ConstProp: Correctly remove const if unknown value assigned to it. #118426
Conversation
|
r? @wesleywiser (rustbot has picked a reviewer for you, use r? to override) |
rustbot
added
S-waiting-on-review
|
Some changes occurred to MIR optimizations cc @rust-lang/wg-mir-opt |
| } | ||
|
|
||
| fn main() { | ||
| assert_eq!(size_of::<u32>(), std::mem::size_of::<u32>()); |
There was a problem hiding this comment.
I don't think this is actually running, and in general I'm not sure what the right way to write a robust test for this is.
There was a problem hiding this comment.
By "running", you mean testing that the assert doesn't fire? If so, then you may want to copy this test into tests/ui/$SOMEWHERE and mark it as a // run-pass test.
(please also give this test file a better name, preferably one that actually describes what it does, here and possibly in tests/ui if you copy it there).
|
Please rebase out the second and fourth commits in this stack, since the latter undoes the former, and it's not particularly useful for git history that way. Either that, or just drop the last commit, since it's possibly useful to keep the instrumentation around, as long as it's debug or trace level. |
rust-cloud-vms
bot
force-pushed
the
const-wat
branch
from
6c06cf4
to
9121a41
Compare
|
Some changes occurred to MIR optimizations cc @rust-lang/wg-mir-opt |
|
r=me when ci is green, since this seems like a pretty nasty regression cc @cjgillot who may be curious about this const prop misoptimization, perhaps you have some ideas to make this code less fragile. |
| @@ -0,0 +1,23 @@ | |||
| // unit-test: ConstProp | |||
| // compile-flags: -O | |||
| // skip-filecheck | |||
There was a problem hiding this comment.
Could you add the filecheck annotations? See other files for example.
Yes. Removing this pass entirely, replaced by gvn. |
rust-cloud-vms
bot
force-pushed
the
const-wat
branch
from
4f310a3
to
0981b7b
Compare
rust-cloud-vms
bot
force-pushed
the
const-wat
branch
from
0981b7b
to
6e956c0
Compare
aDotInTheVoid
added
the
beta-nominated
|
@bors r=compiler-errors,cjgillot |
bors
added
S-waiting-on-bors
…iaskrgr Rollup of 7 pull requests Successful merges: - rust-lang#118157 (Add `never_patterns` feature gate) - rust-lang#118191 (Suggest `let` or `==` on typo'd let-chain) - rust-lang#118231 (also add is_empty to const raw slices) - rust-lang#118333 (Print list of missing target features when calling a function with target features outside an unsafe block) - rust-lang#118426 (ConstProp: Correctly remove const if unknown value assigned to it.) - rust-lang#118428 (rustdoc: Move `AssocItemRender` and `RenderMode` to `html::render`.) - rust-lang#118438 (Update nto-qnx.md) Failed merges: - rust-lang#118268 (Pretty print `Fn<(..., ...)>` trait refs with parentheses (almost) always) r? `@ghost` `@rustbot` modify labels: rollup
Rollup merge of rust-lang#118426 - aDotInTheVoid:const-wat, r=compiler-errors,cjgillot ConstProp: Correctly remove const if unknown value assigned to it. Closes rust-lang#118328 The problematic sequence of MIR is: ```rust _1 = const 0_usize; _1 = const _; // This is an associated constant we can't know before monomorphization. _0 = _1; ``` 1. When `ConstProp::visit_assign` happens on `_1 = const 0_usize;`, it records that `0x0usize` is the value for `_1`. 2. Next `visit_assign` happens on `_1 = const _;`. Because the rvalue `.has_param()`, it can't be const evaled. 3. Finaly, `visit_assign` happens on `_0 = _1;`. Here it would think the value of `_1` was `0x0usize` from step 1. The solution is to remove consts when checking the RValue fails, as they may have contained values that should now be invalidated, as that local was overwritten. This should probably be back-ported to beta. Stable is more iffy, as it's gone unidentified since 1.70, so I only think it's worthwhile if there's another reason for a 1.74.1 release anyway.
rustbot
added
the
beta-accepted
cuviper
removed
the
beta-nominated
[beta] backports - Build pre-coroutine-transform coroutine body rust-lang#117686 - Fix coroutine validation for mixed panic strategy rust-lang#118422 - ConstProp: Remove const when rvalue check fails. rust-lang#118426 - Dispose llvm::TargetMachines prior to llvm::Context being disposed rust-lang#118464 r? ghost
Closes #118328
The problematic sequence of MIR is:
ConstProp::visit_assignhappens on_1 = const 0_usize;, it records that0x0usizeis the value for_1.visit_assignhappens on_1 = const _;. Because the rvalue.has_param(), it can't be const evaled.visit_assignhappens on_0 = _1;. Here it would think the value of_1was0x0usizefrom step 1.The solution is to remove consts when checking the RValue fails, as they may have contained values that should now be invalidated, as that local was overwritten.
This should probably be back-ported to beta. Stable is more iffy, as it's gone unidentified since 1.70, so I only think it's worthwhile if there's another reason for a 1.74.1 release anyway.