[xemacs-hg @ 2002-06-20 21:18:01 by ben] font changes etc.; some 21.4 changes mule/mule-msw-init-late.el: Specify charset->windows-registry conversion. mule/mule-x-init.el: Delete extra mule font additions here. Put them in faces.c. cl-macs.el: Document better. font-lock.el: Move Lisp function regexp to lisp-mode.el. lisp-mode.el: Various indentation fixes: Handle flet functions better. Handle argument lists in defuns and flets. Handle quoted lists, e.g. property lists -- don't indent like function calls. Distinguish between lambdas and other lists. lisp-mode.el: Handle this form. faces.el, font-menu.el, font.el, gtk-faces.el, msw-faces.el, msw-font-menu.el, x-faces.el, x-init.el: Major overhaul of face-handling code: -- Fix lots of bogus code in msw-faces.el, msw-font-menu.el, font-menu.el that was "truenaming" font specs -- i.e. in the process of frobbing a particular field in a general user-specified font spec with wildcarded fields, sticking in particular values for all the remaining wildcarded fields. This bug was rampant everywhere except in x-faces.el (the oldest and only correctly written code). This also means that we need to work with font names at all times and not font instances, because a font instance is essentially a truenamed font. -- Total rewrite of extremely junky code in msw-faces.el. Work with names as well as font instances, and return names; stop truenaming when canonicalizing and frobbing; fix handling of the combined style field, i.e. weight/slant (also fixed in font.el). -- Totally rewrite the frobbing functions in faces.el. This time, we frob all the instantiators rather than just computing a single instance value and working backwards. That way, e.g., `bold' will work for all charsets that have bold available, rather than only for whatever charset was part of the computed font instance (another example of the truename virus). Also fix up code to look at the fallbacks (all of them) when no global value present, so we don't need to put something in the global value. Intelligently handle a request to frob a buffer locale, rather than signalling an error. When frobbing instantiators, try hard to figure out what device type is associated with them, and frob each according to its own proper device type. Correctly handle inheritance vectors given as instantiators. Preserve existing tags when putting back frobbed instantiators. Extract out general specifier-frobbing code into specifier.el. Document everything cleanly. Do lots of other things better, etc. -- Don't duplicatively set a global specification for the default font -- it's already in the fallback and we no longer need a default global specification present. Delete various code in x-faces.el and msw-faces.el that duplicated the lists of fonts in faces.c. -- init-global-faces was not being called at all under MS Windows! Major bogosity. That caused device-specific values to get stuck into all the fonts, making it very hard to change them -- setting global specs caused nothing to happen. -- Correct weight names in font.el. -- Lots more font fixups in objects*.c. Printer.el: Warning fix. specifier.el: Add more args to map-specifier. Add various "heuristic" specifier functions to aid in creation of specifier-munging code such as in faces.el. subr.el: New functions. lwlib.c: Fix warning. config.inc.samp: Clean up, add args to control fastcall (not yet supported! the changes needed are in another ws of mine), profile support, vc6 support, union-type. xemacs.dsp, xemacs.mak: Semi-major overhaul. Fix bug where dump-id was always getting recomputed, forcing a redump even when nothing changed. Add support for fastcall. Support edit-and-continue (on by default) with vc6. Use incremental linking when doing a debug compilation. Add support for profiling. Consolidate the various debug flags. Partial support for "batch-compiling" -- compiling many files on a single invocation of the compiler. Doesn't seem to help that much for me, so it's not finished or enabled by default. Remove HAVE_MSW_C_DIRED, we always do. Correct some sloppy use of directories. s/cygwin32.h: Allow pdump to work under Cygwin (mmap is broken, so need to undefine HAVE_MMAP). s/win32-common.h, s/windowsnt.h: Support for fastcall. Add WIN32_ANY for identifying all Win32 variants (Cygwin, native, MinGW). Both of these are properly used in another ws. alloc.c, balloon-x.c, buffer.c, bytecode.c, callint.c, cm.c, cmdloop.c, cmds.c, console-gtk.c, console-gtk.h, console-msw.c, console-msw.h, console-stream.c, console-stream.h, console-tty.c, console-tty.h, console-x.c, console-x.h, console.c, console.h, device-gtk.c, device-msw.c, device-tty.c, device-x.c, device.c, device.h, devslots.h, dialog-gtk.c, dialog-msw.c, dialog-x.c, dialog.c, dired-msw.c, editfns.c, emacs.c, event-Xt.c, event-gtk.c, event-msw.c, event-stream.c, event-tty.c, event-unixoid.c, events.c, extents.c, extents.h, faces.c, fileio.c, fns.c, frame-gtk.c, frame-msw.c, frame-tty.c, frame-x.c, frame.c, frame.h, glyphs-eimage.c, glyphs-gtk.c, glyphs-msw.c, glyphs-widget.c, glyphs-x.c, glyphs.c, glyphs.h, gui-gtk.c, gui-msw.c, gui-x.c, gui.c, gutter.c, input-method-xlib.c, intl-encap-win32.c, intl-win32.c, keymap.c, lisp.h, macros.c, menubar-gtk.c, menubar-msw.c, menubar-x.c, menubar.c, menubar.h, minibuf.c, mule-charset.c, nt.c, objects-gtk.c, objects-gtk.h, objects-msw.c, objects-msw.h, objects-tty.c, objects-tty.h, objects-x.c, objects-x.h, objects.c, objects.h, postgresql.c, print.c, process.h, redisplay-gtk.c, redisplay-msw.c, redisplay-output.c, redisplay-tty.c, redisplay-x.c, redisplay.c, redisplay.h, scrollbar-gtk.c, scrollbar-msw.c, scrollbar-x.c, scrollbar.c, select-gtk.c, select-msw.c, select-x.c, select.c, signal.c, sound.c, specifier.c, symbols.c, syntax.c, sysdep.c, syssignal.h, syswindows.h, toolbar-common.c, toolbar-gtk.c, toolbar-msw.c, toolbar-x.c, toolbar.c, unicode.c, window.c, window.h: The following are the major changes made: (1) Separation of various header files into an external and an internal version, similar to the existing separation of process.h and procimpl.h. Eventually this should be done for all Lisp objects. The external version has the same name as currently; the internal adds -impl. The external file has XFOO() macros for objects, but the structure is opaque and defined only in the internal file. It's now reasonable to move all prototypes in lisp.h into the appropriate external file, and this should be done. Currently, separation has been done on extents.h, objects*.h, console.h, device.h, frame.h, and window.h. For c/d/f/w, the most basic properties are available in the external header file, with the macros resolving to functions. In the internal header file, the macros are redefined to directly access the structure. Also, the global MARK_FOO_CHANGED macros have been made into functions so that they can be accessed without needing to include lots of -impl headers -- they are used in almost exclusively in non-time-critical functions, and take up enough time that the function overhead will be negligible. Similarly, the function overhead from making the basic properties mentioned above into functions is negligible, and code that does heavy accessing of c/d/f/w structures inevitably ends up needing the internal header files, anyway. (2) More face changes. -- Major rewrite of objects-msw.c. Now handles wildcard specs properly, rather than "truenaming" (or even worse, signalling an error, which previously happened with some of the fallbacks if you tried to use them in make-font-instance!). -- Split charset matching of fonts into two stages -- one to find a font specifically designed for a particular charset (by examining its registry), the second to find a Unicode font that can support the charset. This needs to proceed as two complete, separate instantiations in order to work properly (otherwise many of the fonts in the HELLO page look wrong). This should also make it easy to support iso10646 (Unicode) fonts under X. -- All default values for fonts are now completely specified in the fallbacks. Stuff from mule-x-init.el has all been moved here, merged with the existing specs, and totally rethought so you get sensible results. (HELLO now looks much better!). -- Generalize the "default X/GTK device" stuff into a per-device-type "default device". -- Add mswindows-{set-}charset-registry. In time, charset<->code-page conversion functions will be removed. -- Wrap protective code around calls to compute device specifier tags, and do this computation before calling the face initialization code because the latter may need these tags to be correctly updated. (3) Other changes. EmacsFrame.c, glyphs-msw.c, eval.c, gui-x.c, intl-encap-win32.c, search.c, signal.c, toolbar-msw.c, unicode.c: Warning fixes. config.h.in: #undefs meant to be frobbed by configure *MUST* go inside of #ifndef WIN32_NO_CONFIGURE, and everything else *MUST* go outside! eval.c: Let detailed backtraces be detailed. specifier.c: Don't override user's print-string-length/print-length settings. glyphs.c: New function image-instance-instantiator. config.h.in, sysdep.c: Changes for fastcall. sysdep.c, nt.c: Fix up a previous botched patch that tried to add support for both EEXIST and EACCES. IF THE BOTCHED PATCH WENT INTO 21.4, THIS FIXUP NEEDS TO GO IN, TOO. search.c: Fix *evil* crash due to incorrect synching of syntax-cache code with 21.1. THIS SHOULD GO INTO 21.4.

--- a/src/device.c	Tue Jun 11 19:28:22 2002 +0000
+++ b/src/device.c	Thu Jun 20 21:19:10 2002 +0000
@@ -32,12 +32,13 @@
 
 #include "buffer.h"
 #include "console.h"
-#include "device.h"
+#include "device-impl.h"
 #include "elhash.h"
 #include "events.h"
 #include "faces.h"
-#include "frame.h"
+#include "frame-impl.h"
 #include "keymap.h"
+#include "objects.h"
 #include "redisplay.h"
 #include "specifier.h"
 #include "sysdep.h"
@@ -53,11 +54,13 @@
 /* Vdefault_device is the firstly-created non-stream device that's still
    around.  We don't really use it anywhere currently, but it might
    be used for resourcing at some point.  (Currently we use
-   Vdefault_x_device.) */
+   the default X device -- see Vdefault_device_plist.) */
 Lisp_Object Vdefault_device;
 
 Lisp_Object Vcreate_device_hook, Vdelete_device_hook;
 
+static Lisp_Object Vdefault_device_plist;
+
 /* Device classes */
 /* Qcolor defined in general.c */
 Lisp_Object Qgrayscale, Qmono;
@@ -173,9 +176,7 @@
   d->font_instance_cache =
     make_lisp_hash_table (20, HASH_TABLE_KEY_WEAK, HASH_TABLE_EQUAL);
 #ifdef MULE
-  /* Note that the following table is bi-level. */
-  d->charset_font_cache =
-    make_lisp_hash_table (20, HASH_TABLE_NON_WEAK, HASH_TABLE_EQ);
+  initialize_charset_font_caches (d);
 #endif
   /*
      Note that the image instance cache is actually bi-level.
@@ -210,6 +211,24 @@
   return DFW_DEVICE (object);
 }
 
+Lisp_Object
+device_console (struct device *d)
+{
+  return DEVICE_CONSOLE (d);
+}
+
+int
+device_live_p (struct device *d)
+{
+  return DEVICE_LIVE_P (d);
+}
+
+Lisp_Object
+device_frame_list (struct device *d)
+{
+  return DEVICE_FRAME_LIST (d);
+}
+
 
 DEFUN ("selected-device", Fselected_device, 0, 1, 0, /*
 Return the device which is currently active.
@@ -333,8 +352,6 @@
   return DEVICE_CONSOLE (decode_device (device));
 }
 
-#ifdef HAVE_WINDOW_SYSTEM
-
 static void
 init_global_resources (struct device *d)
 {
@@ -347,8 +364,6 @@
 #endif
 }
 
-#endif
-
 static void
 init_device_resources (struct device *d)
 {
@@ -361,6 +376,45 @@
 #endif
 }
 
+DEFUN ("default-device", Fdefault_device, 0, 1, 0, /*
+Return the default device of type TYPE.
+This is generally the first-created device of that TYPE that still exists.
+It is used for resourcing and certain other things.  On MS Windows, it
+is not very useful because there is generally only one device.
+If TYPE is omitted, it is derived from the selected device.
+If there is no default device of TYPE, nil is returned.
+*/
+       (type))
+{
+  if (NILP (type))
+    type = XDEVICE_TYPE (Fselected_device (Qnil));
+  else
+    /* For errors */
+    decode_console_type (type, ERROR_ME);
+
+  return Fplist_get (Vdefault_device_plist, type, Qnil);
+}
+
+/* Return the default device for a device type. */
+Lisp_Object
+get_default_device (Lisp_Object type)
+{
+  return Fplist_get (Vdefault_device_plist, type, Qnil);
+}
+
+/* Set the default device for a device type. */
+void
+set_default_device (Lisp_Object type, Lisp_Object device)
+{
+  Vdefault_device_plist = Fplist_put (Vdefault_device_plist, type, device);
+}
+
+void
+clear_default_devices (void)
+{
+  Vdefault_device_plist = Qnil;
+}
+
 static Lisp_Object
 semi_canonicalize_device_connection (struct console_methods *meths,
 				     Lisp_Object name, Error_Behavior errb)
@@ -510,27 +564,16 @@
   int speccount = specpdl_depth();
 
   struct gcpro gcpro1, gcpro2, gcpro3;
-#ifdef HAVE_X_WINDOWS
-  /* #### icky-poo.  If this is the first X device we are creating,
-     then retrieve the global face resources.  We have to do it
-     here, at the same time as (or just before) the device face
-     resources are retrieved; specifically, it needs to be done
-     after the device has been created but before any frames have
-     been popped up or much anything else has been done.  It's
-     possible for other devices to specify different global
-     resources (there's a property on each X server's root window
-     that holds some resources); tough luck for the moment.
-
-     This is a nasty violation of device independence, but
-     there's not a whole lot I can figure out to do about it.
-     The real problem is that the concept of resources is not
-     generalized away from X.  Similar resource-related
-     device-independence violations occur in faces.el. */
-  int first_x_device = NILP (Vdefault_x_device) && EQ (type, Qx);
-#endif
-#ifdef HAVE_GTK
-  int first_gtk_device = NILP (Vdefault_gtk_device) && EQ (type, Qgtk);
-#endif
+  /* If this is the first device we are creating of a particular type
+     (e.g. X), then retrieve the global face resources.  We have to do it
+     here, at the same time as (or just before) the device face resources
+     are retrieved; specifically, it needs to be done after the device has
+     been created but before any frames have been popped up or much
+     anything else has been done.  It's possible for other devices to
+     specify different global resources (there's a property on each X
+     server's root window that holds some resources); tough luck for the
+     moment. */
+  int first = NILP (get_default_device (type));
 
   GCPRO3 (device, console, name);
 
@@ -553,7 +596,7 @@
     console = create_console (name, type, conconnect, props);
   }
 
-  record_unwind_protect(delete_deviceless_console, console);
+  record_unwind_protect (delete_deviceless_console, console);
 
   con = XCONSOLE (console);
   d = allocate_device (console);
@@ -571,6 +614,10 @@
 
   /* Do it this way so that the device list is in order of creation */
   con->device_list = nconc2 (con->device_list, Fcons (device, Qnil));
+
+  if (NILP (get_default_device (type)))
+    set_default_device (type, device);
+
   note_object_created (device);
 
   RESET_CHANGED_SET_FLAGS;
@@ -578,25 +625,28 @@
     Vdefault_device = device;
 
   init_device_sound (d);
-#ifdef HAVE_X_WINDOWS
-  if (first_x_device)
-    init_global_resources (d);
-#endif
-#ifdef HAVE_GTK
-  if (first_gtk_device)
-    init_global_resources (d);
-#endif
-  init_device_resources (d);
-
-  MAYBE_DEVMETH (d, finish_init_device, (d, props));
 
   /* If this is the first device on the console, make it the selected one. */
   if (NILP (CONSOLE_SELECTED_DEVICE (con)))
     CONSOLE_SELECTED_DEVICE (con) = device;
 
-  /* #### the following should trap errors. */
+  /* Needed before initialization of resources because they may do things
+     with the tags, esp. the face code.  For example,
+     init-other-random-faces calls face-property-instance, and the
+     specifier code checks inst-pairs by seeing if the device matches the
+     tag; this fails for tags such as `default', if we haven't set up the
+     tags yet. */
   setup_device_initial_specifier_tags (d);
 
+  if (!EQ (type, Qstream))
+    {
+      if (first)
+	init_global_resources (d);
+      init_device_resources (d);
+    }
+
+  MAYBE_DEVMETH (d, finish_init_device, (d, props));
+
   UNGCPRO;
   unbind_to (speccount);
   return device;
@@ -807,6 +857,29 @@
 
   MAYBE_DEVMETH (d, delete_device, (d));
 
+  /* Now see if we're the default device, and thus need to be changed. */
+  {
+    /* Device type still OK, not set to null till down below. */
+    Lisp_Object dt = DEVICE_TYPE (d);
+
+    if (EQ (device, get_default_device (dt)))
+      {
+	Lisp_Object devcons, concons;
+	/* #### handle deleting last device */
+	set_default_device (dt, Qnil);
+	DEVICE_LOOP_NO_BREAK (devcons, concons)
+	  {
+	    if (EQ (dt, XDEVICE_TYPE (XCAR (devcons))) &&
+		!EQ (device, XCAR (devcons)))
+	      {
+		set_default_device (dt, XCAR (devcons));
+		goto double_break;
+	      }
+	  }
+      }
+  }
+ double_break:
+
   CONSOLE_DEVICE_LIST (c) = delq_no_quit (device, CONSOLE_DEVICE_LIST (c));
 
   RESET_CHANGED_SET_FLAGS;
@@ -1219,7 +1292,7 @@
   return Qnil;
 }
 
-void
+Lisp_Object
 call_critical_lisp_code (struct device *d, Lisp_Object function,
 			 Lisp_Object object)
 {
@@ -1227,6 +1300,7 @@
   int count = begin_gc_forbidden ();
   struct gcpro gcpro1;
   Lisp_Object args[3];
+  Lisp_Object retval;
 
   specbind (Qinhibit_quit, Qt);
   record_unwind_protect (unlock_device, wrap_device (d));
@@ -1248,11 +1322,11 @@
 
   /* It's useful to have an error handler; otherwise an infinite
      loop may result. */
-  Fcall_with_condition_handler (!NILP (object) ? 3 : 2, args);
+  retval = Fcall_with_condition_handler (!NILP (object) ? 3 : 2, args);
 
   UNGCPRO;
 
-  unbind_to (count);
+  return unbind_to_1 (count, retval);
 }
 
 
@@ -1280,6 +1354,7 @@
   DEFSUBR (Ffind_device);
   DEFSUBR (Fget_device);
   DEFSUBR (Fmake_device);
+  DEFSUBR (Fdefault_device);
   DEFSUBR (Fdelete_device);
   DEFSUBR (Fdevice_frame_list);
   DEFSUBR (Fdevice_class);
@@ -1369,6 +1444,10 @@
 */ );
   Vdelete_device_hook = Qnil;
 
+  /* Plist of device types and their default devices. */
+  Vdefault_device_plist = Qnil;
+  staticpro (&Vdefault_device_plist);
+
   Vdevice_class_list = list3 (Qcolor, Qgrayscale, Qmono);
   staticpro (&Vdevice_class_list);