Re: [Qemu-devel] [PATCH] block: Simplify a few g_try_malloc() error checks

[Kevin Wolf <kwolf@redhat.com>]

Am 27.01.2015 um 14:42 hat Markus Armbruster geschrieben:
> Paolo Bonzini <pbonzini@redhat.com> writes:
> 
> > On 27/01/2015 13:25, Markus Armbruster wrote:
> >> Unlike malloc(), g_try_malloc() & friends return a null pointer only
> >> on failure, never for a zero size.  Simplify tests for failure
> >> accordingly.  This helps Coverity see returned null pointers can't be
> >> dereferenced.  Also makes the code easier to read.
> >
> > Unfortunately that's not what I see from the source:
> >
> > gpointer
> > g_try_malloc (gsize n_bytes)
> > {
> >   gpointer mem;
> >
> >   if (G_LIKELY (n_bytes))
> >     mem = glib_mem_vtable.try_malloc (n_bytes);
> >   else
> >     mem = NULL;
> >
> >   TRACE (GLIB_MEM_ALLOC((void*) mem, (unsigned int) n_bytes, 0, 1));
> >
> >   return mem;
> > }
> 
> You're right.  Brain fart, please ignore.

Should we consider introducing a qemu_try_malloc() that has the desired
behaviour? The (g_try_)malloc() interface is really easy to misuse and
I've messed up some error checks in the block layer myself.

Kevin