[xemacs-hg @ 2001-06-08 12:21:09 by ben] ------ gc-in-window-procedure fixes ------ alloc.c: Create "post-gc actions", to avoid those dreaded "GC during window procedure" problems. event-msw.c: Abort, clean and simple, when GC in window procedure. We want to flush these puppies out. glyphs-msw.c: Use a post-gc action when destroying subwindows. lisp.h: Declare register_post_gc_action(). scrollbar-msw.c: Use a post-gc action when unshowing scrollbar windows, if in gc. redisplay.c: Add comment about the utter evilness of what's going down here. ------ cygwin setitimer fixes ------ Makefile.in.in: Compile profile.c only when HAVE_SETITIMER. nt.c: Style fixes. nt.c: Move setitimer() emulation to win32.c, because Cygwin needs it too. profile.c: Make sure we don't compile if no setitimer(). Use qxe_setitimer() instead of just plain setitimer(). signal.c: Define qxe_setitimer() as an encapsulation around setitimer() -- call setitimer() directly unless Cygwin or MS Win, in which case we use our simulated version in win32.c. systime.h: Prototype mswindows_setitimer() and qxe_setitimer(). Long comment about "qxe" and the policy regarding encapsulation. win32.c: Move setitimer() emulation here, so Cygwin can use it. Rename a couple of functions and variables to be longer and more descriptive. In setitimer_helper_proc(), send the signal using either mswindows_raise() or (on Cygwin) kill(). If for some reason we are still getting lockups, we'll change the kill() to directly invoke the signal handlers. ------ windows shell fixes ------ callproc.c, ntproc.c: Comments about how these two files must die. callproc.c: On MS Windows, init shell-file-name from SHELL, then COMSPEC, not just COMSPEC. (more correct and closer to FSF.) Don't force a value for SHELL into the environment. (Comments added to explain why not.) nt.c: Don't shove a fabricated SHELL into the environment. See above. ------ misc fixes ------ glyphs-shared.c: Style correction. xemacs-faq.texi: Merge in the rest of Hrvoje's Windows FAQ. Redo section 7 to update current reality and add condensed versions of new changes for 21.1 and 21.4. (Not quite done for 21.4.) Lots more Windows updates. process.el: Need to quote a null argument, too. From Dan Holmsand. startup.el: startup.el: Call MS Windows init function. win32-native.el: Correct comments at top. Correctly handle passing arguments to Cygwin programs and to bash. Fix quoting of zero-length arguments (from Dan Holmsand). Set shell-command-switch based on shell-file-name, which in turn comes from env var SHELL.

--- a/src/callproc.c	Thu Jun 07 06:37:25 2001 +0000
+++ b/src/callproc.c	Fri Jun 08 12:21:27 2001 +0000
@@ -1,4 +1,4 @@
-/* Synchronous subprocess invocation for XEmacs.
+/* Old synchronous subprocess invocation for XEmacs.
    Copyright (C) 1985, 86, 87, 88, 93, 94, 95 Free Software Foundation, Inc.
 
 This file is part of XEmacs.
@@ -21,6 +21,16 @@
 /* Synched up with: Mule 2.0, FSF 19.30. */
 /* Partly sync'ed with 19.36.4 */
 
+
+/* #### This ENTIRE file is only used in batch mode.
+
+   We only need two things to get rid of both this and ntproc.c:
+
+   -- my `stderr-proc' ws, which adds support for a separate stderr
+      in asynch. subprocesses. (it's a feature in `old-call-process-internal'.)
+   -- a noninteractive event loop that supports processes.
+*/
+
 #include <config.h>
 #include "lisp.h"
 
@@ -879,13 +889,40 @@
   {
     /* Initialize shell-file-name from environment variables or best guess. */
 #ifdef WIN32_NATIVE
-    const char *shell = egetenv ("COMSPEC");
-    if (!shell) shell = "\\WINNT\\system32\\cmd.exe";
+    const char *shell = egetenv ("SHELL");
+    if (!shell) shell = egetenv ("COMSPEC");
+    /* Should never happen! */
+    if (!shell) shell = (GetVersion () & 0x80000000 ? "command" : "cmd");
 #else /* not WIN32_NATIVE */
     const char *shell = egetenv ("SHELL");
     if (!shell) shell = "/bin/sh";
 #endif
 
+#if 0 /* defined (WIN32_NATIVE) */
+    /* BAD BAD BAD.  We do not wanting to be passing an XEmacs-created
+       SHELL var down to some inferior Cygwin process, which might get
+       screwed up.
+	 
+       There are a few broken apps (eterm/term.el, eterm/tshell.el,
+       os-utils/terminal.el, texinfo/tex-mode.el) where this will
+       cause problems.  Those broken apps don't look at
+       shell-file-name, instead just at explicit-shell-file-name,
+       ESHELL and SHELL.  They are apparently attempting to borrow
+       what `M-x shell' uses, but that latter also looks at
+       shell-file-name.  What we want is for all of these apps to look
+       at shell-file-name, so that the user can change the value of
+       shell-file-name and everything will work out hunky-dorey.
+       */
+    
+    if (!egetenv ("SHELL"))
+      {
+	CBufbyte *faux_var = alloca_array (CBufbyte, 7 + strlen (shell));
+	sprintf (faux_var, "SHELL=%s", shell);
+	Vprocess_environment = Fcons (build_string (faux_var),
+				      Vprocess_environment);
+      }
+#endif /* 0 */
+
     Vshell_file_name = build_string (shell);
   }
 }