archie/tk3.6/porting.notes

This file contains a collection of notes that various people have
provided about porting Tk to various machines and operating systems.
I don't have personal access to any of these machines, so I make
no guarantees that the notes are correct, complete, or up-to-date.
I'd be happy to receive corrections or updates.

---------------------------------------------
DEC Alphas:
---------------------------------------------

1. There appears to be a compiler/library bug that prevents tkTrig.c
from compiling unless you turn off optimization (remove the -O compiler
switch).  The problem appears to have been fixed in the 1.3-4 version
of the compiler.

---------------------------------------------
HP-UX systems:
---------------------------------------------

1. Configuration:
	    HP-UX Release 7.05 on a series 300 (68k) machine.
	    The native cc has been used for production.
	    X11r4 libraries and include files were taken from
	    internet archives, where as the server came with HPUX release 7.05.
    
    Problems:
	    Symbol table space for cc had to be increased with: -Wc,-Ns3000
	    tkBind.c did not compile under -O:
		    C1 internal error in "GetField": Set Error Detected
		    *** Error code 1
	    tkBind.c did compile without optimization (no -O).

2. Note: if you have trouble getting xauth-style access control to work
(and you'll need xauth if you want to use "send"), be sure to uncomment
the line

# Vuelogin*authorize:         True

in the file /usr/vue/config/Xconfig, so that the server starts up with
authorization enabled.  Also, you may have to reboot the machine in
order to force the server to restart.

---------------------------------------------
SCO Unix:
---------------------------------------------

Getting Tk to run under SCO Unix:

Add a "#undef select" to tkEvent.c, and remove the reference to TK_EXCEPTION
around line 460 of main.c.

Tk uses its own scheme for allocating the border colors for its 3D widgets,
which causes problems when running TK on a system with "PseudoColor"
display class, and a 16-cell colormap.

If you can't go to eight bitplanes, you can instead start the server with a
"-static" (Xsco) or "-analog" (Xsight) option, making the display class 
become "StaticColor".  This makes the entire colormap read-only, and it will
return the color that most closely maps to the desired color as possible.

---------------------------------------------
Silicon Graphics systems:
---------------------------------------------

1. Change the CC variable in the Makefile to:

CC =		cc -xansi -D__STDC__ -signed

2. Change the LIBS variable in the Makefile to use the X11 shared library
("-lX11_s" instead of "-lX11").  Also, if you want to access remote servers
you may need to link with "-lsun", e.g. "-lX11_s -lsun".

3. Under some versions of IRIX (e.g. 4.0.1) you have to turn off
   optimization (e.g.  change "-O" in CFLAGS to "-O0" or remove it
   entirely) because of faulty code generation.  If the Tcl or Tk test
   suites fail, turn off optimization.

4. Add a "-lsun" switch just before "-lm" in the LIBS definition.

---------------------------------------------
IBM RS/6000's:
---------------------------------------------
1. To allow ALT- sequences to work on the RS-6000, the following
line should be changed in tkBind.c:

    OLD LINE:
     {"Alt",	Mod2Mask, 	0},
    NEW LINE:
     {"Alt",	Mod1Mask, 	0},

---------------------------------------------
AT&T SVR4:
---------------------------------------------

1. The first major hurdle is that SVR4's select() subtly differs
from BSD select.  This impacts Tk in two ways, some of the Xlib calls
make use of select() and are inherently broken and Tk itself makes
extensive use of select().  The first problem can't be fixed without
rebuilding one's Xlib, but can be avoided.  I intend to submit part
of my work the XFree86 guys so that the next version of XFree86 for
SVR4 will not be broken.  Until then, it is necessary to comment out
this section of code from Tk_DoOneEvent() (which is near line 1227):

#if !defined(SVR4)
                    void (*oldHandler)();

                    oldHandler = (void (*)()) signal(SIGPIPE, SIG_IGN);
                    XNoOp(display);
                    XFlush(display);
                    (void) signal(SIGPIPE, oldHandler);
#endif /* SVR4 */

if you don't comment it out, some scripts cause wish to go into
an infinite loop of sending no-ops to the X server.

2. As for fixing Tk's calls to select(), I've taken the simple
approach of writing a wrapper for select and then using #define to
replace all calls to select with the wrapper.  I chose tkConfig.h
to load the wrapper.  So at the very end of tkConfig.h, it now looks
like:

#if defined(SVR4)
#  include "BSDselect.h"
#endif 

#endif /* _TKCONFIG */

The file BSDselect.h looks like this:

#include <sys/types.h>
#include <sys/time.h>
#include <sys/select.h>

/*  This is a fix for the difference between BSD's select() and
 *  SVR4's select().  SVR4's select() can never return a value larger
 *  than the total number of file descriptors being checked.  So, if
 *  you select for read and write on one file descriptor, and both
 *  are true, SVR4 select() will only return 1.  BSD select in the
 *  same situation will return 2.
 *
 *	Additionally, BSD select() on timing out, will zero the masks,
 *	while SVR4 does not.  This is fixed here as well.
 *
 *	Set your tabstops to 4 characters to have this code nicely formatted.
 *
 *	Jerry Whelan, guru@bradley.edu, June 12th, 1993
 */


int
BSDselect(nfds, readfds, writefds, exceptfds, timeout)
int nfds;
fd_set *readfds, *writefds, *exceptfds;
struct timeval *timeout;
{
	int		rval,
			i;

	rval = select(nfds, readfds, writefds, exceptfds, timeout);

	switch(rval) {
		case -1:	return(rval);
					break;

		case 0:		if(readfds != NULL)
						FD_ZERO(readfds);
					if(writefds != NULL)
						FD_ZERO(writefds);
					if(exceptfds != NULL)
						FD_ZERO(exceptfds);

					return(rval);
					break;

		default:	for(i=0, rval=0; i < nfds; i++) {
						if((readfds != NULL) && FD_ISSET
(i, readfds)) rval++;
						if((writefds != NULL) && FD_ISSE
T(i, writefds)) rval++;
						if((writefds != NULL) && FD_ISSE
T(i, exceptfds)) rval++;
					}
					return(rval);
		}
/* Should never get here */
}

---------------------------------------------
CDC 4680MP, EP/IX 1.4.3:
---------------------------------------------

The installation was done in the System V environment (-systype sysv)
with the BSD extensions available (-I/usr/include/bsd and -lbsd).  It was
built with the 2.20 level C compiler.  The 2.11 level can be used, but
it is better to match what TCL is built with, which must be 2.20 or
higher (see the porting notes with TCL for the details).

To make the configure script find the BSD extensions, I set environment
variable DEFS to "-I/usr/include/bsd" and LIBS to "-lbsd" before
running it.  I would have also set CC to "cc2.20", but that compiler
driver has a bug that loader errors (e.g. not finding a library routine,
which the script uses to tell what is available) do not cause an error
status to be returned to the shell.

After running configure, I changed the CC definition line in Makefile
from:
	CC=cc
to
	CC=cc2.20
to match the TCL build.  Skip this if the default compiler is already 2.20
(or later).

-------------------------------------------------
Pyramid, OSx 5.1a (UCB universe, GCC installed):
-------------------------------------------------

Instead of typing "./configure" to configure, type

    DEFS="-I/usr/include/X11/attinc" ./configure

to sh to do the configuration.

-------------------------------------------------
NextSTEP 3.1:
-------------------------------------------------

1. Run configure with predefined CPP:
	CPP='cc -E' ./configure
	
2. Edit Makefile: 
  -add '-m' to MATH_LIBS
	MATH_LIBS = -m -lm