2019-12-12

Announcements

@pnkfelix posted an RFC about future-incompat lints RFC 2834
@eddyb landed some additional caching for trait selection which gives some nice speedups #66821
Const eval now has support for if/match and a PR for loop/while is open.
@eddyb is resurrecting their SROA MIR optimization #48300

resolve: Always resolve visibilities on impl items #67236
- Approved for beta backport
resolve: Resolve visibilities on fields with non-builtin attributes #67106
- Approved for beta backport

MIR does not know about static items anymore. Instead we just have a constant whose value is &YOUR_STATIC, &raw mut YOUR_MUT_STATIC or &raw YOUR_EXTERN_STATIC. This simplifies a lot of logic in the MIR and will (together with the next bullet point) allow PlaceBase to just be a Local (so a u32 index for a local to be accessed). #66587
@spastorino is about to finish a PR that removes promoteds from the list of things that are Places in MIR
mir::Body does not contain interior mutability components anymore. It used to have a cache of some information on the MIR CFG that could get cleared and reinitialized as required. This is abstracted away in the BodyCache type.
Promoteds inside constants and statics are now actually separated from their constant’s mir::Body instead of having a scheme around removing StorageDead statements #66642
Cross crate inlining now preserves debug info #66789
There is a new MIR optimization eliminating basic blocks that are unreachable because they are match arms of uninhabited variants #65694
There is a new MIR optimization to work around the problem that LLVM can’t figure out trivial match arms like Ok(x) => Ok(x) #66282
The const prop optimization now runs on all builds (even debug builds) because this gives quite some nice build time improvements #66074
We had a mir-opt room at RustFest, which was a total success and people loved it.