Semi constant expressions

[Doug Gwyn]

In article <242 at ssp1.idca.tds.philips.nl> dolf at idca.tds.PHILIPS.nl (Dolf Grunbauer) writes:
>In article <10885 at smoke.BRL.MIL> gwyn at brl.arpa (Doug Gwyn) writes:
>>In article <1237 at gmdzi.UUCP> wittig at gmdzi.UUCP (Georg Wittig) writes:
>>>	0 * x			(x non-constant)
>>>Does (p)ANSI C say anything about which code should be generated for such
>>>expressions?
>>This falls naturally out of the specification.  The operands ARE evaluated,
>>so if `x' has side-effects they do occur.  If evaluation would produce what
>>the Standard labels as "undefined behavior", then the implementation can do
>>anything it likes including taking short cuts, but in cases where the
>>behavior is well defined the only short cuts permitted are those that
>>produce the same results as doing it the long way.
>How can you tell what will happen when 'doing it the long way' ?

I thought this was clear from what I said and from the context.  In the
case
	0 * x
if evaluation of the expression `x' would have observable side effects
then they MUST occur for a correct implementation of C.  In such a case,
an implementation that replaced the expression "0 * x" with the constant
"0" as an "optimization" would not be standard conforming.

The run-time behavior of a C program is defined in terms of the operation
of a "virtual machine" (with some deliberately fuzzy areas in its
specification, so it's really a class of virtual machines) and input/output
operations (which are usually what makes the program's behavior observable).
A correct implementation must produce observable results identical to those
that would be produced by the virtual machine (actually, by some member of
the class of virtual machines).