Re: unnecessary {de,}alloc wrappers?

[Ondrej Oprala]

On 04/01/2014 05:43 PM, Eric Blake wrote:
> On 04/01/2014 09:28 AM, Ondrej Oprala wrote:
> > Hi,
> >   Could someone please explain to me, why are xrealloc and xfree (or more
> > specifically - their parts) from lib/malloc/xmalloc.c necessary?
> >
> > e.g.
> > if (string)
> >      free(string);
> > Why is this needed? AFAIK the C standard specifies free(NULL) is
> > completely legal.
> Bruno Haible argued that the cost of a redundant if conditional is
> cheaper than the cost of a function call in profiling runs, at least in
> the portions of gnulib where he used that idiom.  Personally, I like
> getting rid of the redundant if (I think the extra CPU cycles spent are
> in the noise, until given a profile dump that proves otherwise).
Hmm, is that still up-to-date information?
Doing
$ git grep -B2 '\bfree\(.*;'
on glibc's and gnulib's upstream repo shows no occurences of code such 
as "if (foobar) free(foobar);".
Sure there are conditional free()-s in the code but the condition 
usually evaluates a wholly different variable,
a combination of them or a function call (often + assignment), which is 
quite common.
> > Also:
> >    temp = pointer ? realloc (pointer, bytes) : malloc (bytes);
> > Again, the C standard says that realloc(0, bytes) is equivalent to
> > malloc(bytes).
> There's no good reason for this one; probably just extremely old legacy
> code back in the day when various libc were not compatible to the
> restrictions added in later versions of the C standard.
That's possible. It might also have been an attempt at further 
optimization - just calling malloc if pointer == NULL, instead going 
through an additional realloc call.

Either way, I find both of these code snippets very redundant.

Thanks,
Ondrej.