🚀 go-pugleaf

  Or can we? Just because the message is gone, doesn't mean that a new
  UAE dialog isn't out there, waiting. 

I can't believe how badly misunderstood this 'No UAE' business has. Pray,
tell me, as a programmer, how in hey do you expect *any* OS to detect 
and correct an error in an application? What do you want Windows to do
when you dereference a Null pointer?

UAE's were terrible because of the following - an application error would
not be detected at the time of the error, but would manifest itself later
on at seemingly completely random times. So for example, Program A could
do something illegal (like delete a resource not belonging to it), things
would continue fine, and then later on, Program B, owner of resource, 
dies with UAE. Now, the poor programmer of Program B is trying to figure 
out what s/he did wrong. (This is probably a bad example, but you get the
point)

The GPF flags an error when it occurs, so that the fault is correctly 
identified to be with Program A.

The following program will die under every OS of the past, present, and
future: whether you call it UAE, GPF, segmentation fault, bus error, or
whatever. 

void main()
{
	int *pnCrash;

	pnCrash = 0;
	*pnCrash = 1;
}

The big improvement in Windows is not that UAE's don't occur (they *must*,
if there is an error in the user program), but that they are *detected*
correctly and at the time the error occurs.

Milan

In article <MMSHAH.92Apr14141355@im5.lcs.mit.edu> mmshah@im.lcs.mit.edu (Milan Shah) writes:
>
>The big improvement in Windows is not that UAE's don't occur (they *must*,
>if there is an error in the user program), but that they are *detected*
>correctly and at the time the error occurs.
>
>Milan
>
>


What I can't understand is why I'm now getting all sorts of new "UAE"s
from programs I've been using successfully under Win3.0.  Xtalk for
Windows is one I regularly get.  

-- 
Pierre von Kaenel	| Skidmore College	     | pvonk@scott.skidmore.edu
Math & CS Dept.		| Saratoga Springs, NY 12866 | (518)584-5000 Ext 2391

"Baseball is 90% mental, the other half is physical" - Yogi

/ hpopd:comp.windows.ms.programmer / pvonk@scott.skidmore.edu (Pierre VonKaenel) /  4:26 pm  Apr 15, 1992 /

>What I can't understand is why I'm now getting all sorts of new "UAE"s
>from programs I've been using successfully under Win3.0.  Xtalk for
>Windows is one I regularly get.  

It may be that Windows 3.1 is uncovering previously hidden defects.

For example, we discovered a defect in our code related to the freeing of
an explicitly loaded DLL. This defect didn't cause a UAE in Windows 3.0a,
but did in Windows 3.1. (We were calling a function from a DLL which then
explicitly freed the DLL; as we returned from the function to hyperspace
we UAE'd. It seems that freeing DLLs is "more synchronous" under 3.1 than
under 3.0a.)

In article <MMSHAH.92Apr14141355@im5.lcs.mit.edu>, mmshah@im.lcs.mit.edu (Milan Shah) writes:
>   Or can we? Just because the message is gone, doesn't mean that a new
>   UAE dialog isn't out there, waiting. 
> 
> I can't believe how badly misunderstood this 'No UAE' business has. Pray,
> tell me, as a programmer, how in hey do you expect *any* OS to detect 
> and correct an error in an application? What do you want Windows to do
> when you dereference a Null pointer?

 [More good examples and discussion of the UAE problem, but then he says:]

> The following program will die under every OS of the past, present, and
> future: whether you call it UAE, GPF, segmentation fault, bus error, or
> whatever. 
> 
> void main()
> {
> 	int *pnCrash;
> 
> 	pnCrash = 0;
> 	*pnCrash = 1;
> }

I can't resist the urge to make a pedantic comment (please forgive me) -
actually this program will very happily compile and run on the PDP-11
RSX system and a number of other systems.  On a lot of the PDP-11 systems
location 0 was the Directive Status Word, used to pass back status results
from system calls from the Executive to the user program.  This sequence
just sets the DSW to 1, a perfectly legal (if strange) operation for a
user program to do - in fact a 1 was the "successful completion" code for
most system calls.

But most protected-mode operating systems have deliberately unmapped
location 0 because it's so convenient for detecting the dereferencing
of a NULL pointer, so it's not a bad example, just that the quantifier
"every" is wrong (should be "most").

Apart from that minor slip (:-) your summary of this subject is quite
good.  For the most part it isn't Windows' fault that UAE's or whatever
exist, but it is (usually) Windows' fault when it allows the UAE in one
app to crash all the other apps on the system.

Bruce C. Wright