Remove the field field from Lvalue::Local
#220
Conversation
src/eval_context.rs
Outdated
| @@ -1353,7 +1347,12 @@ impl<'a, 'tcx> EvalContext<'a, 'tcx> { | |||
| I128 => 16, | |||
| Is => self.memory.pointer_size(), | |||
| }; | |||
| PrimVal::from_i128(self.memory.read_int(ptr, size)?) | |||
| // if we cast a ptr to a usize reading it back into a primval shouldn't panic | |||
There was a problem hiding this comment.
Whoops, copy paste.
src/eval_context.rs
Outdated
| @@ -1366,7 +1365,12 @@ impl<'a, 'tcx> EvalContext<'a, 'tcx> { | |||
| U128 => 16, | |||
| Us => self.memory.pointer_size(), | |||
| }; | |||
| PrimVal::from_u128(self.memory.read_uint(ptr, size)?) | |||
| if size == self.memory.pointer_size() { | |||
| // if we cast a ptr to a usize reading it back into a primval shouldn't panic | |||
There was a problem hiding this comment.
This is the usize cast ;)
| @@ -24,8 +24,6 @@ pub enum Lvalue<'tcx> { | |||
| Local { | |||
| frame: usize, | |||
| local: mir::Local, | |||
| /// Optionally, this lvalue can point to a field of the stack value | |||
| field: Option<(usize, Ty<'tcx>)>, | |||
There was a problem hiding this comment.
Yay. :D TBH I never understood this field, so I'm happy I don't have to.
Looks like |
|
@RalfJung Looks like |
|
Ah, GitHub fooled me. I pressed the "expand code" button and then what Is aw looked almost like a complete type, but it wasn't actually completely expanded. :/ |
oli-obk
changed the title
field field from Lvalue::Local
|
|
|
The |
|
I checked whether we could use the |
| @@ -1353,7 +1347,12 @@ impl<'a, 'tcx> EvalContext<'a, 'tcx> { | |||
| I128 => 16, | |||
| Is => self.memory.pointer_size(), | |||
| }; | |||
| PrimVal::from_i128(self.memory.read_int(ptr, size)?) | |||
| // if we cast a ptr to an isize, reading it back into a primval shouldn't panic | |||
There was a problem hiding this comment.
I am not sure I am happy about using the word "cast" here. This is doing a transmute, i.e., it is reading the same bytes at a different type. The rules for the two should be somewhat consistent, but they are clearly not the same right now (as they are implemented in two different files).
|
Fair enough. |
Undoes #146 and #139 because this was just a lot of overhead working with the field accesses. They also only "helped" when the lvalue that was produced was immediately read again.
We should move to doing an lvalue-read operation similar to https://github.com/solson/miri/pull/139/files#diff-3596aeeadedc2ed352f2c376852f8fb7L117 which optimizes through those cases where you don't want to have a pointer in the end, but only a primval. So basically primval rvalues?
I did not reinstate the mentioned lvalue-read, since i want to get this PR through first and then write a more advanced version that supports more cases.