Re: I don't get how the computer arrives at 2^31

Francis Glassborow <francis@xxxxxxxxxxxxxxxxxxxx> writes:

> In article <43b523f7$0$11073$e4fe514c@xxxxxxxxxxxxxx>, Skarmander
> <invalid@xxxxxxxxxxxxxx> writes
> >To be precise, an implementation could change representations as it
> >sees fit, as long as it makes sure that the view to the program is
> >consistent (e.g. when the value is read through an unsigned char*; it
> >could normalize the representation whenever this happens). I doubt this is
> >what the standard implies, but it is legal nevertheless (per the "as if"
> >clause), and one can come up with implementations that could make use
> >of it.
>
>
> IIRC, we (WG14) decided in Sydney that where a value has multiple
> representations the implementation is entitled to arbitrarily change
> from one representation to another. For example, if the layout of a type
> includes padding, there is no requirement that that padding be the same
> bits on successive reads using a pointer to unsigned char.

If so that has grave implications for the language model.
Accessing objects through unsigned char * would basically
guarantee that a program could not be strictly conforming,
for example.

Do you have a reference for any discussion that went on
around this? I did a search through wg14 documents at
the open-std.org website and didn't find anything.


> I think we went further and declared that an indeterminate vale was
> equivalent to allowing all bit patterns to be legitimate
> representations. This is important for some types of optimisation:
>
> e.g:
>
> int main(){
> int memory[1000000];
> /* processes */
> }
>
> As the array has not been initialised the implementation is free to
> choose to not save the actual bit patterns when paging out the memory,
> because any pattern of bits for the million units of int storage is
> equally valid.

Amazing.

I'd like to offer the opinion that the interests of compiler writers
are over-represented in making these kinds of decisions.


(In any case, thank you Francis for the informative report.)
.

Relevant Pages

  • Re: I dont get how the computer arrives at 2^31
    ... an implementation could change representations as it ... int memory; ... choose to not save the actual bit patterns when paging out the memory, ...
    (comp.std.c)
  • Re: I dont get how the computer arrives at 2^31
    ... an implementation could change representations as it ... >> representations the implementation is entitled to arbitrarily change ... >> includes padding, there is no requirement that that padding be the same ... >> bits on successive reads using a pointer to unsigned char. ...
    (comp.std.c)
  • Re: Perceptual symbol systems
    ... Think of representations as patterns of activation in groups ... derived from the hearing of that phrase are a "representation" ... of some phonological qualities. ...
    (sci.cognitive)