gen-guest-c: optional parameters with null pointers

[guybedford]

In the C guest generator, functions which have optional parameters of the form:

x: func(a: option<a>, b: option<b>) -> _

Currently output signatures like:

void interface_x(interface_option_a_t *a, interface_option_b_t *b);

Where it would be nice to support the more C-like interface:

void interface_x(interface_a_t *a, interface_b_t *b);

where a null pointer could be used to imply the optional parameter value instead.

[guybedford]

As far as I can tell, implementation here would require that OptionLift and OptionLower operations have additional information in core indicating if they are parameters that are being lifted or lowered, so the special casing can apply.

[alexcrichton]

I think that may not be necessary since the context of a generator fixes that, e.g. for C when generating an imported call lifts only happen for results. So given the context of import or export, and C, I think that designates what's being lowered and what's being lifted.

[guybedford]

I think this might still be necessary actually, as when defining an imported function one ends up with a wrapper looking like the following:

void fn (float *maybe_a, float *maybe_b) {
  interface_option_float_t a;
  a.is_some = maybe_a != NULL;
  if (maybe_a) {
    a.val = *maybe_a;
  }
  if (a.is_some) {
    //... logic...
    option = 1;
    option1 = *payload0;
  } else {
    option = 0;
    option1 = 0;
  }

where my concern is that we end up copying the data out of the pointers in this approach when perhaps we should be optimizing this?

For what it is worth, I hit up against similar issues in the JS generator.

Alternatively I can just implement the wrapper approach as described above for now, leaving the door open to further optimizations in future to remove the extra copy.

[alexcrichton]

Ah this is something I think it safe to optimize for later. The set of data possibly being extraneously loaded per call would be a small enough fixed size that I doubt this'd ever be an issue. If it is an issue we could try adding restrict or w/e corresponding attributes C has for "this doesn't alias" or things like that.

[guybedford]

Ok, I'll work to the above for now.

[guybedford]

Landed in #453.