Get to LLVM-lint-clean

[graydon]

Not sure what all is involved, but I gather when we turn on LLVM's internal lint checking, all hell breaks loose. Let's tidy this up.

[Blei]

I would say over 90% (not measured) of lint errors are because of bitcasting function pointers, like the ones mentioned in #6779. I have a patch that removes most of them, but is somewhat hacky and not complete yet. But most importantly, it makes the code bigger and doesn't seem to help time-wise either. According to the LLVM IRC channel, those lint errors are false positives.

[msullivan]

Nothing new to add.

[thestinger]

A big cause of this is #8720.

[thestinger]

I think most of the rest are caused by type_use (similar to the #8720 issue). Replacing it would mergefunc would be the laziest way of fixing them.

[brson]

Not 0.8

[thestinger]

The only remaining lint appears to be many cases of Pessimization: Static alloca outside of entry block.

[steveklabnik]

Triage: I pinged @Aatch about this on IRC, but I am bad at our build system, and so I have nothing specific to add.

[tamird]

@steveklabnik rustc -C passes=lint turns this on.

[steveklabnik]

Triage: to try this today, do

RUSTFLAGS="-C passes=lint" python x.py build

We are very much not close to clean, though it seems like many of these are similar issues, so it's possible that a fix would fix many of them.

[nagisa]

We have 2 broad kinds of warnings coming up while translating rustc:

Undefined behavior: Memory reference address is misaligned
  store i32 %1, i32* %3, !dbg !2710

and

Unusual: noalias argument aliases another argument
  %4 = call zeroext i1 @core::cmp::PartialEq::gt(i8* noalias readonly dereferenceable(1) %1, i8* noalias readonly dereferenceable(1) %3), !dbg !679

The first is very scary, not sure about 2nd.

[matthiaskrgr]

cargo 0.26.0-nightly (91e36aa86 2018-01-25)
rustc 1.25.0-nightly (bacb5c58d 2018-01-26)

I was giving RUSTFLAGS="-C passes=lint" python x.py build a try and these are the kinds that I got:

Unusual: noalias argument aliases another argument
  %3 = call zeroext i1 @_ZN4core3cmp10PartialOrd2gt17h599e81f888a1fe30E(i8* noalias readonly dereferenceable(1) %0, i8* noalias readonly dereferenceable(1) %2)

Undefined behavior: Buffer overflow
  %3 = load i64, i64* %2, align 1, !range !0

Unusual: Return statement in function with noreturn attribute
  ret void

While doing a debug crater build, ( RUSTFLAGS="-C passes=lint" cargo build in crater repo) this showed up:

Unusual: unreachable immediately preceded by instruction without side effects
  unreachable, !dbg !69