Re: [PATCH] Replace NULL symbol with 0

["Tomasz Kupczyk" <kupczyk tomasz gmail com>]

Dnia 01-08-2011 o 13:43:42 Dodji Seketeli <dodji seketeli org> napisał(a):

> Tomasz Kupczyk <kupczyk tomasz gmail com> a écrit:
>
> > Hello all,
>
> Hello Tomasz.  :-)
>
> > I'm new on the Nemiver mailing list,
>
> You are welcome!
>
> > I haven't written any code in Nemiver yet, so maybe I shouldn't
> > comment on the code here,
>
> I disagree.  You are free to comment here, no matter who you are.  I for
> one am always interested in technical matters related to Nemiver.  :-)

Good to hear that :)

> > wondering what is the idea behind changing all NULL's to 0's?
> >
> > As far as I know NULL equal to 0 is platform dependent, but on most
> > platforms probably it's indeed 0.
>
> The C++ specification says, in [conv.ptr]/1:
>
>     A null pointer constant is an integral constant expression (5.19)
>     prvalue of integer type that evaluates to zero
>
> So, the constant 0 is a perfectly valid platform independent null
> pointer constant in C++, unlike in C.

Ok, I've also checked, this explains a lot.

> When you use g++, NULL is a macro defined to __null, which is a special,
> g++ specific symbol that does the right thing, i.e, it evaluates to
> zero.
>
> Furthermore, NULL being a macro, if you want to see its integral
> representation in a debugger, you have to have to e.g, have some arcane
> compiler flag enabled, as just -g is not enough.  If you use zero, in
> the debugger, you know easily that it is zero.
>
> > Different thing is that IMO using NULL makes sometimes code easier to
> > read,
>
> I understand that conveying the idea that "this is a null pointer
> constant" is more maintainable (as in, "it tells a better story to the
> reader's mind") than just 0, when using a context that expects a
> pointer.  But NULL is not necessarily correct.  That why c++0x
> introduces the null pointer constant nullptr, of type nullptr_t.

When is NULL not necessarily correct? I suppose it's better to read
some rationale why nullptr is introduced :)

> For now, my take was that it's less confusing to separate NULL which
> originates from C and which is platform dependant, as you pointed out,
> from 0.  And use 0 instead.  When you are used it, it zero can also tell
> a nice story to the reader's mind, I believe.  :-)
>
> I am happy to let pieces of code that are compiled with a C compiler use
> NULL.
>
> > and for example if in the future compilers will start to enforce
> > semantically null pointers being really NULL then all 0's will need to
> > be turned into NULL's once again.
>
> I don't understand this.  nullptr of c++0x comes with an implicit
> conversion that can convert it to 0.  So using 0 will still work in
> c++0x.

I was thinking about type checking, for example when you write void* ptr =  
0,
0 is of integral type, and ptr is of void* type, but the conversion  
succeeds,
but when you write void* ptr = 10 you'll get an error (gcc gives a warning  
only):

error: invalid conversion from ‘int’ to ‘void*’

I thought that in future it could happen that assigning 0 to a pointer type
will be treated the same way as assinging 10, because there will be no  
special
case for 0.

But generally yeah, it's a digression, thank you for your answers :)

Best regards,
Tomasz Kupczyk