Re: Counterexamples in C programming and library documentation (was: [PATCH v3] NULL.3const: Add documentation for NULL)

[Alejandro Colomar]

Hi Branden,

On 8/3/22 17:58, G. Branden Robinson wrote:
> At 2022-08-03T01:54:18+0200, Alejandro Colomar wrote:
> > Could you show an example?  I'm curious.
>
> Sure.  Let me quote Andrew Koenig, _C Traps and Pitfalls_,
> Addison-Wesley, 1989.
>
> "I once talked to someone who was writing a C program to run stand-alone
> in a microprocessor.  When this machine was switched on, the hardware
> would call the subroutine whose address was stored in location zero.
>
> In order to simulate turning power on, we had to devise a C statement
> that would call this subroutine explicitly.  After some thought, we came
> up with the following.
>
>    (*(void(*)())0)();

That piece of code is beautiful.

> Expressions like these strike terror into the hearts of C programmers."
>
> I, too, initially viewed it with dismay.  But I stared at it for a
> while, and once I parsed it, I cackled with pleasure.[1]
>
> However, my pleasure would turn to anger if some jackanapes stuck a
> comment above it reading as follows.
>
>    /* You are not expected to understand this. */

:)

> This comment from the V6 Unix kernel is famous, more famous
> unfortunately than the fact that Dennis Ritchie himself repented of this
> comment, and explained himself beautifully.
>
> https://www.bell-labs.com/usr/dmr/www/odd.html
>
> Regards,
> Branden
>
> [1] Here, at least, the addressing mode of the resulting instruction
>      _is_ more or less visible in the C statement.  In this instance I
>      don't think "NULL" would make it clearer, and I would come down on
>      the McIlroy side of the argument.  But this circumstance has nearly
>      vanished from the domain of practical application.  Thanks largely
>      to C and the generations of programmers it has fascinated with its
>      charming and attractive footguns, null pointers have been chased
>      with such neurotic compulsion that it is common these days to
>      program the MMU to raise an invalid address fault on any access to
>      the first page of memory.  Although I guess as recently as 2009,
>      Linux didn't do this.
>
>        https://lwn.net/Articles/342330/
>
>      On 64-bit architectures where the backing of the address space by
>      actual storage is once again sparse, I don't know of a reason to use
>      the page at the bottom of memory for storage.

I think this is one of those cases where the all-0s bit pattern doesn't 
necessarily correspond to a NULL pointer.  That's using a pointer to 
something that exists and can be dereferenced, and happens to be at 
address 0.  So, I agree with you that using the address 0 in this case 
makes much more sense than dereferencing the NULL pointer, which is 
supposed to be an invalid pointer, one that doesn't point to anything.

Cheers,

Alex


--
Alejandro Colomar
<http://www.alejandro-colomar.es/>

OpenPGP_signature
Description: OpenPGP digital signature