C++: Placeholder CFG for coroutines #16187
Conversation
| // TODO: Use a new opcode to represent "awaiting the value" | ||
| expr instanceof CoAwaitExpr and result instanceof Opcode::CopyValue | ||
| or | ||
| // TODO: Use a new opcode to represent "awaiting the value" |
There was a problem hiding this comment.
Shouldn't this be:
| // TODO: Use a new opcode to represent "awaiting the value" | |
| // TODO: Use a new opcode to represent "yielding a value" |
?
There was a problem hiding this comment.
Actually no. co_yield is just syntactic sugar for:
co_await promise.yield_value(expr)(see here) and the "yielding" is done by the call to yield_value.
So whatever instruction we come up with to use the above TODO should also be used here since a co_yield ultimately does something equivalent to a special kind of co_await.
| #-----| m0_9(promise_type) = Chi : total:m87_14, partial:m0_8 | ||
| # 88| v88_1(void) = NoOp : | ||
| #-----| v0_10(void) = NoOp : | ||
| #-----| Goto (back edge) -> Block 1 |
There was a problem hiding this comment.
I'm somewhat confused by this. Why are there two blocks here and why is this Goto marked as a back edge?
There was a problem hiding this comment.
Oh, I didn't spot this either. I'm also confused by this. Will investigate!
There was a problem hiding this comment.
I should've checked the raw IR before opening this PR >.< The IR is still very messed up! I'll fix that
There was a problem hiding this comment.
Aha! I've found the problem:
private predicate foo(UnaryOperation uo) {
not exists(uo.getOperand())
}has 4 results in the co_return_void function 🤔 So I think there's some extractor issue going on here.
There was a problem hiding this comment.
I've created an issue for this now.
| @@ -1501,3 +1501,41 @@ class TranslatedVlaDeclarationStmt extends TranslatedStmt { | |||
|
|
|||
| override Instruction getChildSuccessorInternal(TranslatedElement child, EdgeKind kind) { none() } | |||
| } | |||
|
|
|||
| class TranslatedCoReturnStmt extends TranslatedStmt { | |||
| override CoReturnStmt stmt; | |||
There was a problem hiding this comment.
I did not know you can do this (override the type of a member variable in QL).
There was a problem hiding this comment.
It's extremely useful! We do it a lot in both C/C++ IR construction and in the Swift CFG construction code
This PR adds CFG support for C++20 coroutines. It's basically complete, except for the fact that we don't yet have an instruction to represent "awaiting a value". So instead we're simply emitting a
NoOpinstruction for now.Eventually, that
NoOpinstruction should be replaced with a new instruction, and subsequent analyses should decide what to do with it. (For example, dataflow should use field flow to ensure that yielded values propagate to the caller.)Commit-by-commit review recommended.