Re: Ada's Slide To Oblivion ...

[David Brown <cla@davidb.org>]

Ian Wild <ian@cfmu.eurocontrol.be> wrote:
> David Brown wrote:

>> Legality aside, I have yet to find a compiler that doesn't accept this.
> 
> I'll agree that most compilers will accept it, but it's not
> C and quite often doesn't do what you're expecting.

The "quite often" part is really going to depend on what you're doing.
Most processors that are used for general purposes these days have the
same memory spaces for code and data.  Every processor that I've seen
posted to this list certainly does.

A good counter example is the Atmel AVR Risc processor.  This is a
microcontroller where instructions are fetched from flash memory, and
data comes from SRAM.  There are separate load instructions to retrieve
data from the two types of memory.  This is an example where data to
function pointer conversions in C aren't going to work.

On most architectures, C compilers will implement the pointer
conversions in the obvious way.  If they didn't, their compilers would
probably be unable to compile operating systems and system libraries.

BTW, I believe that GNAT assumes the address spaces are the same to
construct trampolines for nested procedure pointers.  gcc for C also
does if you nest functions (a gcc extension).

Yes, using this feature is unportable, but hopefully someone
constructing opcodes, writing them in memory, and trying to execute them
isn't expecting it to be portable.

There are a lot of other issues you have to worry about when you do
this, such as cache coherency, since many processors have separate I and
D caches.

David Brown