[move-only] Implement very initial work for move only #60187
Conversation
Currently it doesn't have any effect.
…ve only type". I also created a SILType::isMoveOnly() helper that returns true if a type is a move only wrapped type or a first class move only type. The verifier check that move only types aren't copied in canonical SIL was rewired to use that as well.
This is just the very beginning... I still need to implement more parts of SILGen for this. But all great things start small. I am going to iterate on top of this and just wanted to get some initial parts of the work in as I go.
|
@swift-ci test |
| // We are pattern matching against these patterns: | ||
| // | ||
| // bb0(%0 : @guaranteed $T): | ||
| // %1 = copy_value %0 |
There was a problem hiding this comment.
@gottesmm if an argument is marked @moveOnly, then we need to be sure no copy_value exists in the argument, at least by the time this diagnostic runs. This is true for both owned and guaranteed.
There was a problem hiding this comment.
I talked with Andy about this, what actually happens here is that as part of the diagnostic I eliminate the copy_value and the mark_must_check. I did it this way since today copy propagation doesn't support guaranteed parameters (although we could extend it, I decided to just hack around it).
There was a problem hiding this comment.
The right thing here is I think to just see if we can get copy propagation working on guaranteed things and rip this out.
There was a problem hiding this comment.
@gottesmm for future reference, copy propagation always handled guaranteed function arguments. It's done in RewriteInnerBorrowUses and CanonicalizeBorrowScope::visitBorrowScopeUses.
|
|
||
| // CHECK-LABEL: sil [ossa] @$s8moveonly15useKlassConsumeyyAA0C0CnF : $@convention(thin) (@owned Klass) -> () { | ||
| // CHECK: bb0([[ARG:%.*]] : @owned $Klass): | ||
| // TODO: Is this move_value [lexical] necessary? |
There was a problem hiding this comment.
TODO: Is this move_value [lexical] necessary?
The move_value is not necessary. Function arguments have implicit "lexical" scope
There was a problem hiding this comment.
I did this b/c I was worried that we had baked in this notion that owned arguments needed an implicit lexical scope. This can be seen by how we for other cases insert a begin_borrow [lexical] on the owned argument.
There was a problem hiding this comment.
I am fine fixing this in a follow on commit.
I need to add more tests/expand it to more of SILGen. But for today, this is enough to get started/be incremental. Of course this is behind the experimental move only flag.