An assertion I added recently to check property overrides in
the ASTVerifier uncovered some bugs in this area:
- We did not synthesize a materializeForSet for properties
defined in extensions of imported classes unless they
witnessed a protocol requirement.
This triggered an assertion if the property had an
override that was checked before the protocol conformance,
since the override's materializeForSet would not be marked
as an override of the base materializeForSet.
- materializeForSet for properties defined in extensions would
statically dispatch the getter and setter instead of dynamically
dispatching. This is incorrect since we statically dispatch
to the materializeForSet in this case, and we can in fact
override it in a subclass.
Fixes <rdar://problem/31334272>.
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).
This fixes a crash when referencing partially-applied methods
from @_inlineable functions.
Also, curry thunks for private methods do not need shared
linkage; private is sufficient.
Also, add a third [serializable] state for functions whose bodies we
*can* serialize, but only do so if they're referenced from another
serialized function.
This will be used for bodies synthesized for imported definitions,
such as init(rawValue:), etc, and various thunks, but for now this
change is NFC.
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
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>.
Everything here should be NFC after the ownership model eliminator except for 1
change where translation of unowned parameters is made more
correct. Specifically:
1. In manageParam, we make it so that if we allow PlusZero, we begin an actual
begin_borrow, end_borrow sequence. We can do this unconditionally since if the
passed in SILValue is already borrowed, we just return early.
2. In TranslateArguments::translateSingle(), we used to handle owned, unowned,
and guaranteed parameters all the same way. This is of course incorrect. Now we
do the following:
a. If our final translated value is guaranteed, but we want an unowned or
owned parameter, then we perform a copyUnmanaged().
b. If our final translated value is unowned and our argument must be a
guaranteed value, then we first transition the unowned value to an owned value
using SILGen::emitManagedRetain() and then transition from owned to guaranteed
using a emitBeginBorrow().
c. If our final translated value is owned and our argument must be a
guaranteed value, then we perform an emitBeginBorrow().
3. In forwardFunctionArguments(), if our argument requires a guaranteed
argument, we begin a begin borrow sequence.
rdar://29791263
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
Being generic, the '_unwrapped' intrinsics force trafficking through memory, and while they're transparent so always get inlined, we don't do memory promotion in -Onone. Emitting the branch inline lets loadable optionals stay values leading to better -Onone codegen. (It also lets us throw away a surprising amount of support code for these optional intrinsics.)
Similarly to how we've always handled parameter types, we
now recursively expand tuples in result types and separately
determine a result convention for each result.
The most important code-generation change here is that
indirect results are now returned separately from each
other and from any direct results. It is generally far
better, when receiving an indirect result, to receive it
as an independent result; the caller is much more likely
to be able to directly receive the result in the address
they want to initialize, rather than having to receive it
in temporary memory and then copy parts of it into the
target.
The most important conceptual change here that clients and
producers of SIL must be aware of is the new distinction
between a SILFunctionType's *parameters* and its *argument
list*. The former is just the formal parameters, derived
purely from the parameter types of the original function;
indirect results are no longer in this list. The latter
includes the indirect result arguments; as always, all
the indirect results strictly precede the parameters.
Apply instructions and entry block arguments follow the
argument list, not the parameter list.
A relatively minor change is that there can now be multiple
direct results, each with its own result convention.
This is a minor change because I've chosen to leave
return instructions as taking a single operand and
apply instructions as producing a single result; when
the type describes multiple results, they are implicitly
bound up in a tuple. It might make sense to split these
up and allow e.g. return instructions to take a list
of operands; however, it's not clear what to do on the
caller side, and this would be a major change that can
be separated out from this already over-large patch.
Unsurprisingly, the most invasive changes here are in
SILGen; this requires substantial reworking of both call
emission and reabstraction. It also proved important
to switch several SILGen operations over to work with
RValue instead of ManagedValue, since otherwise they
would be forced to spuriously "implode" buffers.
Having a separate address and container value returned from alloc_stack is not really needed in SIL.
Even if they differ we have both addresses available during IRGen, because a dealloc_stack is always dominated by the corresponding alloc_stack in the same function.
Although this commit quite large, most changes are trivial. The largest non-trivial change is in IRGenSIL.
This commit is a NFC regarding the generated code. Even the generated SIL is the same (except removed #0, #1 and @local_storage).
And include some supplementary mangling changes:
- Give the first generic param (depth=0, index=0) a single character mangling. Even after removing the self type from method declaration types, 'Self' still shows up very frequently in protocol requirement signatures.
- Fix the mangling of generic parameter counts to elide the count when there's only one parameter at the starting depth of the mangling.
Together these carve another 154KB out of a debug standard library. There's some awkwardness in demangled strings that I'll clean up in subsequent commits; since decl types now only mangle the number of generic params at their own depth, it's context-dependent what depths those represent, which we get wrong now. Currying markers are also wrong, but since free function currying is going away, we can mangle the partial application thunks in different ways.
Swift SVN r32896
Canonical dependent member types are always based from a generic parameter, so we can use a more optimal mangling that assumes this. We can also introduce substitutions for AssociatedTypeDecls, and when a generic parameter in a signature is constrained by a single protocol, we can leave that protocol qualification out of the unsubstituted associated type mangling. These optimizations together shrink the standard library by 117KB, and bring the length of the longest Swift symbol in the stdlib down from 578 to 334 characters, shorter than the longest C++ symbol in the stdlib.
Swift SVN r32786
'Ss' appears in manglings tens of thousands of times in the standard library and is also incredibly frequent in other modules. This alone is enough to shrink the standard library by 59KB.
Swift SVN r32409
This allows SILGen to recognize them as foreign declarations, fixing a bug when classes with subscripts defined in Objective-C were extended to conform to Swift protocols with subscript requirements. rdar://problem/20371661
Swift SVN r27998
The only caveat is that:
1. We do not properly recognize when we have a let binding and we
perform a guaranteed dynamic call. In such a case, we add an extra
retain, release pair around the call. In order to get that case I will
need to refactor some code in Callee. I want to make this change, but
not at the expense of getting the rest of this work in.
2. Some of the protocol witness thunks generated have unnecessary
retains or releases in a similar manner.
But this is a good first step.
I am going to send a large follow up email with all of the relevant results, so
I can let the bots chew on this a little bit.
rdar://19933044
Swift SVN r27241
We no longer need or use it since we can always refer to the same bit on
the applied function when deciding whether to inline during mandatory
inlining.
Resolves rdar://problem/19478366.
Swift SVN r26534
This will have an effect on inlining into thunks.
Currently this flag is set for witness thunks and thunks from function signature optimization.
No change in code generation, yet.
Swift SVN r24998
Most tests were using %swift or similar substitutions, which did not
include the target triple and SDK. The driver was defaulting to the
host OS. Thus, we could not run the tests when the standard library was
not built for OS X.
Swift SVN r24504
as passing self by value, not by inout. This is the correct representation at
the AST level, and we now lower self references as the new @in_guaranteed
parameter convention. This allows SIL clients (like DI) to know that a nonmutating
protocol method does not mutate the pointee passed into the method.
This fixes:
<rdar://problem/19215313> let properties don't work with protocol method dispatch
<rdar://problem/15821762> Self argument of generic curried nonmutating instance methods is inout
Swift SVN r23864
Doing so is safe even though we have mock SDK. The include paths for
modules with the same name in the real and mock SDKs are different, and
the module files will be distinct (because they will have a different
hash).
This reduces test runtime on OS X by 30% and brings it under a minute on
a 16-core machine.
This also uncovered some problems with some tests -- even when run for
iOS configurations, some tests would still run with macosx triple. I
fixed the tests where I noticed this issue.
rdar://problem/19125022
Swift SVN r23683
Teaches TryAddFinal to descend into public and objc classes. Only
tries to add final if we're either whole compilation mode, or we're
processing the primary source buffer.
Updates test cases. Includes workarounds for <rdar://problem/17860781>
and <rdar://problem/17862348>.
Swift SVN r20790
This ensures that if the witness is a foreign-to-native, curry, or other lazily-generated thunk, that the thunk gets generated. This fixes natively-ObjC witnesses to Swift protocols.
Swift SVN r17739
