[Python-Dev] How to test for stuff like fsync?

[Guido van Rossum]

> Guido van Rossum wrote:
> > I don't like this.  We're not taking fsync's address for a very good
> > reason, just to save some code duplication.  Removing fsync from
> > those platforms just because our code-saving trick doesn't work there
> > doesn't strike me as good service for those platforms.

[MvL]
> It's not only fsync, though - there would be a lot of duplicated code if
> we were to convert all function pointer passing back to function calls.

Well, too bad.  We could use macros instead.  I believe I started
taking the function address many years ago, and I just did it as a
trick to save some typing, not to save code size.  I think it's wrong
to use it as an argument against including a function on a platform
because the platform doesn't support that trick.

> Also, if the system is lacking a function prototype, at the minimum, the
> compiler will give a warning - at worst, we invoke the function 
> incorrectly, causing a crash.

For fsync(), invoking it wrongly sounds unlikely.  I'm sure the users
of fsync() would gladly take the warning if they can have the
function.

Is there really no function prototype in the system headers for
fsync() on those systems, or don't we see it because we're using
feature selection macros (isn't that the right term?) that hide it?

--Guido van Rossum (home page: http://www.python.org/~guido/)