[ccan] Valgrind + tests

[Rusty Russell]

On Tuesday 24 March 2009 23:18:25 Tim Post wrote:
> Hi,
> 
> On Tue, 2009-03-24 at 23:00 +1030, Rusty Russell wrote:
> 
> > Sure.  Or you can submit something that uses it, and it'll break my compile
> > (-Wundef) and I'll update config.h automatically :)
> > 
> 
> Can you please explain the sanity of using -Wundef when knowingly using
> something that explicitly points to (and encourages modification of)
> config.h?
> 
> If HAVE_VALGRIND_VALGRIND_H is not set, it would make little difference
> to someone using a module that they grabbed. However, someone using
> autotools would have this on or off depending on what configure found.
> 
> Or am I missing something? Quite possibly, I am :)

OK, it works like this.  Instead of using the standard:

	#ifdef HAVE_FOO

CCAN modules use:

	#if HAVE_FOO

The C preprocessor treats unknown symbols as 0, so as long as the convention
is "#define HAVE_FOO 1", these two are equivalent.

*Except* that I compile here with -Wundef -Werr: -Wundef warns about undefined
preprocessor symbols, and -Werr makes sure I notice it by erroring out.  That
means that when someone makes up a new HAVE_ symbol and I am going to spot it
and figure out whether I should set it or not.

It avoids the classic error where you miss new features simply because you
didn't update config.h

Hope that clarifies?
Rusty.