Re: [Qemu-devel] [PATCH v2 03/16] sockets: allow port to be NULL when listening on IP address

[Markus Armbruster]

Peter Maydell <address@hidden> writes:

> On 22 October 2015 at 10:43, Daniel P. Berrange <address@hidden> wrote:
>> For reasons I don't understand, it appears that even string
>> fields which are marked as optional get a 'has_XXX' flag,
>> to distinguish betweeen a NULL string and an unset string.
>> I struggle to imagine why we need this. It makes sense for
>> integer/boolean types, but I would have thought just checking
>> for NULL was sufficient for string types.
>
> I think the argument here is pure consistency. *Every* optional
> field should be implemented as has_fieldname (bool) plus fieldname
> (actual value). Special casing string types because we happen
> to have a special value we can use to indicate 'no string'
> seems to me like it would cause more confusion than it would
> be worth.

Having a has_FOO for pointer-valued FOOs also causes confusion, because
it begs questions like "does has_FOO imply FOO != NULL?" and "if
!has_FOO, do I have to worry about freeing FOO?".

I believe we have these has_FOOs not because someone tried to minimize
confusion, but due to laziness: it was simpler to blindly generate the
has_FOO for every FOO.

Since I don't think distinguishing "pointer argument omitted" from
"pointer argument is null" makes sense, I want the has_FOO killed for
pointer FOOs.  Perhaps it's even in the QAPI review queue already, which
Eric keeps filling faster than I can drain it :)