[4.1] SIL: Fix zealous assert in verifier #14134
Conversation
A materialize for set / write-back sequence is not really mutating the opened existential rdar://36799163
|
@swift-ci Please test |
| // Non-Mutating set pattern that allows a inout (that can't really | ||
| // write back. | ||
| if (auto *AI = dyn_cast<ApplyInst>(inst)) { | ||
| if (isa<PointerToThinFunctionInst>(AI->getCallee())) { |
There was a problem hiding this comment.
I don't get it. What does the function representation have to do with consuming-or-mutating arguments? I think this warrants a little more explanation.
|
The only producer of this instruction is SILGen when it generates a materialize for set. |
|
I will follow-up with an updated comment on the master branch. |
|
Build failed |
|
Build failed |
|
@swift-ci Please test |
|
@swift-ci Please test source compatibility |
|
Build failed |
|
Build failed |
|
@swift-ci Please test |
|
Build failed |
|
Build failed |
|
@swift-ci please nominate |
A materialize for set / write-back sequence is not really mutating the opened existential on an Rvalue existential.
rdar://36799163
Explanation: We generate a write back for a non-mutating set on an r-value. This write back looks like a mutation on an open_existenial that only allows immutable access. The verifier of open_existential instructions will look at the use and complain when this use really is safe. The patch disable verification in this case.
Scope: A fix that makes the SIL verifier not assert on a correct code pattern.
Risk: Low. This disables an invalid assertion.
Testing: A Swift CI test was added