Subj : segments
To   : ALL
From : GENE KWIECINSKI
Date : Sun Dec 28 2003 01:51 am

Seems like a simple problem, but might be a long description of it,
*hopefully* a simple fix.  (Using plain vanilla MASM.)

In short, I want to use multiple segments not to *have* different
segments, /per se/, ie, not to be able to have >64k of anything, either
singly or combined, but to just group together sections of
code/data/etc. by *function*.  Everything's a tiny-model .com file.

Am trying to reorganise code at the *source* level, though, so that
instead of a random mix of code, constant data, and volatile data
scattered throughout a .com, I want to declare each in its own segment.
So rather'n having in source file 1 (eg, string ops) code, constants,
data, then in source file 2 (eg, wrapper functions for DOS/BIOS calls)
more code, more constants, more data, etc., I want it to assemble so
that the final .com file will be code, code, code, constants, constants,
constants, and data, data, data.  I tried declaring

        code    segment byte public
        code    ends

        bss     segment byte public
        bss     ends

        data    segment byte public
        data    ends

        tiny    group   code,bss,data

        code    segment
                assume  cs:tiny,ds:tiny,es:tiny
                ...
                <code>
                ...
        code    ends

        <various include files, each w/ segments defined>
        etc.

and that seems to work just fine, with one exception!  I print out
"messages", eg, a help screen, which is preordered as an array of
strings, then table of pointers, terminated with a null pointer, eg

        bss     segment

        h00     db      'line 1 of help msg',0
        h01     db      'line 2 of help msg',0
        h02     db      'line 3 of help msg',0

        htbl    dw      h00,h01,h02,0

        bss     ends

and print the message by feeding 'htbl' to a subroutine that groks this,
eg,

        code    segment

        lea     ax,hbtl
        push    ax
        call    _putmsg

        code    ends

Not a problem, as 'htbl' *will* be resolved by the assembler relative to
'code'/'tiny', in this case, say, 046DH.  Pointers 'h00', 'h01', and
'h02' *should* be resolved to 04xxH, but are left alone relative to
segment 'bss', so that values in the list 'htbl' are in the range 00xxH
(ie, counter starts at 0 at first 'bss' data), *not* 04xxH were they to
be resolved relative to the 'code'/'tiny' segment/group.

*If* any pointer in 'data' or 'bss' is referenced by something in the
'code' section, it's properly resolved to its real binary address as it
would appear in the .com file.  If left alone and only self-referenced
within 'data' or 'bss', it's left as its unresolved value relative only
to the beginning of the respective segment that it's declared in.

Changing everything to the same segment (which defeats the whole
purpose) fixes it.  I know of no way to force resolution to the final
value, like maybe

        htbl    dw      tiny:h00, tiny:h01, tiny:h02, 0

so that the values will *not* be 0000,0019,0032,0000, but in fact be
something like 046D,0486,049F,0000.  Did try defining a constant to be
added to each pointer (eg, "bss - code"), but the assembler would have
none of it.

I *could* put dummy instructions like 'lea ax,h00', 'lea ax,h01', and
'lea ax,h02', just to "touch" those values, but again, that defeats the
whole purpose.

I can't believe I'm the first/only one to run up against this problem.
*ANY* ideas??
---
 þ SLMR 2.0 #åñjw þ For a good time, call F0F0:F0F0.
 * Origin: FONiX Info Systems * Berkshire UK * www.fonix.org (2:252/171)

.