Support for @noescape SILFunctionTypes.
These are the underlying SIL changes necessary to implement the new
closure capture ABI.
Note: This includes a change to function name mangling that
primarily affects reabstraction thunks.
The new ABI will allow stack allocation of non-escaping closures as a
simple optimization.
The new ABI, and the stack allocation optimization, also require
closure context to be @guaranteed. That will be implemented as the
next step.
Many SIL passes pattern match partial_apply sequences. These all
needed to be fixed to handle the convert_function that SILGen now
emits. The conversion is now needed whenever a function declaration,
which has an escaping type, is passed into a @NoEscape argument.
In addition to supporting new SIL patterns, some optimizations like
inlining and SIL combine are now stronger which could perturb some
benchmark results.
These underlying SIL changes should be merged now to avoid conflicting
with other work. Minor benchmark discrepancies can be investigated as part of
the stack-allocation work.
* Add a noescape attribute to SILFunctionType.
And set this attribute correctly when lowering formal function types to SILFunctionTypes based on @escaping.
This will allow stack allocation of closures, and unblock a related ABI change.
* Flip the polarity on @noescape on SILFunctionType and clarify that
we don't default it.
* Emit withoutActuallyEscaping using a convert_function instruction.
It might be better to use a specialized instruction here, but I'll leave that up to Andy.
Andy: And I'll leave that to Arnold who is implementing SIL support for guaranteed ownership of thick function types.
* Fix SILGen and SIL Parsing.
* Fix the LoadableByAddress pass.
* Fix ClosureSpecializer.
* Fix performance inliner constant propagation.
* Fix the PartialApplyCombiner.
* Adjust SILFunctionType for thunks.
* Add mangling for @noescape/@escaping.
* Fix test cases for @noescape attribute, mangling, convert_function, etc.
* Fix exclusivity test cases.
* Fix AccessEnforcement.
* Fix SILCombine of convert_function -> apply.
* Fix ObjC bridging thunks.
* Various MandatoryInlining fixes.
* Fix SILCombine optimizeApplyOfConvertFunction.
* Fix more test cases after merging (again).
* Fix ClosureSpecializer. Hande convert_function cloning.
Be conservative when combining convert_function. Most of our code doesn't know
how to deal with function type mismatches yet.
* Fix MandatoryInlining.
Be conservative with function conversion. The inliner does not yet know how to
cast arguments or convert between throwing forms.
* Fix PartialApplyCombiner.
The interface type for materializeForSet does not mention the
element type of a subscript, so Sema was not inferring
requirements that were implied by the element type but not
explicitly stated in the generic signature.
This led to assertion failures in SILGen because the
materializeForSet calls the getter and setter, which does
have these requirements.
Fixes <rdar://problem/33219034>, <https://bugs.swift.org/browse/SR-5420>.
We used to give witness thunks public linkage if the
conforming type and the protocol are public.
This is completely unnecessary. If the conformance is
fragile, the thunk should be [shared] [serialized],
allowing the thunk to be serialized into callers after
devirtualization.
Otherwise for private protocols or resilient modules,
witness thunks can just always be private.
This should reduce the size of compiled binaries.
There are two other mildly interesting consequences:
1) In the bridged cast tests, we now inline the witness
thunks from the bridgeable conformances, which removes
one level of indirection.
2) This uncovered a flaw in our accessibility checking
model. Usually, we reject a witness that is less
visible than the protocol; however, we fail to
reject it in the case that it comes from an
extension.
This is because members of an extension can be
declared 'public' even if the extended type is not
public, and it appears that in this case the 'public'
keyword has no effect.
I would prefer it if a) 'public' generated a warning
here, and b) the conformance also generated a warning.
In Swift 4 mode, we could then make this kind of
sillyness into an error. But for now, live with the
broken behavior, and add a test to exercise it to ensure
we don't crash.
There are other places where this "allow public but
ignore it, kinda, except respect it in some places"
behavior causes problems. I don't know if it was intentional
or just emergent behavior from general messiness in Sema.
3) In the TBD code, there is one less 'failure' because now
that witness thunks are no longer public, TBDGen does not
need to reason about them (except for the case #2 above,
which will probably require a similar workaround in TBDGen
as what I put into SILGen).
We had some non-deterministic behavior where depending on
validation order, synthesized accessors would end up in
different places because we would sometimes just add them
at the end of the member list.
Now add the getter right after the storage, the setter
right after the getter and the materializeForSet right
after the setter.
This changes some test output where the declaration order
did not make sense before but should otherwise have no
functional effect.
Instead of appending a character for each substitution, we now prefix the substitution with the repeat count, e.g.
AbbbbB -> A5B
The same is done for known-type substitutions, e.g.
SiSiSi -> S3i
This significantly shrinks mangled names which contain large lists of the same type, like
func foo(_ x: (Int, Int, Int, Int, Int, Int, Int, Int, Int, Int, Int, Int))
rdar://problem/30707433
First, use the correct generic environment to compute the substituted
storage type. Substitutions derived from 'self' are not enough,
because we also want the archetypes of the generic subscript's
innermost generic parameters.
Also, use the method and witness_method calling conventions for the
materializeForSet callback, depending on if we have a protocol
witness or concrete implementation.
Since the materializeForSet callback is called with a more
abstract type at the call site than the actual function type
of the callback, we used to rely on these two SIL types being
ABI compatible:
@convention(thin) <Self : P, T, U) (..., Self.Type) -> ()
@convention(thin) <T, U> (..., Foo<T, U>.Type) -> ()
The IRGen lowering is roughly the following -- the call site
passes two unused parameters, but that's fine:
(..., Self.Type*, Self.Type*, Self.P*)
(..., Foo<T, U>.Type*)
However if the callback has its own generic parameters because
the subscript is generic, we might have SIL types like so,
@convention(thin) <Self : P, T, U, V) (..., Self.Type) -> ()
@convention(thin) <T, U, V> (..., Foo<T, U>.Type) -> ()
And the IRGen lowering is the following:
(..., Self.Type*, Self.Type*, Self.P*, V.Type*)
(..., Foo<T, U>.Type*, V.Type*)
The parameters no longer line up, because the caller still passes
the two discarded arguments, and type metadata for V cannot be
derived from the Self metadata so must be passed separately.
The witness_method calling convention is designed to solve this
problem; it puts the Self metadata and protocol conformance last,
so if you have these SIL types:
@convention(witness_method) <Self : P, T, U, V) (..., swiftself Self.Type) -> ()
@convention(witness_method) <T, U, V> (..., swiftself Foo<T, U>.Type) -> ()
The IRGen lowering is the following:
(..., Self.Type*, V.Type*, Self.Type*, Self.P*)
(..., Foo<T, U>.Type*, V.Type*, Self.Type*, unused i8*)
However, the problem is now that witness_method and thin functions
are not ABI compatible, because thin functions don't have a
distinguished 'self', which is passed differently in LLVM's swiftcc
calling convention:
@convention(witness_method) <Self : P, T, U, V) (..., Self.Type) -> ()
@convention(thin) <T, U, V> (..., Foo<T, U>.Type) -> ()
So instead of using 'thin' representation for the concrete callback
case, use 'method', which is essentially the same as 'thin' except if
the last parameter is pointer-size, it is passed as the 'self' value.
This makes everything work out.
We can get the generic signature from the generic environment
now, and for generic subscript protocol witnesses, using the
signature of the conformance is wrong; it won't have the
generic parameters of the subscript itself.
Also, emit the materializeForSet callback in the right place in
SILModule. Instead of adding it at the end, put it before the
materializeForSet itself. This makes tests a bit easier to write.
Previously, we were emitting these cleanups at the end of the lexical scope
instead of at the end of the formal evaluation scope. This change ensures that
we always emit the cleanup immediately at the end of the formal evaluation
scope.
Previously in most cases we got away with this due to the +0 self
hack. Basically we would emit a get for a self parameter and then immediately
use that self parameter as a guaranteed parameter. Then the hack would insert
the destroy value forwarding the lexical scope level cleanup at the same time.
rdar://29791263
In 74d979f0ac, the policy was changed
so that only value type accessors are ever marked transparent, and
not class accessors.
This was intended to fix a bug where inlining an accessor of an
Objective-C-derived class across module boundaries caused a linker
failure because the accessor referenced a field offset variable,
which has hidden visibility.
However, this also caused a performance regression for Swift native
classes. Bring back the old behavior for Swift native classes in
non-resilient modules.
Fixes <rdar://problem/29884727>.
The problem is that the derived property's materializeForSet was
being synthesized after recordOverride() was called, so the new
accessor never got setOverridenDecl() called on it.
As a result, SIL didn't know that the derived materializeForSet
should replace the vtable entry for the base class materializeForSet.
The more fundamental problem here is that even after some recent
cleanups, accessors are still sometimes type checked before
the AbstractStorageDecla and sometimes after. So things like
inferring final, dynamic and overrides have to be duplicated in
multiple places.
Fixes <https://bugs.swift.org/browse/SR-3840> and
<rdar://problem/30336146>.
The problem is that the derived property's materializeForSet was
being synthesized after recordOverride() was called, so the new
accessor never got setOverridenDecl() called on it.
As a result, SIL didn't know that the derived materializeForSet
should replace the vtable entry for the base class materializeForSet.
The more fundamental problem here is that even after some recent
cleanups, accessors are still sometimes type checked before
the AbstractStorageDecla and sometimes after. So things like
inferring final, dynamic and overrides have to be duplicated in
multiple places.
Fixes <https://bugs.swift.org/browse/SR-3840> and
<rdar://problem/30336146>.
Textual SIL was sometimes ambiguous when SILDeclRefs were used, because the textual representation of SILDeclRefs was the same for functions that have the same name, but different signatures.
Textual SIL was sometimes ambiguous when SILDeclRefs were used, because the textual representation of SILDeclRefs was the same for functions that have the same name, but different signatures.
Keep in mind that these are approximations that will not impact correctness
since in all cases I ensured that the SIL will be the same after the
OwnershipModelEliminator has run. The cases that I was unsure of I commented
with SEMANTIC ARC TODO. Once we have the verifier any confusion that may have
occurred here will be dealt with.
rdar://28685236
By moving the is-setter-newValue logic into a TypeCheckOption, we
allow more local reasoning inside of resolve type and can plumb the
special knowledge directly into applyNonEscapingFromContext. This is a
little cleaner, but mainly has the advantage that it will keep future
refactorings of the AST and escapability more localized.
Added test cases for SILGen as well, to make sure we don't crash
(enforces getter/setter escapability parity).
From the Swift documentation:
"If you define an optional variable without providing a default value,
the variable is automatically set to nil for you."
This flips the switch to have @noescape be the default semantics for
function types in argument positions, for everything except property
setters. Property setters are naturally escaping, so they keep their
escaping-by-default behavior.
Adds contentual printing, and updates the test cases.
There is some further (non-source-breaking) work to be done for
SE-0103:
- We need the withoutActuallyEscaping function
- Improve diagnostics and QoI to at least @noescape's standards
- Deprecate / drop @noescape, right now we allow it
- Update internal code completion printing to be contextual
- Add more tests to explore tricky corner cases
- Small regressions in fixits in attr/attr_availability.swift
Strict aliasing only applies to memory operations that use strict
addresses. The optimizer needs to be aware of this flag. Uses of raw
addresses should not have their address substituted with a strict
address.
Also add Builtin.LoadRaw which will be used by raw pointer loads.
We need to wrap SelfInterfaceType and SubstSelfType in a metatype.
Previously this was only done correctly when emitting concrete
method implementations of materializeForSet.
The function pointer is a thin function and possibly polymorphic,
so it does not really have an AST type. Instead of pretending it has
an AST type, just return a RawPointer and remove some casts in the
process.
We don't really need to mangle the AST type of the callback, besides
it doesn't have an AST type at all, because it is polymorphic.
Also, mangle the closure as if it were parented by the requirement
and not the witness, for consistency with witness thunks.
The mangling should still be unique since it includes the conformance.
NFC other than updating tests for new mangling.
When we emit calls to getters and setters with emitApply(), the call
emission code ends up re-abstracting the result of a getter or the
parameter to a setter.
As a result, getOrigFormalType() was incorrect for logical path
components. This was worked around by only adding an OrigToSubst
path component to an lvalue if the last path component was physical.
This caused a problem in the following case:
1) There was an abstraction difference between the storage of the
protocol requirement and the storage of the protocol witness
2) There was an abstraction difference between the storage of the
protocol witness, and the fully-substituted type of the witness
An example is when the witness is in a protocol extension, and uses
'Self' or some other associated type which is concrete and loadable
in the conformance.
Fix this properly by splitting up getStorageTypeData() into two
functions, one used when adding physical path components and another
one used for logical.
As a result, we can now give the correct abstraction pattern to
logical path components.
