Sunday May 28 1995 09:11, Bill Grimsley wrote to Paul Edwards:

Intereresting that your original message didn't make it here. 

 PE>> Ok, it's taken me all day, but I found out what was stopping me
 PE>> using Watcom C for DOS (after Rowan told me how to make the
 PE>> assembler stuff work - THANKS ROWAN!!!).  The bug was in
 PE>> normal.asm, it has a "ret 4" instead of a
"ret". This is a bug
 PE>> that would have existed in all 3.x versions of MSGED, and I do not
 PE>> know why it hasn't affected anyone.  It only affected the dos
 PE>> version of course.  Anyway, I have no idea whether this is causing
 PE>> the Qld woes or not, but instead of just sending up the Borland
 PE>> executable with this bug squashed, I'm sending up the Watcom
 PE>> executable instead.  So that is two changes that I am sending, and
 PE>> hopefully one of them will fix the problem you are having, and
 PE>> subsequenly hopefully everyone else's.

    I have NORMAL.ASM in another window right now, and indeed it has
"ret 4". This means return [far], and discard 4 bytes (2
parameters) from the stack. Except no parameters were passed to it!

    This has the effect of "misaligning" the stack pointer (SP)
by +4 every time this function is called. Eventually SP will grow into code
or data space, or wrap at SS:FFFF and cause a stack fault (exception #12)
or possibly a GPF (#13).

    How often is _normalize called ?

    As I said in my last message... errors in ASM are very subtle and can
take a lot of work to track down. I should know. I'm writing an entire
bloody compiler.

---