Re: sprintf warning on overlapping output

[Florian Weimer <fw at deneb dot enyo dot de>]

* Jeff Law:

>>>> sprintf(buf, "%s %d", buf, x);

>> Or we could make this well-defined because it is such a useful
>> extension.

> That just encourages developers to write non-portable code.  I'd
> rather see this kind of thing halt the program in its tracks before
> wandering into the realm of undefined or implementation defined
> behavior.

But then the programmer will likely write something like this:

  sprintf(buf + strlen (buf), "%d", x);

Or perhaps even:

  snprintf(buf + strlen (buf), sizeof (buf), "%d", x);

(This passes an incorrect size.)

As far as I can tell, this idiom currently prevents
__builtin_object_size from providing useful data and makes us lose
fortify protection, while Bernd's original code has it.