Move recursion check for zsts back to read site instead of access check site

[oli-obk]

Reverts #71140 (comment)

Fix #71612
Fix #71709

r? @RalfJung

[RalfJung]

Hm... honestly I am not convinced if that is the right fix long-term. I anyway plan to change check_offset_align to return an Option<&Allocation> or so, as part of cleaning up the Allocation API in preparation for rust-lang/miri#841. There's just so many other things that keep preempting this task.^^

However, there is no reason to do this re-work now, and this adds a testcase to make sure I find a good way to handle that when I do my planned changes. So, fine for me.

[oli-obk]

Hm... honestly I am not convinced if that is the right fix long-term.

Neither am I. I did try to fix it in memory.rs first, but that ended up looking very hacky, so I chose to go for a simple fix here and leave future refactorings to... well the future.

I anyway plan to change check_offset_align to return an Option<&Allocation> or so, as part of cleaning up the Allocation API in preparation for rust-lang/miri#841. There's just so many other things that keep preempting this task.^^

However, there is no reason to do this re-work now, and this adds a testcase to make sure I find a good way to handle that when I do my planned changes. So, fine for me.

[RalfJung]

Are we sure "check" is enough?

[oli-obk]

Yes, the const prop one occurred during mir optimization of fn new, which will still happen in check mode, and the static one happened during evaluation of the static's initializer, which also happens in check mode.

[RalfJung]

r=me with that last question resolved.

[oli-obk]

@bors r=RalfJung

[bors]

📌 Commit c64c776 has been approved by RalfJung

[bors]

⌛ Testing commit c64c776 with merge 6318d24...

[bors]

☀️ Test successful - checks-azure
Approved by: RalfJung
Pushing 6318d24 to master...