024 10:11:30 -0400
> 
> Eli Zaretskii <eliz@gnu.org> writes:
> 
> > Querying the cursor position won't help in this case because it is
> > Emacs that moves the cursor when you type C-f, not the terminal.
> 
> I'm not talking about C-f, but simply displaying the characters on the
> screen.  Emacs assumes the width is 4 when it prints this character, and
> so it thinks that the cursor moved over 4 places.  When the terminal
> actually only moves the cursor over 2 spaces, emacs gets out of sync
> with the terminal, and massive breakage occurs.

I understand what you are saying, but this is not how Emacs display
code works.  It needs to know the width of every character displayed
on the screen, and it needs to be able to determine that even without
actually displaying the character.

When Emacs is about to redraw some portion of the screen, it moves the
cursor to that place.  To be able to move the cursor there, it needs
to be able to compute the coordinates on the screen of every character
that is currently shown, so it can construct the command for the
terminal driver to move cursor to that place.  If Emacs were to rely
on displaying characters for that, it would have needed to constantly
redraw large portions of the screen, and that would both be much
slower and cause unpleasant flickering of the display, due to
redrawing of screen portions that don't actually change.

So this technique is out of the question for Emacs.

> By reading back the cursor position from the terminal after displaying a
> grapheme cluster, it would learn how the terminal displayed it and
> update its idea of where the cursor is correctly.

I understand.  But Emacs needs this information also long after the
characters were already drawn.  For example, imagine that Emacs
displays these characters on the screen, and then leaves most of the
screen intact and periodically redraws some small portion of the
screen, like updating current time in the lower-right corner of the
screen when Emacs is otherwise idle.  To do that, Emacs needs to move
the cursor from its current position somewhere on the screen to the
lower-right corner, redraw the time there, then move the cursor back
to where it was.  These cursor moves are based on the ability to
calculate the geometry of each character on display without actually
writing the characters to the screen.

In addition, if Emacs had to query the cursor position after each
written character, its redisplay would be much slower than it is now.

> I originally ran into this problem not with a ZWJ, but with an emoji
> followed by alternate selector 16 that someone used in a subject line of
> an email, and when browsing my inbox with notmuch, the terminal went
> FUBAR.

Yes, that's a known issue with some of the terminal emulators that
compose Emoji and other similar character sequences into grapheme
clusters, while ignoring the width that is expected from the result.
I'm not aware of any good solution, unfortunately.  Sometimes,
disabling auto-composition-mode helps, but even that cannot solve all
the problems, especially when each of the characters composed by the
terminal into a single grapheme cluster has non-zero width according
to the Unicode tables.  (If only the first character in the composed
sequence has non-zero width and the rest are zero-width, disabling
auto-composition-mode might produce a correct display.)

The bottom line is what I said at the beginning: we need some protocol
by which a terminal emulator could be queried about whether it
supports character composition, and if so, what is the screen width of
a given sequence of codepoints that will be composed, without actually
displaying them.  Better yet, some standard table of such widths could
be accepted by complying terminal emulators, and then Emacs could use
such a table to know the width in advance (similarly to how it knows
that from the Unicode data files).

Until such protocols or tables exist, Emacs will be unable to produce
correct display on these terminal emulators.

bug#70000: 29.2; Grapheme handling incorrect

[Phillip Susi]

Eli Zaretskii <eliz@gnu.org> writes:

> I understand.  But Emacs needs this information also long after the
> characters were already drawn.  For example, imagine that Emacs

Yes, it would have to learn the width the first time it displays each
grapheme and build a list of known widths to remember for future use.

> In addition, if Emacs had to query the cursor position after each
> written character, its redisplay would be much slower than it is now.

It would only need to query when printing a grapheme cluster, and only
the first time.  After that, it could remeber.