🚀 go-pugleaf

I've discovered this bug in Lattice C, version 4.01 and reported it to
the Lattice customer support folks.  It seems that if you subscript an
array (or pointer, same thing) with an unsigned short, the subscript is
still sign extended.  This makes it impossible to have a subscript greater
than 32K using unsigned shorts.

The Lattice guy reproduced the problem, and thought that there would be
another patch to version 4 of the compiler before the next major release
(which is 5.0, I think along with the symbolic debugger).

Anyway, simple test program:

	int *array;
	unsigned short us;

	func() {
		int i;
	
		i = array[us];
		return i;
	}

compile this with or without -b0, and look at the object file created
with OMD.  You'll see an EXT.L; bad news.

LATTICE OBJECT MODULE DISASSEMBLER V2.00

Amiga Object File Loader V1.00
68000 Instruction Set

EXTERNAL DEFINITIONS

_foo 0000-00    _array 0000-02    _us 0004-02

SECTION 00 "test.o" 00000020 BYTES
int *array;
unsigned short us;

foo() {
0000 4E55FFFC                   LINK      A5,FFFC
0004 BFEC 0000-XX.2             CMPA.L    __base(A4),A7
0008 6500                       
000A  0000-XX.1                 
	int i;

	i = array[us];
000C 302C 0004-02.2             MOVE.W    02.00000004(A4),D0
0010 48C0                       EXT.L     D0		<=== WRONG! ===
0012 E580                       ASL.L     #2,D0
0014 206C 0000-02.2             MOVEA.L   02.00000000(A4),A0
0018 D1C0                       ADDA.L    D0,A0
001A 2010                       MOVE.L    (A0),D0
	return i;
001C 4E5D                       UNLK      A5
001E 4E75                       RTS

SECTION 01 "__MERGED" 00000000 BYTES

SECTION 02 "__MERGED" 00000008 BYTES


Extra bonus bug:

	When compiling a module with -b0 (absolute reference, rather
than off of A4), the compiler will not allow you to have more than
64K of data declared in the BSS section (at least).  It seems logical
to prevent you from doing with if you don't specify -b0, since it would
be difficult to reference the data, but doesn't make sense otherwise.



	This makes my port of TeX-in-C have to malloc() the various
arrays;  the first bug makes it difficult to create non-trivial
documents using LaTeX with a 32K memory size.  Once there problems are
worked out, I'll probably distribute my port, along with the dvi-to-
postscript translator for the amiga.  My port of TeX doesn't run as fast
as AmigaTeX does, nor does it have loop mode and all that good stuff, but
its free.  It also probably won't fit on a 512K machine either.

	I suppose that if I really begin to do serious TeX work, I'll
probably spend the $200 and buy AmigaTeX.  If the demo version on one
of the Fish disks is any indication, its is a wonderful product.

	Once I hear back from Lattice, I'll follow up with what they've
decided to do.



Louis A. Mamakos  WA3YMH    Internet: louie@TRANTOR.UMD.EDU
University of Maryland, Computer Science Center - Systems Programming