Emacs with Cocoa/GNUstep

Emacs with Cocoa/GNUstep

[Germán Arias]

Hi, I'm trying to run emacs with GNUstep, but it seems to be totally
broken. So my question is if someone is running emacs with Cocoa and if
it works fine. Because with latest gnustep packages don't work (compile
but can't run). And after the last update in file configure.in, the
class nsmenu.m can't compile. So I can guess there isn't testing in
GNUstep/Cocoa. Is just a question before trying to get working emacs
with gnustep. Thanks.

Re: Emacs with Cocoa/GNUstep

[David Reitter]

On Apr 26, 2011, at 8:39 PM, Germán Arias wrote:
> 
> broken. So my question is if someone is running emacs with Cocoa and if
> it works fine. 

Yes, it works as intended, at least as of April 1st, when the last sync to the Git repository took place.

Re: Emacs with Cocoa/GNUstep

[CHENG Gao]

It works for me. I build from bzr repo frequently and now I am using
Apr. 25 build.

,----
| GNU Emacs 24.0.50.1 (x86_64-apple-darwin10.7.0, NS apple-appkit-1038.35)
|  of 2011-04-25
`----

Re: Emacs with Cocoa/GNUstep

[Germán Arias]

On mié, 2011-04-27 at 11:56 +0800, CHENG Gao wrote:
> It works for me. I build from bzr repo frequently and now I am using
> Apr. 25 build.
> 
> ,----
> | GNU Emacs 24.0.50.1 (x86_64-apple-darwin10.7.0, NS apple-appkit-1038.35)
> |  of 2011-04-25
> `----

Hi, I have the revision:

revision-id:
bkey76@gmail.com-20110427021744-cs2b8xl1kzfv0sj4                  
date: 2011-04-26 21:17:44 -0500
build-date: 2011-04-26 22:24:07 -0600
revno: 104022
branch-nick: trunk

In gNewSense Delta H with GCC 4.2 and latest GNUstep packages. Compiling
with:

./autogen.sh
./configure --with-ns
make bootstrap (if necessary)
make install


First I noticed some methods that are not present on GNUstep:

nsterm.m: In function ‘ns_update_end’:
nsterm.m:738: warning: ‘NSView’ may not respond to
‘-unlockFocusNeedsFlush:’
nsterm.m:738: warning: (Messages without a matching method signature
nsterm.m:738: warning: will be assumed to return ‘id’ and accept
nsterm.m:738: warning: ‘...’ as arguments.)

...
...

nsfns.m: In function ‘Fns_convert_utf8_nfd_to_nfc’:
nsfns.m:1974: warning: ‘NSString’ may not respond to
‘-precomposedStringWithCanonicalMapping’
nsfns.m:1974: warning: (Messages without a matching method signature
nsfns.m:1974: warning: will be assumed to return ‘id’ and accept
nsfns.m:1974: warning: ‘...’ as arguments.)


This is not a problem, for the moment, the problem is:

gcc -c  -Demacs -DHAVE_CONFIG_H  -I. -I/home/german/Instalados/emacs/src -I../lib -I/home/german/Instalados/emacs/src/../lib    -D_REENTRANT -fPIC -fno-strict-aliasing    -I/usr/include/libxml2         -MMD -MF deps/nsmenu.d   -Wimplicit-function-declaration -Wold-style-definition -Wdeclaration-after-statement  -g -O2 -fgnu-runtime -Wno-import -fconstant-string-class=NSConstantString -DGNUSTEP_BASE_LIBRARY=1 -DGNU_GUI_LIBRARY=1 -DGNU_RUNTIME=1 -DGSWARN -DGSDIAGNOSE nsmenu.m
nsmenu.m: In function ‘ns_update_menubar’:
nsmenu.m:229: error: ‘struct Lisp_Vector’ has no member named ‘size’
nsmenu.m:230: error: ‘struct Lisp_Vector’ has no member named ‘size’
nsmenu.m:231: error: ‘struct Lisp_Vector’ has no member named ‘size’
nsmenu.m:233: error: ‘struct Lisp_Vector’ has no member named ‘size’
nsmenu.m:235: error: ‘struct Lisp_Vector’ has no member named ‘size’
nsmenu.m:349: error: ‘struct Lisp_Vector’ has no member named ‘size’
nsmenu.m:410: error: ‘struct Lisp_Vector’ has no member named ‘size’
nsmenu.m:437: error: ‘struct Lisp_Vector’ has no member named ‘size’
nsmenu.m: In function ‘-[EmacsMenu fillWithWidgetValue:]’:
nsmenu.m:690: warning: passing argument 1 of ‘setAction:’ from
incompatible pointer type
nsmenu.m: In function ‘-[EmacsMenu addSubmenuWithTitle:forFrame:]’:
nsmenu.m:711: warning: passing argument 2 of
‘addItemWithTitle:action:keyEquivalent:’ from incompatible pointer type
make[2]: *** [nsmenu.o] Error 1
make[2]: Leaving directory `/home/german/Instalados/emacs/src'
make[1]: *** [src] Error 2
make[1]: Leaving directory `/home/german/Instalados/emacs'
make: *** [bootstrap] Error 2


I get this error after the update I did today. Before it, the app
compiled OK, but had problems to launch. I was investigating that issue,
but after the update I can't compile the app again. Because I get the
error above.

Re: Emacs with Cocoa/GNUstep

[CHENG Gao]

Just found out I forgot to mention something. I mean it works for me on
MacOSX. 

Re: Emacs with Cocoa/GNUstep

[Paul Eggert]

On 04/26/11 21:38, Germán Arias wrote:
> nsmenu.m:229: error: ‘struct Lisp_Vector’ has no member named ‘size’

That problem was due to my change to lisp.h in revno 104021.
Sorry about that.  I committed the following fix in revno 104024.
This patch can be applied independently of the earlier
lisp.h change, and it isolates nsmenu.m from that implementation
detail of lisp.h so it should be a win independently of the
lisp.h change.

2011-04-27  Paul Eggert  <eggert@cs.ucla.edu>

	* nsmenu.m: Replace all uses of XVECTOR with ASIZE and AREF.
	This makes this file independent of the recent pseudovector change.

--- src/nsmenu.m	2011-03-27 09:23:52 +0000
+++ src/nsmenu.m	2011-04-27 06:01:43 +0000
@@ -218,7 +218,7 @@
 
       /* Save the frame's previous menu bar contents data */
       if (previous_menu_items_used)
-	memcpy (previous_items, XVECTOR (f->menu_bar_vector)->contents,
+	memcpy (previous_items, &AREF (f->menu_bar_vector, 0),
 		previous_menu_items_used * sizeof (Lisp_Object));
 
       /* parse stage 1: extract from lisp */
@@ -226,19 +226,19 @@
 
       menu_items = f->menu_bar_vector;
       menu_items_allocated = VECTORP (menu_items) ? ASIZE (menu_items) : 0;
-      submenu_start = (int *) alloca (XVECTOR (items)->size * sizeof (int *));
-      submenu_end = (int *) alloca (XVECTOR (items)->size * sizeof (int *));
-      submenu_n_panes = (int *) alloca (XVECTOR (items)->size * sizeof (int));
+      submenu_start = (int *) alloca (ASIZE (items) * sizeof (int *));
+      submenu_end = (int *) alloca (ASIZE (items) * sizeof (int *));
+      submenu_n_panes = (int *) alloca (ASIZE (items) * sizeof (int));
       submenu_top_level_items
-	= (int *) alloca (XVECTOR (items)->size * sizeof (int *));
+	= (int *) alloca (ASIZE (items) * sizeof (int *));
       init_menu_items ();
-      for (i = 0; i < XVECTOR (items)->size; i += 4)
+      for (i = 0; i < ASIZE (items); i += 4)
 	{
 	  Lisp_Object key, string, maps;
 
-	  key = XVECTOR (items)->contents[i];
-	  string = XVECTOR (items)->contents[i + 1];
-	  maps = XVECTOR (items)->contents[i + 2];
+	  key = AREF (items, i);
+	  string = AREF (items, i + 1);
+	  maps = AREF (items, i + 2);
 	  if (NILP (string))
 	    break;
 
@@ -311,11 +311,11 @@
             /* FIXME: this ALWAYS fails on Buffers menu items.. something
                  about their strings causes them to change every time, so we
                  double-check failures */
-            if (!EQ (previous_items[i], XVECTOR (menu_items)->contents[i]))
+            if (!EQ (previous_items[i], AREF (menu_items, i)))
               if (!(STRINGP (previous_items[i])
-                    && STRINGP (XVECTOR (menu_items)->contents[i])
+                    && STRINGP (AREF (menu_items, i))
                     && !strcmp (SDATA (previous_items[i]),
-                               SDATA (XVECTOR (menu_items)->contents[i]))))
+				SDATA (AREF (menu_items, i)))))
                   break;
           if (i == previous_menu_items_used)
             {
@@ -346,10 +346,10 @@
       /* Parse stage 2a: now GC cannot happen during the lifetime of the
          widget_value, so it's safe to store data from a Lisp_String */
       wv = first_wv->contents;
-      for (i = 0; i < XVECTOR (items)->size; i += 4)
+      for (i = 0; i < ASIZE (items); i += 4)
 	{
 	  Lisp_Object string;
-	  string = XVECTOR (items)->contents[i + 1];
+	  string = AREF (items, i + 1);
 	  if (NILP (string))
 	    break;
 /*           if (submenu && strcmp (submenuTitle, SDATA (string)))
@@ -407,7 +407,7 @@
 
 
       /* check if no change.. this mechanism is a bit rough, but ready */
-      n = XVECTOR (items)->size / 4;
+      n = ASIZE (items) / 4;
       if (f == last_f && n_previous_strings == n)
         {
           for (i = 0; i<n; i++)
@@ -434,9 +434,9 @@
         }
 
       [menu clear];
-      for (i = 0; i < XVECTOR (items)->size; i += 4)
+      for (i = 0; i < ASIZE (items); i += 4)
 	{
-	  string = XVECTOR (items)->contents[i + 1];
+	  string = AREF (items, i + 1);
 	  if (NILP (string))
 	    break;
 
@@ -794,7 +794,7 @@
   i = 0;
   while (i < menu_items_used)
     {
-      if (EQ (XVECTOR (menu_items)->contents[i], Qnil))
+      if (EQ (AREF (menu_items, i), Qnil))
 	{
 	  submenu_stack[submenu_depth++] = save_wv;
 	  save_wv = prev_wv;
@@ -802,21 +802,21 @@
 	  first_pane = 1;
 	  i++;
 	}
-      else if (EQ (XVECTOR (menu_items)->contents[i], Qlambda))
+      else if (EQ (AREF (menu_items, i), Qlambda))
 	{
 	  prev_wv = save_wv;
 	  save_wv = submenu_stack[--submenu_depth];
 	  first_pane = 0;
 	  i++;
 	}
-      else if (EQ (XVECTOR (menu_items)->contents[i], Qt)
+      else if (EQ (AREF (menu_items, i), Qt)
 	       && submenu_depth != 0)
 	i += MENU_ITEMS_PANE_LENGTH;
       /* Ignore a nil in the item list.
 	 It's meaningful only for dialog boxes.  */
-      else if (EQ (XVECTOR (menu_items)->contents[i], Qquote))
+      else if (EQ (AREF (menu_items, i), Qquote))
 	i += 1;
-      else if (EQ (XVECTOR (menu_items)->contents[i], Qt))
+      else if (EQ (AREF (menu_items, i), Qt))
 	{
 	  /* Create a new pane.  */
 	  Lisp_Object pane_name, prefix;
@@ -906,7 +906,7 @@
 	     make the call_data null so that it won't display a box
 	     when the mouse is on it.  */
 	  wv->call_data
-	      = !NILP (def) ? (void *) &XVECTOR (menu_items)->contents[i] : 0;
+	      = !NILP (def) ? (void *) &AREF (menu_items, i) : 0;
 	  wv->enabled = !NILP (enable);
 
 	  if (NILP (type))

Re: Emacs with Cocoa/GNUstep

[Germán Arias]

On mar, 2011-04-26 at 23:13 -0700, Paul Eggert wrote:
> On 04/26/11 21:38, Germán Arias wrote:
> > nsmenu.m:229: error: ‘struct Lisp_Vector’ has no member named ‘size’
> 
> That problem was due to my change to lisp.h in revno 104021.
> Sorry about that.  I committed the following fix in revno 104024.
> This patch can be applied independently of the earlier
> lisp.h change, and it isolates nsmenu.m from that implementation
> detail of lisp.h so it should be a win independently of the
> lisp.h change.

Thanks, now compile again.

Re: Emacs with Cocoa/GNUstep

[Pascal J. Bourguignon]

Germán Arias <german@xelalug.org> writes:

> Hi, I'm trying to run emacs with GNUstep, but it seems to be totally
> broken. So my question is if someone is running emacs with Cocoa and if
> it works fine. Because with latest gnustep packages don't work (compile
> but can't run). And after the last update in file configure.in, the
> class nsmenu.m can't compile. So I can guess there isn't testing in
> GNUstep/Cocoa. Is just a question before trying to get working emacs
> with gnustep. Thanks.

Emacs works very well in MacOSX:
http://emacsformacosx.com/

I don't know about gnustep, it's been a long time.

-- 
__Pascal Bourguignon__                     http://www.informatimago.com/
A bad day in () is better than a good day in {}.

Re: Emacs with Cocoa/GNUstep

[David De La Harpe Golden]

On 27/04/11 01:39, Germán Arias wrote:
> Hi, I'm trying to run emacs with GNUstep, but it seems to be totally
> broken. So my question is if someone is running emacs with Cocoa and if
> it works fine. Because with latest gnustep packages don't work (compile
> but can't run). And after the last update in file configure.in, the
> class nsmenu.m can't compile. So I can guess there isn't testing in
> GNUstep/Cocoa.

Probably not a whole lot.... FWIW, I built it successfully several times 
a few months back, and though it was not really especially useful once 
built, it was not completely unusable either, it was working enough for 
me to debug some pasteboard interaction issues without having access to 
macosx.

I'm not personally up to doing much it about it at the moment, so here 
are some notes (Stefan: mostly same ones I passed offlist to you at the 
time) as a braindump, some of the below is probably obsolete:

First and foremost, DO NOT use any debian packages of GNUstep at time of 
writing, they're obsolete and hopelessly buggy, especially on 64-bit.  I 
wasted basically a weekend's worth of emacs-time that way, I switched to 
an svn checkout of GNUstep installed to /usr/local/GNUstep and was up 
and running fairly rapidly.

GNUstep itself is fairly painless to build from source (though n.b. I am 
used to underdocumented build scripts from hell from work, my perception 
may be skewed), the most tricky bit was getting building the 
gnustep-back backend right, so that I could use cairo.

http://wiki.gnustep.org/index.php/GNUstep_Installation_Process

The other build-time issue is that emacs "./configure --with-ns" doesn't 
seem to discover the ObjC headers properly in the gnustep case - 
specifying CPPFLAGS "fixed" that, but I suspect TRT would be to have 
emacs' configure use "gnustep-config --objc-flags" to discover the flags 
(or maybe just $GNUSTEP_MAKEFILES and the right part of gnustep's 
Makefiles, but gnustep-config seems interesting 'cos it apparently works 
much like other blah-config type tools).

I wouldn't say it is _totally_ unusable, but neither is it at all 
pleasant .  There are probably plenty more smaller issues that I'm 
presently not noticing given the big ones:

0. Make sure the relevant gnustep daemons are running.

I mean the gpbs / gdnc /gdomap  daemons. Not really an emacs problem, 
and documented in the gnustep docs, just something to be aware of.

gpbs is particularly important, as it's the GNUstep PasteBoard Server, 
and without it, clipboard/primary/secondary just ain't gonna work.

http://gnustep.made-it.com/BuildGuide/index.html#GNUSTEP.SERVICES

1. frame resizing.

The single most major annoyance (or at least tied with repaint) is that 
it doesn't handle being resized by the window manager, you have to 
(set-frame-width/height) from within emacs for it to work.  I guess 
macosx handles resizing differently.

2. repaint

There are nasty repaint issues sometimes too upon partial scrolling, a 
bit like the ones you used to see on w32 emacs under wine.  A 
page-scroll up and down has thus far largely dispelled them when they 
occur. Some of the colors seem way off (my yellow-on-black fringes come 
out cyan-on-white).

3. toolbar/scrollbar.

The toolbar doesn't work, and the scrollbars only work sometimes, but 
it's not like I use either much.  The menu bar is fine.

4. choice of fonts and gui backend:

Wrong choice of font can render it difficult to use too - its metric 
computation isn't always right I guess, with one font I ended up with 
something like a 3-pixel high minibuffer.

Remember that GNUstep also has multiple graphics backends with different 
font behaviours, I used the cairo backend which reputedly has slightly 
poorer font rendering than art, but OTOH worked.

Remember to try with "openapp Emacs -Q", because your ~/.emacs (which 
gnustep will see) might be setting a problematic font, mine initially was.

It sometimes makes "interesting" font choices itself, I don't know where 
it found some sort of comic-sans alike on my system but it did.

5. Non-latin chars.

Doesn't seem to do well here at all.

6. keyboard modifier mapping

One thing that helped a lot was to reconfigure the gnustep-level 
keyboard modifier mapping so that the keys in ns emacs ultimately wound 
up similar to x11 emacs with its out-of-box defaults.

http://www.gnustep.org/resources/documentation/Developer/Back/General/DefaultsSummary.html

I used (with Help on Super_R because I wanted to see what it did, turns 
out ns emacs treats it as Hyper...):

NSGlobalDomain GSFirstControlKey Control_L
NSGlobalDomain GSSecondControlKey Control_R

NSGlobalDomain GSFirstAlternateKey Alt_L
NSGlobalDomain GSSecondAlternateKey NoSymbol

NSGlobalDomain GSFirstCommandKey Super_L
NSGlobalDomain GSSecondCommandKey NoSymbol

NSGlobalDomain GSFirstHelpKey Help
NSGlobalDomain GSSecondHelpKey Super_R


7. [new since I passed this to Stefan]. No timers/idle???

Timers and idle stuff mostly not running, I think. Emacs processes stuff 
when there's input i.e. wiggle the mouse to make stuff happen ?!

Re: Emacs with Cocoa/GNUstep

[Germán Arias]

On mié, 2011-04-27 at 20:13 +0100, David De La Harpe Golden wrote:
> 
> Probably not a whole lot.... FWIW, I built it successfully several times 
> a few months back, and though it was not really especially useful once 
> built, it was not completely unusable either, it was working enough for 
> me to debug some pasteboard interaction issues without having access to 
> macosx.
> 
> I'm not personally up to doing much it about it at the moment, so here 
> are some notes (Stefan: mostly same ones I passed offlist to you at the 
> time) as a braindump, some of the below is probably obsolete:

Thanks, Stefan sent me a copy of that email some days ago.

> 
> First and foremost, DO NOT use any debian packages of GNUstep at time of 
> writing, they're obsolete and hopelessly buggy, especially on 64-bit.  I 
> wasted basically a weekend's worth of emacs-time that way, I switched to 
> an svn checkout of GNUstep installed to /usr/local/GNUstep and was up 
> and running fairly rapidly.
> 
> GNUstep itself is fairly painless to build from source (though n.b. I am 
> used to underdocumented build scripts from hell from work, my perception 
> may be skewed), the most tricky bit was getting building the 
> gnustep-back backend right, so that I could use cairo.
> 
> http://wiki.gnustep.org/index.php/GNUstep_Installation_Process
> 

I'm a contributor to GNUstep project since 2 years ago. So I really
don't have problems to install it from source. In fact I have more or
less (with some additions from SVN) the latest stable release of
GNUstep. And many apps that run perfectly.


> The other build-time issue is that emacs "./configure --with-ns" doesn't 
> seem to discover the ObjC headers properly in the gnustep case - 
> specifying CPPFLAGS "fixed" that, but I suspect TRT would be to have 
> emacs' configure use "gnustep-config --objc-flags" to discover the flags 
> (or maybe just $GNUSTEP_MAKEFILES and the right part of gnustep's 
> Makefiles, but gnustep-config seems interesting 'cos it apparently works 
> much like other blah-config type tools).

./configure --with-ns works fine form me (with stable release of
GNUstep).

> 
> I wouldn't say it is _totally_ unusable, but neither is it at all 
> pleasant .  There are probably plenty more smaller issues that I'm 
> presently not noticing given the big ones:

I tried Emacs.app like a year ago. I don't remember exactly what
version. I tested it and worked (with problems and sometimes crashed).
But currently, when I try to launch it I get the error:

german@german-desktop:~/Instalados/emacs$ openapp ./nextstep/Emacs.app/
/home/german/Instalados/emacs/nextstep/Emacs.app/Emacs: Uncaught
exception NSInternalInconsistencyException, reason: NSApp's run called
recursively


> 
> 0. Make sure the relevant gnustep daemons are running.
> 
> I mean the gpbs / gdnc /gdomap  daemons. Not really an emacs problem, 
> and documented in the gnustep docs, just something to be aware of.
> 
> gpbs is particularly important, as it's the GNUstep PasteBoard Server, 
> and without it, clipboard/primary/secondary just ain't gonna work.
> 
> http://gnustep.made-it.com/BuildGuide/index.html#GNUSTEP.SERVICES
> 
> 1. frame resizing.
> 
> The single most major annoyance (or at least tied with repaint) is that 
> it doesn't handle being resized by the window manager, you have to 
> (set-frame-width/height) from within emacs for it to work.  I guess 
> macosx handles resizing differently.
> 
> 2. repaint
> 
> There are nasty repaint issues sometimes too upon partial scrolling, a 
> bit like the ones you used to see on w32 emacs under wine.  A 
> page-scroll up and down has thus far largely dispelled them when they 
> occur. Some of the colors seem way off (my yellow-on-black fringes come 
> out cyan-on-white).
> 
> 3. toolbar/scrollbar.
> 
> The toolbar doesn't work, and the scrollbars only work sometimes, but 
> it's not like I use either much.  The menu bar is fine.
> 
> 4. choice of fonts and gui backend:
> 
> Wrong choice of font can render it difficult to use too - its metric 
> computation isn't always right I guess, with one font I ended up with 
> something like a 3-pixel high minibuffer.
> 
> Remember that GNUstep also has multiple graphics backends with different 
> font behaviours, I used the cairo backend which reputedly has slightly 
> poorer font rendering than art, but OTOH worked.
> 
> Remember to try with "openapp Emacs -Q", because your ~/.emacs (which 
> gnustep will see) might be setting a problematic font, mine initially was.
> 
> It sometimes makes "interesting" font choices itself, I don't know where 
> it found some sort of comic-sans alike on my system but it did.
> 
> 5. Non-latin chars.
> 
> Doesn't seem to do well here at all.
> 
> 6. keyboard modifier mapping
> 
> One thing that helped a lot was to reconfigure the gnustep-level 
> keyboard modifier mapping so that the keys in ns emacs ultimately wound 
> up similar to x11 emacs with its out-of-box defaults.
> 
> http://www.gnustep.org/resources/documentation/Developer/Back/General/DefaultsSummary.html
> 
> I used (with Help on Super_R because I wanted to see what it did, turns 
> out ns emacs treats it as Hyper...):
> 
> NSGlobalDomain GSFirstControlKey Control_L
> NSGlobalDomain GSSecondControlKey Control_R
> 
> NSGlobalDomain GSFirstAlternateKey Alt_L
> NSGlobalDomain GSSecondAlternateKey NoSymbol
> 
> NSGlobalDomain GSFirstCommandKey Super_L
> NSGlobalDomain GSSecondCommandKey NoSymbol
> 
> NSGlobalDomain GSFirstHelpKey Help
> NSGlobalDomain GSSecondHelpKey Super_R
> 
> 
> 7. [new since I passed this to Stefan]. No timers/idle???
> 
> Timers and idle stuff mostly not running, I think. Emacs processes stuff 
> when there's input i.e. wiggle the mouse to make stuff happen ?!
> 

Re: Emacs with Cocoa/GNUstep

[Stefan Monnier]

> I tried Emacs.app like a year ago. I don't remember exactly what
> version. I tested it and worked (with problems and sometimes crashed).
> But currently, when I try to launch it I get the error:

The problem with the GNUstep version of Emacs is that it has never
worked satisfactorily, not noone uses it, so we never learn about
regressions, such as ones introduced by changes for Mac OS X using
features that don't work quite the same under GNUstep.
Any help in getting the GNUstep version to a usable state would be very
welcome, and subsequently to keep it usable as well, of course.


        Stefan