could not evaluate static initializer

[Mark-Simulacrum]

https://crater-reports.s3.amazonaws.com/beta-1.39-1/beta-2019-09-28/reg/vptr-0.2.0/log.txt
looks like this is inside a proc macro so would be good to figure out what the raw expression is

cc #64962

[Centril]

cc @oli-obk @RalfJung -- my guess is the crate is doing something UB in the static.

[ecstatic-morse]

This proc macro creates a reference to address 0 here. It should switch to memoffset, whose behavior is not as undefined 😄.

[Centril]

cc @ogoffart

Closing as wontfix because of UB.

[RalfJung]

Well, there's no way to do offsetof in CTFE right now. Previously, there were some approaches that seemed to work but actually were UB, like this one. But now our const-time UB checker got a bit better, hence the crater regression.

#63810 is the key ingredient to enable a less insane version of offsetof in CTFE. (Basically, the CTFE variant is then just as unsound as memoffset. In other words, it is UB but works in practice because right now we do not exploit that UB in the compiler.)

[RalfJung]

I opened an issue in the crate to give the author a heads-up: ogoffart/vptr#1.

@Mark-Simulacrum were there other regressions that look similar? offsetof is one of these notorious operations that Rust doesn't support but people want it, and they want it so badly that instead of making the language support offsetof they write programs with UB that happen to work at the time of implementation... so I'd not be surprised to see that kind of snippet come up a few times.

[Mark-Simulacrum]

I didn't see any other cases of this.

[ecstatic-morse]

Ugh, I missed the static declaration.