-At this point, I ended up in a lengthy discussion with @eddyb and @arielb1, who both know approximately infinitely more about LLVM and rustc than I do, and this is how the third option in the list arose:
+At this point, I ended up in a lengthy discussion with @eddyb and @arielb1 and some folks in #rustc, who know approximately infinitely more about LLVM and rustc than I do, and this is how the third option in the list arose:
When performing `StorageLive` on a variable that already is live, forget the value that is currently in the local variable, but otherwise keep it live.
This is consistent with what we have caught LLVM doing.
It is hard to get any more definite than this.
When performing `StorageLive` on a variable that already is live, forget the value that is currently in the local variable, but otherwise keep it live.
This is consistent with what we have caught LLVM doing.
It is hard to get any more definite than this.
The good news is that with this choice of MIR semantics, miri's test suite passes.
We can thus be sure (well, insofar as the test suite is representative -- this will hopefully get better over time) that rustc produces code that follows our new rules.
The good news is that with this choice of MIR semantics, miri's test suite passes.
We can thus be sure (well, insofar as the test suite is representative -- this will hopefully get better over time) that rustc produces code that follows our new rules.
projects / web.git / blobdiff