[cxx-interop] Allow importing templated functions when template args do not appear in function signature

[guitard0g]

Currently if we try to import a templated function when the template parameters do not appear in the function signature, the corresponding Swift function will have extra unused generic parameters, which means we can't call those functions. This patch fixes the problem by making those template params into explicit metatype function parameters to the corresponding Swift function. E.g.

template <typename T>
void foo()

becomes:

func foo<T>(T: T.Type) 

instead of:

func foo<T>() // can't be called

This extra argument is only used to deduce the generic type needed to specialize the C++ function. The corresponding function call made under the hood will cut out the extra argument and forward along the proper parameters to the C++ function.

[guitard0g]

WIP because there seems to be some wonkiness with how this interacts with default template arguments, still investigating a fix there.

[zoecarver]

What a great patch! Thank you, Josh.

I really like how you commented all the code you added so it's super easy to understand and follow. Also, thanks for factoring out that lambda into a general helper function.

I have one or two small nits, but overall, this patch seems good to me :)

[zoecarver]

Let's link this to the SR: XXXX.

[zoecarver]

Oops, meant to actually give you the SR. Just update the TODO to read TODO(SR-13809):. This way we can grep for the SR and find all the places we need to update whenever we add support for dependent types.

[zoecarver]

I know I named this function, but it's kind of ugly, so while you're updating the signature maybe change it to generateSpecializedCXXFunctionTemplate? WDYT?

[zoecarver]

Not in this patch, but at some point we should move all synthesizers to their own file. They just take up space right now and clutter the files they're scattered across.

[zoecarver]

Can you add an expectEquals somewhere in both of these tests. Otherwise (very theoretically), they might just get optimized away. Also it's not a bad idea to just test that the correct thing is returned and there wasn't an argument mismatch or something.

[zoecarver]

@guitard0g let me know if you need help running the bots.

[egorzhdan]

This approach looks great to me! I only noticed one small issue that prevents the C++ stdlib from being imported.

[guitard0g]

@swift-ci please test

[guitard0g]

@swift-ci please test

[guitard0g]

@egorzhdan @zoecarver Thanks for the review! I believe I've addressed all the comments in the most recent revision. It changed a decent bit because I realized there were some quirks with making sure we manually invoke the generic substitutions on the signature of the synthesized thunk. Should be ready for another round of reviews.

[zoecarver]

🚢 it

(I left three very small nits. All are optional.)

[zoecarver]

I wish we could use auto [a, b] = 😢

[zoecarver]

I think various methods/static methods might take a meta type. We need to verify if this will break that.

[zoecarver]

The tests are passing, so I guess this isn't an issue.

[zoecarver]

Swift Test Windows Platform — Build failed.

Now you see why I'm so excited about #36082 :P

[egorzhdan]

Looks great!

[guitard0g]

@swift-ci please smoke test and merge