Update rust toolchain to 2022-12-11 #2045
Conversation
|
One of the harnesses in s2n-quic ( I'm investigating. |
zhassan-aws
force-pushed
the
toolchain-2022-12-11
branch
from
4b979e9
to
b0da68e
Compare
zhassan-aws
force-pushed
the
toolchain-2022-12-11
branch
from
a46380e
to
cc0c3ba
Compare
|
@celinval I confirmed that regressions pass with this PR with CBMC's latest commit (diffblue/cbmc@f76d7fa): https://github.com/zhassan-aws/kani/actions/runs/3944422079 |
zhassan-aws
force-pushed
the
toolchain-2022-12-11
branch
from
b5894ed
to
6365cd9
Compare
| targets: &SwitchTargets, | ||
| loc: Location, | ||
| ) -> Stmt { | ||
| let v = self.codegen_operand(discr); | ||
| let switch_ty = self.monomorphize(switch_ty); | ||
| let switch_ty = v.typ().clone(); |
There was a problem hiding this comment.
Just confirming my reading of this: we remove the monomorphize because we are dealing with a "cbmc type" not a rust one anymore, and presumably the monomorphize happened as part of codegen_operand right?
There was a problem hiding this comment.
Not really. The switch_ty field was removed from the SwitchInt struct in rust-lang/rust#105234 because it's redundant: its type is always the same as the type of the discr field. So my change just copies the type that we've already created for the discr field.
See this comment for more info: https://github.com/rust-lang/rust/pull/105234/files#diff-3a6077a453e9ef35b5b85b4419025066cc7a0f3bad1207ae98860902c817a252L531
There was a problem hiding this comment.
But yes, AFAIK, the monomorphization happens in codegen_operand.
| @@ -865,10 +878,10 @@ impl<'tcx> GotocCtx<'tcx> { | |||
| fn codegen_alignment_padding( | |||
| &self, | |||
| size: Size, | |||
| layout: &Layout, | |||
| layout: &LayoutS<VariantIdx>, | |||
There was a problem hiding this comment.
Are you sure we should be switching to LayoutS? It seems like the sort of thing there's probably a better type we should be using here.
There was a problem hiding this comment.
I brought in those changes from #1983. @adpaco-aws: do you know if there's a better type to use?
There was a problem hiding this comment.
Given the little changes to the code that uses this type, this seems like a good fit. Further improvement should probably be orthogonal to updating the toolchain. I think this has to do with this PR: rust-lang/rust#103693
| @@ -392,7 +390,7 @@ impl<'tcx> GotocCtx<'tcx> { | |||
| let cases = targets | |||
| .iter() | |||
| .map(|(c, bb)| { | |||
| Expr::int_constant(c, self.codegen_ty(switch_ty)) | |||
| Expr::int_constant(c, switch_ty.clone()) | |||
There was a problem hiding this comment.
nit: I don't think you need the clone here.
There was a problem hiding this comment.
I also thought it shouldn't be needed, but the compiler corrected me :)
error[E0507]: cannot move out of `switch_ty`, a captured variable in an `FnMut` closure
--> kani-compiler/src/codegen_cprover_gotoc/codegen/statement.rs:393:43
|
372 | let switch_ty = v.typ().clone();
| --------- captured outer variable
...
392 | .map(|(c, bb)| {
| --------- captured by this `FnMut` closure
393 | Expr::int_constant(c, switch_ty)
| ^^^^^^^^^ move occurs because `switch_ty` has type `cprover_bindings::goto_program::Type`, which does not implement the `Copy` trait
It's because it's used in a closure which might be called multiple times.
| @@ -865,10 +878,10 @@ impl<'tcx> GotocCtx<'tcx> { | |||
| fn codegen_alignment_padding( | |||
| &self, | |||
| size: Size, | |||
| layout: &Layout, | |||
| layout: &LayoutS<VariantIdx>, | |||
There was a problem hiding this comment.
Given the little changes to the code that uses this type, this seems like a good fit. Further improvement should probably be orthogonal to updating the toolchain. I think this has to do with this PR: rust-lang/rust#103693
Compiletest behavior has changed to always failing fast after model-checking#2045. This PR introduces a `--no-fail-fast` flag to compiletest which will execute the entire suite regardless of failure. The regression script uses `--no-fail-fast` so all failures in a suite are printed as part of the CI.
Compiletest behavior has changed to always failing fast after #2045. This PR introduces a --no-fail-fast flag to compiletest which will execute the entire suite regardless of failure. The regression script uses --no-fail-fast so all failures in a suite are printed as part of the CI.
2022-12-04GenericArgNestedMetaItemConstKind::Exprreturn on reach visitorLayoutDescription of changes:
This PR builds on @adpaco-aws's commits in #1983, but bumps the version further to 2022-12-11 (instead of 2022-12-04), to include some of the later commits related to generators, which are necessary to fix how we codegen them.
The relevant PR is rust-lang/rust#105082 which was merged on 2022-12-10.
Another PR that required changes is rust-lang/rust#105234.
Resolved issues:
Towards #2044
Related RFC:
Optional #ISSUE-NUMBER.
Call-outs:
I suggest reviewing by commit because the clippy fix commits touch many files.
Testing:
How is this change tested? Current regressions
Is this a refactor change? No
Checklist
By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 and MIT licenses.