Make pointer field of Pin private

[joboet]

This PR is an alternative to #119562.

By using the macro hygiene of macros 2.0 inside the pin macro, the pointer field of Pin does no longer need to be public. Hiding it avoids the issue outlined in #119562, against which a regression test is added.

[rustbot]

r? @Mark-Simulacrum

(rustbot has picked a reviewer for you, use r? to override)

[Noratrieb]

Back when this macro was added, @danielhenrymantilla did originally want to use a macro 2.0, but if I remember it correctly, it was decided that relying on such a macro was not good and that it should use a macro_rules with a hack instead.
Given that we can always go back to a hacked macro_rules (by renaming the field), I don't immediately see the problem, but it would be good to revisit that as an explicit decision.

[mohe2015]

What does cargo expand (or the rust equivalent) do now? Because I remember that it expanded to code with the pointer field.

[danielhenrymantilla]

Removal of turbofish type is orthogonal to field privacy / hygiene, and technically allows for code such as let p: Pin<*mut _> = pin!(...); (letting a coërcion happen on the pointer: ... assignment).
Since this whole thing is already quite tricky/subtle, let's not risk unnecessary flexibility (which appears to be unsound⚠️ ):

Suggested change

	Pin{pointer:&mut{ $value }}
	Pin::<&mut_>{pointer:&mut{ $value }}

[joboet]

I've readded this and added a comment explaining why it should not be removed.

[joboet]

I wonder if I should also add a test for this case in this PR, so that others cannot stumble over the same thing.

[fmease]

cc @LegionMammal978

[fmease]

What does cargo expand (or the rust equivalent) do now?

Well, it's going to show pointer. However, cargo expand / rustc -Zunpretty=expanded doesn't come with any stability guarantees.

[mohe2015]

What does cargo expand (or the rust equivalent) do now?

Well, it's going to show pointer. However, cargo expand / rustc -Zunpretty=expanded doesn't come with any stability guarantees.

I sometimes compile the expanded code to get better source line information etc. and to my understanding I can't easily make it compile anymore then by adding the internal feature flag? Not a too big issue but just came to my mind.

[Mark-Simulacrum]

r? @Nilstrieb

Perhaps you remember which team / group made the decision referenced here:

Given that we can always go back to a hacked macro_rules (by renaming the field), I don't immediately see the problem, but it would be good to revisit that as an explicit decision.

[Noratrieb]

I went through the discussion and it looks like this was a libs-api team decision in a meeting: #93176 (comment)
The status of macro 2.0 hasn't change in the meantime. What has changed is that we now have a concrete case of this hack going wrong in the issue this PR tries to fix, where the field name leaks through deref.
With this knowledge, is it worth revisiting the original choice of using macro 2.0, or is it deemed "good enough" to just give the pointer field a "more obscure" name like __pointer or something even more obscure?
With that question, r? @m-ou-se (or @rust-lang/libs-api in general)

[bors]

☔ The latest upstream changes (presumably #119722) made this pull request unmergeable. Please resolve the merge conflicts.

[LegionMammal978]

I noticed today that there's also a comment in compiler/rustc_hir_typeck/src/fn_ctxt/suggestions.rs:

// Skip suggestions for unstable public fields (for example `Pin::pointer`)

The example should probably be omitted. Apart from Pin::pointer, there don't appear to be any other unstable public fields in the standard library.

[m-ou-se]

I don't think we're supposed to use full macro 2.0 hygiene in a public macro yet. cc @petrochenkov

[m-ou-se]

Considering #119562 is merged, macros 2.0 hygiene isn't ready yet, and the temporary lifetime rfc work progressing towards a solution for the pin!() macro without public Pin fields, I'm closing this.