Once we move to a copy-on-write implementation of existential value buffers we
can no longer consume or destroy values of an opened existential unless the
buffer is uniquely owned.
Therefore we need to track the allowed operation on opened values.
Add qualifiers "mutable_access" and "immutable_access" to open_existential_addr
instructions to indicate the allowed access to the opened value.
Once we move to a copy-on-write implementation, an "open_existential_addr
mutable_access" instruction will ensure unique ownership of the value buffer.
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.
A change was recently made to canonicalize replacement types in
GenericSignature::getSubstitutions().
This resulted in ParenType being stripped off, which triggered
the 'tuple splat' diagnostic on code that was accepted in Swift 3.0.
I believe this canonicalization step is unnecessary; we
canonicalize using a brand-new ArchetypeBuilder that has no
generic signature added to it, so this is just equivalent to a
call to getCanonicalType().
Also adding the generic signature in question to the builder is
not the right answer either; the replacement types might be
written in terms of a different generic signature, or possibly
in terms of archetypes.
Taking this out seems to have no effect except changing a few
SIL dumps to contain sugared types, which should be harmless.
Part of fixing <rdar://problem/29739905>.
When enumerating requirements, always use the archetype anchors to
express requirements. Unlike "representatives", which are simply there
to maintain the union-find data structure used to track equivalence
classes of potential archetypes, archetype anchors are the
ABI-stable canonical types within a fully-formed generic signature.
The test case churn comes from two places. First, while
representatives are *often* the same as the archetype anchors, they
aren't *always* the same. Where they differ, we'll see a change in
both the printed generic signature and, therefore, it's
mangling.
Additionally, requirement inference now takes much greater
care to make sure that the first types in the requirement follow
archetype anchor ordering, so actual conformance requirements occur in
the requirement list at the archetype anchor---not at the first type
that is equivalent to the anchor---which permits the simplification in
IRGen's emission of polymorphic arguments.
If a generic parameter was substituted for Self, we have to be careful
to not erase Self.Type down to a concrete metatype.
Also, teach the devirtualizer that a metatype of Self type does not
have an exact static type.
Previously, SIL type lowering would assume a MetatypeType was
completely lowered if it had a representation, but this is not
quite right; after substitution, we can have a MetatypeType
whose instance type is a DynamicSelfType. Strip these away more
eagerly, since they show up in SILType::subst(), where we first
substitute AST-level generic parameters, and then lower the
result to get the final SIL type.
Fixes <https://bugs.swift.org/browse/SR-2733>.
Officially kick SILBoxType over to be "nominal" in its layout, with generic layouts structurally parameterized only by formal types. Change SIL to lower a capture to a nongeneric box when possible, or a box capturing the enclosing generic context when necessary.
Use a syntax that declares the layout's generic parameters and fields,
followed by the generic arguments to apply to the layout:
{ var Int, let String } // A concrete box layout with a mutable Int
// and immutable String field
<T, U> { var T, let U } <Int, String> // A generic box layout,
// applied to Int and String
// arguments
This ensures that ownership is properly propagated forward through the use-def
graph.
This was the work that was stymied by issues relating to SILBuilder performing
local ARC dataflow. I ripped out that local dataflow in 6f4e2ab and added a
cheap ARC guaranteed dataflow pass that performs the same optimization.
Also in the process of doing this work, I found that there were many SILGen
tests that were either pattern matching in the wrong functions or had wrong
CHECK lines (for instance CHECK_NEXT). I fixed all of these issues and also
expanded many of the tests so that they verify ownership. The only work I left
for a future PR is that there are certain places in tests where we are using the
projection from an original value, instead of a copy. I marked those with a
message SEMANTIC ARC TODO so that they are easy to find.
rdar://28685236
When DynamicSelfType occurs outside of a class body (for example,
inside of a SIL function), it is not enough to simply utter 'Self',
because then we lose the underlying type.
Instead, print it out as '@dynamic_self Foo', where 'Foo' is the
underlying class type or archetype, and add parser support for
the same.
Fixes <rdar://problem/27735857>.
If a closure captures the dynamic 'Self' type, but no value of type 'Self'
(for example, it is possible to have a weak capture of 'self'; if the weak
reference becomes nil, there's no way for the closure to get the dynamic
'Self' type from the value).
In this case, add a hidden argument of type $Self.Type, and pass in the
Self metatype.
Fixes <https://bugs.swift.org/browse/SR-1558> / <rdar://problem/22299905>.
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.
The SIL parser used PolymorphicFunctionType in two places:
- Internals of SILFunctionType parsing
- Overload selection for class_method / super_method / dynamic_method
instructions
It is better to have Sema construct GenericFunctionType directly
in SIL mode. In particular, the overload selection logic is simpler
now, since it does not have to deal with the fact that
PolymorphicFunctionTypes do not canonicalize.
Mostly NFC, except the SIL printer output is a bit different; for a
generic method on a generic type, the type parameters all come first,
like ``<T><U> G<T> -> (U) -> ()'' -vs- ``<T> G<T> -> <U> (U) -> ()''.
Also, generic constraints look different, instead of ``<`Self` : P>``
we now have ``<Self where Self : P>''.
This patch has two consequences that will become important later:
- While code that constructs PolymorphicFunctionType still exists in
Sema, the SIL parser was the last major component that *consumed*
PolymorphicFunctionType.
- Everywhere we set SILFunction::ContextGenericParams, we now have
a well-formed context GenericSignature available, allowing
ContextGenericParams to be replaced by a GenericSignature
eventually.
And use project_box to get to the address value.
SILGen now generates a project_box for each alloc_box.
And IRGen re-uses the address value from the alloc_box if the operand of project_box is an alloc_box.
This lets the generated code be the same as before.
Other than that most changes of this (quite large) commit are straightforward.
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).
Parameters (to methods, initializers, accessors, subscripts, etc) have always been represented
as Pattern's (of a particular sort), stemming from an early design direction that was abandoned.
Being built on top of patterns leads to patterns being overly complicated (e.g. tuple patterns
have to have varargs and default parameters) and make working on parameter lists complicated
and error prone. This might have been ok in 2015, but there is no way we can live like this in
2016.
Instead of using Patterns, carve out a new ParameterList and Parameter type to represent all the
parameter specific stuff. This simplifies many things and allows a lot of simplifications.
Unfortunately, I wasn't able to do this very incrementally, so this is a huge patch. The good
news is that it erases a ton of code, and the technical debt that went with it. Ignoring test
suite changes, we have:
77 files changed, 2359 insertions(+), 3221 deletions(-)
This patch also makes a bunch of wierd things dead, but I'll sweep those out in follow-on
patches.
Fixes <rdar://problem/22846558> No code completions in Foo( when Foo has error type
Fixes <rdar://problem/24026538> Slight regression in generated header, which I filed to go with 3a23d75.
Fixes an overloading bug involving default arguments and curried functions (see the diff to
Constraints/diagnostics.swift, which we now correctly accept).
Fixes cases where problems with parameters would get emitted multiple times, e.g. in the
test/Parse/subscripting.swift testcase.
The source range for ParamDecl now includes its type, which permutes some of the IDE / SourceModel tests
(for the better, I think).
Eliminates the bogus "type annotation missing in pattern" error message when a type isn't
specified for a parameter (see test/decl/func/functions.swift).
This now consistently parenthesizes argument lists in function types, which leads to many diffs in the
SILGen tests among others.
This does break the "sibling indentation" test in SourceKit/CodeFormat/indent-sibling.swift, and
I haven't been able to figure it out. Given that this is experimental functionality anyway,
I'm just XFAILing the test for now. i'll look at it separately from this mongo diff.
SILPrinter was printing uses for all SIL values, except for SIL basic blocks arguments. Fill the gap and print uses for BB arguments as well. This makes reading and analyzing SIL easier.
Basic blocks may have multiple arguments, therefore print uses of each BB argument on separate lines - one line per BB argument.
The comment containing information about uses of a BB argument is printed on the line just above the basic block name, following the approach used for function_ref and other kinds of instructions, which have additional information printed on the line above the actual instruction.
The output now looks like:
// %0 // user: %3
// %1 // user: %9
bb0(%0 : $Int32, %1 : $UnsafeMutablePointer<UnsafeMutablePointer<Int8>>):
rdar://23336589
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
This makes it clearer that expressions like "foo.myType.init()" are creating new objects, instead of invoking a weird-looking method. The last part of rdar://problem/21375845.
Swift SVN r29375
to not drop optionals in memory all the time. We now generate a lot better code
for them in many cases. This makes generated SIL more readable and should help
-O0 perf.
This is progress towards <rdar://problem/20642198> SILGen shouldn't be dropping optionals into memory all the time
Swift SVN r28102
registers instead of eagerly dumping them in memory and operating on
them by-address. This avoids a lot of temporaries and traffic to
manipulate them.
As part of this, add some new SGF::getOptionalNoneValue/getOptionalSomeValue
helper methods for forming loading optional values.
Many thanks to JoeG for helping with the abstraction difference change in
getOptionalSomeValue.
Swift SVN r27537
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
Try to emit the existential as a guaranteed value, and if we succeed, only +1 the bound opaque value if it's needed as a consumed value. This lets us avoid retaining or copying the existential if the existential can be produced and its contained value consumed at +0.
Swift SVN r27200
This means:
1. In_Guaranteed when preparing accessor base args is like @in not
@inout. This is because @in_guaranteed parameters are immutable. We
were not miscompiling since we were not inserting cleanups for these
parameters. Now with 2, we perform the copy so we have the immutable
property and then destroy_addr the result after the call.
2. If we have a guaranteed parameter, we put the destroy value right
after the call instead of at the end of expression.
The reason 2 is necessary is that if we destroy the value at the end of
scope situations like the following cause COW to fail:
struct Foo {
let object: AnyObject
var rawObject: Builtin.RawPointer {
return Builtin.bridgeToRawPointer(object) // Psuedoname of builtin.
}
mutating func isUnique() -> Bool {
return isUnique(rawObject)
}
}
What happens is that because Foo.isUnique is mutating, Foo is passed in
@inout. Since @inout is a guarantee related to memory, SILGen has to
increment the refcount of self to guarantee self's lifetime. Before this
patch we would have (in pseudo-sil).
%self = load %ptr_self
retain_value %self
%0 = getRawObject() // guaranteed call.
%result = isUnique(%0)
release_value %self
This causes the COW check to always fail. There is no reason to extend
the lifetime of %self so far, guaranteed only means that the object's
lifetime is guaranteed over the call in question. So now instead, we
release after the call.
<rdar://problem/20094305>
<rdar://problem/20234910>
Swift SVN r26351
Leave a cleanup to deinit the container after the uniquely-referenced opaque value is taken out of it. While we're here, stub out support for boxed existentials (though we can't test it since _ErrorType doesn't have any Self-returning methods, and we currently only produce OpenExistentialExprs in the AST for method calls involving covariant Self or metatypes).
Swift SVN r26284
For better consistency with other address-only instruction variants, and to open the door to new exciting existential representations (such as a refcounted boxed representation for ErrorType).
Swift SVN r25902
This lets us disambiguate the symbols for static and instance properties, and enables us to eventually leave the useless "self" type mangling out of method symbols. Fixes rdar://19012022 and dupes thereof, including crasher #1341.
Swift SVN r25111
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
Before this patch there was no dependence visible to the optimizer between a
open_existential and the witness_method allowing the optimizer to reorder the
two instruction. The dependence was implicit in the opened archetype but this
is not a concept model by the SIL optimizer.
%2 = open_existential %0 : $*FooProto to $*@opened("...") FooProto
%3 = witness_method $@opened("...") FooProto,
#FooProto.bar!1 : $@cc(...)
%4 = apply %3<...>(%2)
This patch changes the SIL representation such that witness_methods on opened
archetypes take the open_existential (or the producer of the opened existential)
as an operand preventing the optimizer from reordering them.
%2 = open_existential %0 : $*FooProto to $*@opened("...") FooProto
%3 = witness_method $@opened("...") FooProto,
#FooProto.bar!1,
%2 : $*@opened("...") FooProto : $@cc(...)
%4 = apply %3<...>(%2)
rdar://18984526
Swift SVN r23438
