Allow Size to be any valid u64

[oli-obk]

cc rust-lang/miri#378 (comment)

The alternative is to make mir::interpret's pointer offsets not be Size

fixes#50917

r? @eddyb

[RalfJung]

LLVM doesn't support allocations or types larger than half the address space (and I am not sure about exactly half the address space). Is there any risk of those coming up after this patch?

[oli-obk]

We do have tests for these kinds of types. Not sure how to do tests for allocations of such sizes xD

[eddyb]

Shouldn't this be a From impl?

[oli-obk]

I considered that. But i think being explicit is better. I don't feel too strongly about that though, so if you want I'll change it

[eddyb]

I think zero is misleading, I initially assumed it'd be a null pointer.

[eddyb]

Also, random thought: maybe MemoryPointer should be AllocPointer or AbstractPointer?

[oli-obk]

We could just name it Pointer and rename the current Pointer struct to PrimvalPointer

[eddyb]

That seems worse. Maybe this?

enumScalar{Undef,Bits(u128),Ptr(Pointer),}enumValue{Scalar(Scalar),ScalarPair(Scalar,Scalar),ByRef(Pointer,Align),}

[oli-obk]

What about the current struct Pointer(Primval) type? It's just a wrapper around a PrimVal, but helps us not confuse normal PrimVals with those whose type is a pointer type.

[eddyb]

The type doesn't really matter, the value does (unless there are pointer-specific methods on the wrapper?). And I prefer Scalar over Primval, but I also forgot the latter is really a thing.

[eddyb]

cc @alexcrichton@nagisa@arielb1@rkruppe
Does any of you remember what failure mode LLVM had when we started adding type size checks?
This PR does not affect any of the checks done when computing sizes of ADTs, arrays, etc. - but Size can now contain byte counts that could show up as negative if placed in signed integers.
(or overflow if multiplied by 8 other than through the .bits() method)

[alexcrichton]

Hm I don't recall myself, but others might!

[eddyb]

@bors r+

[bors]

📌 Commit 9f79a19 has been approved by eddyb

[bors]

⌛ Testing commit 9f79a19 with merge 29ffe51...

[bors]

☀️ Test successful - status-appveyor, status-travis
Approved by: eddyb
Pushing 29ffe51 to master...