[troff]: Fix Savannah #64484.

* src/roff/troff/input.cpp (encode_char_for_troff_output): Annotate the
  function's purpose.  Initially assume the character to be encoded as
  valid.  If the current token is a plain space, write a space (U+0020)
  to the output.  (This is necessary because the `device` request no
  longer reads its arguments in copy mode; see below.)  Move the `sc`
  local variable to a higher scope.  Update the new `is_char_valid`
  Boolean instead of issuing an error diagnostic at each point of
  validation failure.  When done processing the character, test
  `is_char_valid` and emit different diagnostics depending on whether
  the input was a special character escape sequence we can't handle, or
  something else.  Emit a self-quoted escape character _as a backslash_,
  not as the current *roff escape character.

  (device_request): Rewrite to operate in interpretation mode, not copy
  mode.

* doc/groff.texi (Postprocessor Access):
* man/groff.7.man (Request short reference):
* NEWS: Document it.

Fixes <https://savannah.gnu.org/bugs/?64484>.

5 files changed, 107 insertions, 68 deletions

diff --git a/ChangeLog b/ChangeLog
index 45db977cc..446043386 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,5 +1,31 @@
 2024-01-13  G. Branden Robinson <g.branden.robinson@gmail.com>
 
+	[troff]: Fix Savannah #64484.
+
+	* src/roff/troff/input.cpp (encode_char_for_troff_output):
+	Annotate the function's purpose.  Initially assume the character
+	to be encoded as valid.  If the current token is a plain space,
+	write a space (U+0020) to the output.  (This is necessary
+	because the `device` request no longer reads its arguments in
+	copy mode; see below.)  Move the `sc` local variable to a higher
+	scope.  Update the new `is_char_valid` Boolean instead of
+	issuing an error diagnostic at each point of validation failure.
+	When done processing the character, test `is_char_valid` and
+	emit different diagnostics depending on whether the input was a
+	special character escape sequence we can't handle, or something
+	else.  Emit a self-quoted escape character _as a backslash_, not
+	as the current *roff escape character.
+	(device_request): Rewrite to operate in interpretation mode, not
+	copy mode.
+
+	* doc/groff.texi (Postprocessor Access):
+	* man/groff.7.man (Request short reference):
+	* NEWS: Document it.
+
+	Fixes <https://savannah.gnu.org/bugs/?64484>.
+
+2024-01-13  G. Branden Robinson <g.branden.robinson@gmail.com>
+
 	* src/roff/groff/tests/\
 	device-control-special-character-handling.sh: Add unit test for
 	this feature.  We want to be able to consistently pass (some)
@@ -172,7 +198,7 @@
 	"default", instead of permitting it to have a null name.
 
 	* doc/groff.texi (Colors):
-	* man/groff.7.man (Read-only regsiters):
+	* man/groff.7.man (Read-only registers):
 	* NEWS: Document it.
 
 	Fixes <https://savannah.gnu.org/bugs/?64592>.  Thanks to Deri
diff --git a/NEWS b/NEWS
index adf27d093..7f5e8894c 100644
--- a/NEWS
+++ b/NEWS
@@ -41,6 +41,15 @@ o The read-only registers `.m` and `.M` now interpolate "default" when
   the default color is selected as the stroke or fill color,
   respectively, rather than interpolating nothing.
 
+o The `device` request no longer reads its arguments in copy mode; this
+  change makes it more consistent with the `\X` device control command
+  escape sequence.  This request also no longer emits a self-quoted
+  *roff escape character as itself, but instead as a backslash.
+  (troff's input and output languges are not the same thing.)  These
+  changes are to enable postprocessors to reliably interpret device
+  control commands that wish to express arbitrary byte sequences.  For
+  example, PDF bookmarks need to be expressed in UTF-16LE.
+
 eqn
 ---
 
diff --git a/doc/groff.texi b/doc/groff.texi
index 93a86b202..6e2e1cc3e 100644
--- a/doc/groff.texi
+++ b/doc/groff.texi
@@ -16421,36 +16421,34 @@ returned by the function @cite{getenv@r{(3)}}.
 @cindex access to postprocessor
 
 Two escape sequences and two requests enable documents to pass
-information directly to a postprocessor.  These are useful for
-exercising device-specific capabilities that the @code{groff} language
-does not abstract or generalize; examples include the embedding of
-hyperlinks and image files.  Device-specific functions are documented in
-each output driver's man page, such as @cite{gropdf@r{(1)}},
-@cite{grops@r{(1)}}, or @cite{grotty@r{(1)}}.
+information directly to an output driver or other postprocessor.  These
+are useful for exercising device-specific capabilities that the
+@code{groff} language does not abstract or generalize; examples include
+the embedding of hyperlinks and image files.  Device-specific functions
+are documented in each output driver's man page, such as
+@cite{gropdf@r{(1)}}, @cite{grops@r{(1)}}, or @cite{grotty@r{(1)}}.
 
 @DefreqList {device, xxx @r{@dots{}}}
 @DefescListEndx {\\X, @code{'}, xxx @r{@dots{}}, @code{'}}
 Embed all @var{xxx} arguments into GNU @command{troff} output as
-parameters to a device control command @w{@samp{x X}}.  The meaning and
-interpretation of such parameters is determined by the output driver or
-other postprocessor.
-
-@cindex @code{device} request, and copy mode
-@cindex copy mode, and @code{device} request
-@cindex mode, copy, and @code{device} request
-The @code{device} request processes its arguments in copy mode
-(@pxref{Copy Mode}).  An initial neutral double quote in @var{contents}
-is stripped to allow embedding of leading spaces.
-@cindex @code{\&}, in @code{\X}
-@cindex @code{\)}, in @code{\X}
-@cindex @code{\%}, in @code{\X}
+parameters to an @w{@samp{x X}} device control
+command.@footnote{@xref{gtroff Output}.}  The meaning and interpretation
+of such parameters is determined by the output driver or other
+postprocessor.
+
+The @code{device} request strips an initial neutral double quote from
+@var{contents} to allow embedding of leading spaces.
+
+@cindex @code{\&}, in device control commands
+@cindex @code{\)}, in device control commands
+@cindex @code{\%}, in device control commands
 @ifnotinfo
-@cindex @code{\:}, in @code{\X}
+@cindex @code{\:}, in device control commands
 @end ifnotinfo
 @ifinfo
-@cindex @code{\@r{<colon>}}, in @code{\X}
+@cindex @code{\@r{<colon>}}, in device control commands
 @end ifinfo
-By contrast, within @code{\X} arguments, the escape sequences @code{\&},
+Within a device control command, the escape sequences @code{\&},
 @code{\)}, @code{\%}, and @code{\:} are ignored; @code{\@key{SPC}} and
 @code{\~} are converted to single space characters; and @code{\\} has
 its escape character stripped.  So that the basic Latin subset of the
@@ -16458,13 +16456,15 @@ Unicode character set@footnote{that is, ISO@tie{}646:1991-IRV or,
 popularly, ``US-ASCII''} can be reliably encoded in device control
 commands, seven special character escape sequences (@samp{\-},
 @samp{\[aq]}, @samp{\[dq]}, @samp{\[ga]}, @samp{\[ha]}, @samp{\[rs]},
-and @samp{\[ti]},) are mapped to basic Latin characters; see the
+and @samp{\[ti]}) are mapped to basic Latin characters; see the
 @cite{groff_char@r{(7)}} man page.  For this transformation, character
 translations and special character definitions are
 ignored.@footnote{They are bypassed because these parameters are not
 rendered as glyphs in the output; instead, they remain abstract
-characters---in a PDF bookmark or a URL, for example.} The use of any
-other escape sequence in @code{\X} parameters is normally an error.
+characters---in a PDF bookmark or a URL, for example.}
+
+Escape sequences other than the foregoing in device control command
+may be ignored, or produce an error.
 
 A device control command issued with the @code{device} request will not
 be reflected in the output unless a partially collected line exists at
@@ -16483,6 +16483,9 @@ discussed above).  @code{use_charnames_in_special} is currently employed
 only by @code{grohtml}.
 @endDefesc
 
+GNU @command{troff} also permits the interpolatation of macro contents
+as a device control command.
+
 @DefreqList {devicem, name}
 @DefescItemx {\\Y, , n, }
 @DefescItem {\\Y, (, nm, }
@@ -16490,12 +16493,12 @@ only by @code{grohtml}.
 This is approximately equivalent to @samp{\X'\*[@var{name}]'}
 (one-character name@tie{}@var{n}, two-character name @var{nm}).
 However, the contents of the string or macro @var{name} are not
-interpreted; also it is permitted for @var{name} to have been defined as
-a macro and thus contain newlines (it is not permitted for the argument
-to @code{\X} to contain newlines).  The inclusion of newlines requires
-an extension to the @acronym{AT&T} @command{troff} output format, and
-confuses drivers that do not know about this extension (@pxref{Device
-Control Commands}).
+interpreted; it is also permitted for @var{name} to have been defined as
+a macro and thus contain newlines.  (There is no way to embed a newline
+in the arguments to @code{device} or @code{\X}.)  The inclusion of
+newlines requires an extension to the @acronym{AT&T} @command{troff}
+output format; their presence confuses drivers that do not know about it
+(@pxref{Device Control Commands}).
 @endDefesc
 
 @DefreqList {tag, name}
diff --git a/man/groff.7.man b/man/groff.7.man
index 6e8e6974a..8bb3bfd84 100644
--- a/man/groff.7.man
+++ b/man/groff.7.man
@@ -2983,8 +2983,7 @@ with compatibility mode disabled when the macro is interpreted.
 .TPx
 .REQ .device anything
 Write
-.IR anything ,
-read in copy mode,
+.I anything
 to
 .I @g@troff
 output as a device control command.
diff --git a/src/roff/troff/input.cpp b/src/roff/troff/input.cpp
index a2bb11d4e..d3d700c0c 100644
--- a/src/roff/troff/input.cpp
+++ b/src/roff/troff/input.cpp
@@ -5571,14 +5571,19 @@ static node *do_non_interpreted()
   return new non_interpreted_node(mac);
 }
 
+// In troff output, we translate the escape character to '\', but it is
+// up to the postprocessor to interpret it as such.  (This mostly
+// matters for device control commands.)
 static void encode_char_for_troff_output(macro *mac, const char c)
 {
+  bool is_char_valid = true;
+  const char *sc = 0 /* nullptr */;
   if ('\0' == c) {
-    if (tok.is_stretchable_space()
-	     || tok.is_unstretchable_space())
+    if (tok.is_space()
+	|| tok.is_stretchable_space()
+	|| tok.is_unstretchable_space())
       mac->append(' ');
     else if (tok.is_special()) {
-      const char *sc;
       if (font::use_charnames_in_special) {
 	charinfo *ci = tok.get_char(true /* required */);
 	sc = ci->get_symbol()->contents();
@@ -5612,29 +5617,32 @@ static void encode_char_for_troff_output(macro *mac, const char c)
 	    mac->append(']');
 	  }
 	  else
-	    error("special character '%1' cannot be used within a"
-	          " device control escape sequence", sc);
+	    is_char_valid = false;
 	}
 	else
-	  error("special character '%1' cannot be used within a device"
-		" control escape sequence", sc);
+	  is_char_valid = false;
       }
     }
     else if (!(tok.is_hyphen_indicator()
 	       || tok.is_dummy()
 	       || tok.is_transparent_dummy()
 	       || tok.is_zero_width_break()))
-      error("%1 is invalid within device control escape sequence",
-	    tok.description());
+      is_char_valid = false;
+    if (!is_char_valid) {
+      if (sc != 0 /* nullptr */)
+	error("special character '%1' is invalid within a device"
+	      " control command", sc);
+      else
+	error("%1 is invalid within a device control command",
+	      tok.description());
+    }
   }
   else {
-    if ((font::use_charnames_in_special) && ('\\' == c)) {
-      /*
-       * add escape escape sequence
-       */
-      mac->append(c);
+    if (c == escape_char) {
+      mac->append('\\');
     }
-    mac->append(c);
+    else
+      mac->append(c);
   }
 }
 
@@ -5678,34 +5686,28 @@ static node *do_special()
 
 static void device_request()
 {
-  // We can't use `has_arg()` here because we want to read in copy mode.
-  int c;
-  for (;;) {
-    c = input_stack::peek();
-    if (' ' == c)
-      (void) get_copy(0 /* nullptr */);
-    else
-      break;
+  if (!has_arg()) {
+    warning(WARN_MISSING, "device request expects arguments");
+    skip_line();
+    return;
   }
-  if (('\n' == c) || (EOF == c)) {
-    warning(WARN_MISSING, "device control request expects arguments");
+  if (tok.is_newline() || tok.is_eof()) {
+    warning(WARN_MISSING, "device request expects arguments");
     skip_line();
     return;
   }
+  if ('"' == tok.ch()) {
+    tok.next();
+  }
   macro mac;
   for (;;) {
-    c = get_copy(0 /* nullptr */);
-    if ('"' == c) {
-      c = get_copy(0 /* nullptr */);
-      break;
-    }
-    if (c != ' ' && c != '\t')
+    if (tok.is_newline() || tok.is_eof())
       break;
+    encode_char_for_troff_output(&mac, tok.ch());
+    tok.next();
   }
-  for (; c != '\n' && c != EOF; c = get_copy(0 /* nullptr */))
-    mac.append(c);
   curenv->add_node(new special_node(mac));
-  tok.next();
+  skip_line();
 }
 
 static void device_macro_request()