We weren't looking at the length of an opaque archetype's type parameter,
which could lead to unbounded growth in the number of emitted specializations.
Fixes rdar://problem/121867690.
This adds SIL-level support and LLVM codegen for normal results of a coroutine.
The main user of this will be autodiff as VJP of a coroutine must be a coroutine itself (in order to produce the yielded result) and return a pullback closure as a normal result.
For now only direct results are supported, but this seems to be enough for autodiff purposes.
I thought that we would never actually hit this test case, but since we do not
have opaque values this pattern is treated like a projection in certain cases. I
updated the code as appropriate and added a test.
We should make this as a transfer... but this is just temporary until I get
around to it. For now this is always an out parameter, so we should be safe.
I was correct that it is lookthrough, but it has multiple parameters, so the
normal "baked" implementation cannot be used since the "baked" implementation
assumes that any insts it handles has a single operand.
A name diagnostic looks as follows:
Some notes:
1. VariableNameUtils still doesn't pattern match all cases. In the cases where
we fail to pattern match, I fall back to the old diagnostics. I am going to be
adding tests/etc specific to VariableNameUtils in a later patch.
2. I took this as an opportunity to begin preparing to change the diagnostics
into diags of the following form:
a. The main diagnostic is changed to something short "transferring non-Sendable
%0 could yield races with later accesses".
b. I added a longer note at the same location that explains that our caller is
in a specific isolation domain and our callee is isolated differently.
c. I added a note on the definition location of the identified value.
I am going to reuse this for TransferNonSendable. In the process I made a few
changes to make it more amenable to both use cases and used the current set of
tests that we have for noncopyable types to validate that I didn't break
anything.
Decls with a package access level are currently set to public SIL
linkages. This limits the ability to have more fine-grained control
and optimize around resilience and serialization.
This PR introduces a separate SIL linkage and FormalLinkage for
package decls, pipes them down to IRGen, and updates linkage checks
at call sites to include package linkage.
Resolves rdar://121409846
Unreachable blocks possess some challenges to autodiff since in reverse pass (pullback generation) we need to execute the function backwards, pushing the values from return BB back to entry block. As a result, unreachable blocks might become reachable from the return BB and this might cause all kind of issues.
I also made ref_to_unmanaged and unmanaged_to_ref look through. They should have
always been look through, but I believe early on they were marked assign and I
never touched fixed it. To get the test for strong_copy_unmanaged_value to work,
I needed them to obey these semantics so I fixed them at the same time.
