`Fuse` impls expose use of specialization (`default fn`)

[Centril]

Context: #70750 (comment)

We have some uses of default fn in src/libcore/adapters/fuse.rs which allow users to specialize those functions. Instead, these default fns should be moved into internal (private) traits so that the specialization isn't exposed.

This issue has been assigned to @rakshith-ravi via this comment.

[rakshith-ravi]

Supp. Can I take this up?

Oh also, the file is in src/libcore/iter/adapters/fuse.rs not src/libcore/adapters/fuse.rs. Had me confused for a while, but picked it up from context of the other PR, just saying.

[cuviper]

@rustbot assign @rakshith-ravi

[cuviper]

I fear what the extra indirection will do to #70332, but I guess we'll just have to see...

[rakshith-ravi]

Alright. I'll go through that as well and suggest changes there, in case I can think of any.

Unlike my last PR, this time, I'd like to give this a shot without any mentoring. However, if I get stuck somewhere, I might request for help. Bear with me please.

Anyways, it's a little too late for me here (IST). Will get on it first thing tomorrow morning. Thanks

[cuviper]

It's also possible that could work out better if there's a way for Chain to directly reach the non-specialized version, rather than shimming its own Defuse to avoid FusedIterator effects.

[rakshith-ravi]

So if I understand the situation correctly, we need to have Iterator inherit from IteratorInternal or something like that, which allows for specialization and perform all our specialized work on that, while not exposing the default fns to anything outside the crate, yes?

This will also mean that Iterator will provide a blanket implementation that uses the IteratorInternal's implementation.

[rakshith-ravi]

It's also possible that could work out better if there's a way for Chain to directly reach the non-specialized version, rather than shimming its own Defuse to avoid FusedIterator effects.

I don't think I entirely understand what you're saying. Could you help me clarify that please?

[cuviper]

We definitely don't want to change the Iterator trait, if that's what you mean. I would look at Zip and its ZipImpl to see how that's privately specialized.

It's also possible that could work out better if there's a way for Chain to directly reach the non-specialized version, rather than shimming its own Defuse to avoid FusedIterator effects.

I don't think I entirely understand what you're saying. Could you help me clarify that please?

My thought was that we could make #70332 use some internal part of your work, e.g. FuseImpl, to bypass some of the nested calls. But now I'm also trying another Chain independent of Fuse in #70896 which I think is promising...