Revert revert of constness in #86003

[usbalbin]

Re-constify mem::swap, mem::replace, ptr::write which were marked as not const in #86003

Once the checks pass, this should solve #86236

[rust-highfive]

r? @Mark-Simulacrum

(rust-highfive has picked a reviewer for you, use r? to override)

[usbalbin]

Would it make sense to make a tracking issue for const_ptr_write?

[RalfJung]

@pnkfelix any objections to this? I am not sure why these were de-constified to begin with, so just making sure I am not missing anything here. :)

[RalfJung]

@usbalbin we also lost src/test/ui/consts/copy-intrinsic.rs in #86003; could you add that back as well?

[RalfJung]

LGTM -- r=me, except that it'd be good to hear back from @pnkfelix. Let's give them a few more days.

r? @RalfJung

[usbalbin]

Sorry but I am not quite sure what to do with the errors (or rather lack of errors)

[usbalbin]

Are those things that are no longer checked? Should I remove those tests or somehow make sure the errors are detected again?

[RalfJung]

Something is strange, it looks like none of the errors showed up -- but the exit status is still 1...?

[RalfJung]

Ah, I think I know what happens: the errors now have the wrong span; they are shown with spans in the standard library.

To fix this, you should call the intrinsics directly in this test. You will probably need to add a extern "rust-intrinsic" block to import them.

[RalfJung]

So, something like this. But at least on the playground this doesn't change the spans that are shown...

[usbalbin]

Thanks :)

What should I do about

error: module has missing stability attribute
--> src/main.rs:2:1

?
Same error in your playground

[RalfJung]

Ah... well we'll need to add a stability attribute. This should do it:

#![stable(feature = "dummy", since = "1.0.0")]

[usbalbin]

So if you write code which calls unstable code than you have to be explicit about the stabilitiy of your code?

Again thanks :)

Would you like me to squash the two latest commits (2faa57a and d1ae4e2) onto one?

[RalfJung]

A chain of things happens here:

• To call copy_nonoverlapping, it needs to be const fn
• To make an intrinsic const fn, we need to add rustc_const_unstable
• To be allowed to use that attribute we need to enable the staged_api feature
• Once we enable that feature we need to give stability attributes to everything; the easiest thing is to make is all stable so that we don't have to enable feature gates to call our own stuff. (Maybe we don't actually need that, and #![unstable(...)] would also work.

Would you like me to squash the two latest commits (2faa57a and d1ae4e2) onto one?

Seems fine to me to keep this separate.

[usbalbin]

Ah, okay

[RalfJung]

@pnkfelix confirmed that as long as we don't touch copy/copy_nonoverlapping, we should be good. So, let's (re-)land this. Thanks @usbalbin :)
@bors r+

[bors]

💔 Test failed - checks-actions

[RalfJung]

Looks like #86194 changed the output of some tests added here.

[usbalbin]

@pnkfelix confirmed that as long as we don't touch copy/copy_nonoverlapping, we should be good. So, let's (re-)land this. Thanks @usbalbin :)
@bors r+

Thank you :)

[RalfJung]

@bors r+

[bors]

📌 Commit 4aa1267 has been approved by RalfJung

[bors]

⌛ Testing commit 4aa1267 with merge 49ba936...

[bors]

☀️ Test successful - checks-actions
Approved by: RalfJung
Pushing 49ba936 to master...