Path: utzoo!attcan!uunet!midway!ncar!zaphod.mps.ohio-state.edu!usc!rutgers!att!watmath!maytag!watdragon!lotus!sekoppenhoef
From: sekoppenhoef@lotus.uwaterloo.ca
Newsgroups: comp.sys.amiga
Subject: Manx Bug (3.6 + 5.0a + 5.0b)
Message-ID: <1990Jul6.022133.29265@watdragon.waterloo.edu>
Date: 6 Jul 90 02:21:33 GMT
Sender: daemon@watdragon.waterloo.edu (Owner of Many System Processes)
Distribution: comp
Organization: University of Waterloo
Lines: 59

I have found a case where the MANX C compiler (any version) generates bad code.
I discovered this while porting GNU awk 2.11 from MSDOS to the AMIGA (it works
now).

They do some error checking?? using a macro in similar form to:

#define MAC(a,b) { \
     /* some processing */    \
     if (b) {   \
       fprintf (stderr, b);   \
	} \
   }


In some(all) cases, b is of the form "error message here"  ie it is used
in calls such as MAC(code, "can't open file")

For the resulting code

	if ("can't open file")
	...

MANX generates a    tst.l  #.12   (12 arbitrary)  which is an instruction
that the assembler doesn't recognize  (Opcode operands did not match,
src=0, dest=400)   ie the immediate mode of this operand is not recognized.

Is this construct legal in C (I believe there is nothing wrong with it)?

Could somebody pass this onto the people at MANX so that they may correct it?

Also, SDB 5.0b does NOT like long source files.  I had to debug the source
file that was generated by bison (~3000 lines) and when SDB was displaying
the file it looks as though it did the following:

	open source file
	read in n lines of text
	display source line
	close file

	open source file
	read in n+1 lines of text
	display source line
	close file

	...  for each line of text in the source window.

This is not acceptable!!!  When debugging around line 2700, it took about
10 seconds to refresh the window, and this had to be done every time the
source window scrolled.


It was also slower than 5.0a on skipping over library routines such as
open(), tmpnam().  (These took approx 15 seconds one time)

Enough of these gripes.  Hopefully MANX will see it in there infinite wisdom
to at least fix the way SDB displays the source code.

-- Scotte
(sekoppenhoef@lotus.waterloo.edu)   A friend's account