Promoteds can contain raw pointers, but these must still only point to immutable allocations

[oli-obk]

fixes#67601

r? @RalfJung

cc @wesleywiser in order to not change behaviour in this PR, const prop uses the constant rules for interning, but at least there's an explicit mode for it now that we can think about this in the future

[bors]

☔ The latest upstream changes (presumably #67327) made this pull request unmergeable. Please resolve the merge conflicts.

[SimonSapin]

Friendly ping: what’s next for this PR? Can I help?

[oli-obk]

I just haven't gotten to it. Probably Tuesday or some other day next week

[oli-obk]

I addressed all review comments and marked the issue as a stable to nightly regression

[RalfJung]

Right, I was thinking about this branch here when asking about nested allocations in constants. Isn't this unreachable then?

[oli-obk]

It is reachable, but I'm entirely unclean on why it is reachable

constRAW_TRAIT_OBJ_CONTENT_INVALID:*constdynTrait = &unsafe{BoolTransmute{val:3}.bl}as*const_;

hits the InternKind::Constant arm.

See the MIR on https://play.rust-lang.org/?version=stable&mode=debug&edition=2018&gist=3bc6c8b165b614bbe6cf5566d30752de there's no StorageDead for the allocation containing the bool with bit pattern 3.

I'm not really sure what's going on there. Technically, since we're inside a constant, shouldn't this trigger promotion and promote the transmuted value?

[oli-obk]

Under unleash it's also reached in

constMUTABLE_BEHIND_RAW:*muti32 = &UnsafeCell::new(42)as*const_as*mut_;

but that triggers the delay_span_bug as expected.

Also under unleash, but without an ICE (just triggering the dynamic checks in interp):

constMUTABLE_BEHIND_RAW:*muti32 = &UnsafeCell::new(42)as*const_as*mut_;//~^ WARN: skipping const checksconstMUTATING_BEHIND_RAW:() = {// Test that `MUTABLE_BEHIND_RAW` is actually immutable, by doing this at const time.unsafe{*MUTABLE_BEHIND_RAW = 99//~ ERROR any use of this value will cause an error}};

[RalfJung]

I'm not really sure what's going on there. Technically, since we're inside a constant, shouldn't this trigger promotion and promote the transmuted value?

Ah, this must be the "not-promotion" also described in this document (Ctrl-F "looks like"). That makes sense.

[RalfJung]

What I don't understand though is why RAW_TRAIT_OBJ_CONTENT_INVALID doesn't trigger the ICE. Why is the allocation already immutable?

[RalfJung]

(The PR doesn't change anything here so this doesn't block landing, but I'd really like to understand this and then see it documented.)

[wesleywiser]

I would find documentation here immensely helpful.

[oli-obk]

If you have a holistic view of what happens, it would really help to write that down.

I don't, which is why my answers are so confusing. They are just a brain dump of me discovering what is going on, not me understanding it entirely

[oli-obk]

These issues are preexisting, so after this PR is merged I'll open a new one to write docs and actually figure out what is going on

[RalfJung]

I don't, which is why my answers are so confusing. They are just a brain dump of me discovering what is going on, not me understanding it entirely

Fair enough. :)

[bors]

☔ The latest upstream changes (presumably #68078) made this pull request unmergeable. Please resolve the merge conflicts.

[bors]

☔ The latest upstream changes (presumably #67000) made this pull request unmergeable. Please resolve the merge conflicts.

[RalfJung]

r=me with the if let Some(_) = ecx.tcx.alloc_map.lock().get(alloc_id) branch removed.

[oli-obk]

@bors r=RalfJung

[bors]

📌 Commit 69ffe7b has been approved by RalfJung

[RalfJung]

@bors p=1 (regression fix, and this has been sitting for a while)

[bors]

⌛ Testing commit 69ffe7b with merge faf45c5...

[bors]

☀️ Test successful - checks-azure
Approved by: RalfJung
Pushing faf45c5 to master...

[rust-highfive]

📣 Toolstate changed by #67603!

Tested on commit faf45c5.
Direct link to PR: #67603

💔 rls on linux: test-pass → test-fail (cc @Xanewok, @rust-lang/infra).

[SimonSapin]

Confirmed that this fixes #67601 for Servo. Thanks!