Notes on writing portable C/Unix programs


Experienced C programmers probably know all this stuff but I've tripped
on these things myself.  Many of them are also no longer valid since
the establishment of the ANSI C and POSIX standards.  For what it's worth...



Include files and defines
=========================
Use #include <string.h>, not #include <strings.h>

If you need SEEK_SET, SEEK_CUR, etc don't depend on <stdio.h> to define them.

If you need M_PI, don't depend on <math.h> to define it.

The prototype for memcpy() may be in <string.h> or <stdlib.h> and even then
	varies from system to system.

The prototypes for malloc(), free(), etc may be in <stdlib.h> or <memory.h>

The O_RDONLY and related symbols (man open) can be defined in a variety of
	places including <fcntl.h>, <sys/file.h>, <sys/fcntl.h>, etc.

Include <stdio.h> to define NULL.



Math
====
Don't use sinf(), cosf(), etc. they're not always available.

Be aware that pow(0,0)==1 on some systems but causes a domain error on others.

Be aware that divide by zero isn't always caught on modern RISC systems.
Look for a man page on handling floating point exceptions.

The prototypes for atoi() and atof() may be in <math.h> or <stdlib.h>



The C language
==============
Under K&R C be aware that float function parameters are always passed
as doubles.  For example, this is bad:

	void foo( x )
	float x;
	{
	   float *y;
	   y = &x;              /* x is really a double! */
	   printf("%f\n", *y);
	}

Similarly, be aware that char parameters are always passed as ints.

Empty initializers are bad:  static int x[] = { };

Not all C compilers handle automatic aggregate initializations.  For example,
	void foo()
	{
	   float vector[3] = { 1.0, 2.0, 3.0 };
	   ...
	}
vector could be defined as static:
	void foo()
	{
	   static float vector[3] = { 1.0, 2.0, 3.0 };
	   ...
	}

realloc( NULL, n ) doesn't always work as by ANSI on some Suns (note the NULL).



Architecture dependencies
=========================
Be aware that "long int" is not always the same size as "int".

Be aware that sizeof(any pointer) is not four bytes on 64-bit systems.



System-specific pitfalls
========================
strdup() is not available on DEC Ultrix systems and not in ANSI C spec.

The prototype for strdup() is in stdlib.h on DEC alpha systems, not string.h

On HP's you usually need to define -D_HPUX_SOURCE in your CFLAGS



X11, GL and OpenGL
==================
If need to include <X11/Xlib.h> AND <gl/gl.h>, include them in that order.

Be sure to use -I$(SOMETHING) in your makefile to find X11 include files if
	not in /usr/include/X11



ranlib
======
Some Unix systems need to have ranlib applied to library (.a) files,
others don't.  Here's what I've found:

	configuration		need ranlib?
	----------------	------------
	IBM RS/6000, AIX	no
	Amiga, AMIX		no
	Data General, DGUX	no
	FreeBSD			yes
	generic GCC		yes
	HP, HPUX		no
	SGI, IRIX		no
	PC, Linux		yes
	MacIntosh, MachTen	yes
	PC, NetBSD		yes
	DEC, OSF/1		no
	PC, SCO unix		no
	SunOS 4.x		yes
	SunOS 5.x 		no
	DEC, Ultrix		no
	PC, Unixware		no
	Stardent, Vistra	no
	Stardent, Stellix	no


Miscellaneous
=============

Don't name a function "index" as there's sometime a string function with
that name.


Avoid the use of bcopy and bzero, and use memcpy and memset instead.
Notice the arg order from memcpy is the reverse of bcopy.  (thanks PH)


On IRIX Use the -signed flag on your cc lines to force char declarations
to be signed, or make sure that all char declarations are preceded by
"signed" or"unsigned".  Most comilers on Linux and Windows/NT default
chars to signed char.  (thanks PH)



See also Portability.txt in Mark Kilgard's GLUT distribution.