Add forwarders for 'name()' method for Funcs, ImageParam

[steven-johnson]

(followup to #6772)

[abadams]

Why this change? Is the macro used later?

[steven-johnson]

Oh, sorry, drive-by change -- apparently a holdover from way back when there were multiple forwarding macros, just dead code now

[steven-johnson]

OK, I obviously didn't even test this locally because I thought it was a no-brainer, silly me

[steven-johnson]

Had to rename the inherited GIOBase::name() method (as I did for type() yesterday) to avoid ambiguity with the new forwards, PTAL

[abadams]

Hrm, I guess the name of an Output doesn't have to match the name of the Func you assign to it. As things currently stand, does calling name() return the Output name (i.e. gio_name)? Does this change it to return the Func name?

[steven-johnson]

Hrm, I guess the name of an Output doesn't have to match the name of the Func you assign to it.

Correct -- in this case we take pains to ensure that the name specified in the metadata matches the gio_name rather than the func name (especially since the latter is often uniquified with $2, etc)

As things currently stand, does calling name() return the Output name (i.e. gio_name)?

Yes.

Does this change it to return the Func name?

Yes. (The goal here being to try to get the Input and Output wrappers to be as close to the underlying object as possible.)

FWIW, though, I didn't find any usage inside Halide tests, apps, etc of code inspecting an Input/Output name -- this change was to satisfy a small bit of code inside Google.

[abadams]

That seems surprising. My expectation would have been that if you declare an Output with a name, calling name() on it gives you that name back, and you have to do something like my_output_buffer.func().name() or Func(my_output_buffer).name() to get the name of a Func.

[steven-johnson]

My expectation would have been that if you declare an Output with a name, calling name() on it gives you that name back

OK, but keep in mind that the (current) Python generator design has no special name() method on inputs or outputs -- the only such method available is the one on the underlying Func, ImageParam, etc. -- so this change brings the C++ behavior closer to the (proposed) Python design.

Also keep in mind that even in C++, the gio_name of all Inputs and Outputs is 100% fixed at compiletime; there is no way to replace those names via GeneratorParam or whatnot, so having a name() getter is something that would rarely be used.

[steven-johnson]

Monday Morning Review Ping

[steven-johnson]

I'm just going to drop this PR entirely: the only usage of the GIOBase::name() function that I'm aware of was in one piece of Google code, and we changed it to use a hardcoded name instead (because that's essentially what it was using). I presume that if/when other folks realize that the name() method is no longer public, they will let us know whether it's a problem for them.