If a local function tried to reference a local typealias, we were trying to emit a SIL variable reference and freaking out, causing <rdar://problem/15360605>. For now, don't emit any capture arguments for local typealiases. Any context a local typealias requires should be handled by capturing the external context's generic parameters.
Swift SVN r9865
Whatever kind of Swift decl we cons up for a Clang enum, add it to the externals list so we can pick it up and emit Swift metadata for it in IRGen. Fixes <rdar://problem/15242452>.
Swift SVN r9801
so they can be easily identified by the inout deshadowing pass. This does
not mark the alloc_box as autogenerated though, since it is a user variable
when the variable escapes.
- Fix some wonky indentation in SILLocation.
Swift SVN r9700
Introduces a new flag in SILLocation: InPrologue to mark instructions
that setup the stack and allocate storage for local variables/arguments.
Fixes rdar://problem/15290023: Breakpoint set on prologue - crash ensues.
Swift SVN r9686
Pull the implicit 'Self' associated type out of the protocol and into
an implicitly-declared generic parameter list for the protocol. This
makes all of the methods of a protocol polymorphic, e.g., given
protocol P {
typealias Assoc
func getAssoc() -> Assoc
}
the type of P.getAssoc is:
<Self : P> (self : @inout P) -> () -> Self.Assoc
This directly expresses the notion that protocol methods are
polymorphic, even though 'Self' is always implicitly bound. It can be
used to simplify IRgen and some parts of the type checker, as well as
laying more of the groundwork for default definitions within
protocols as well as sundry other improvements to the generics
system.
There are a number of moving parts that needed to be updated in tandem
for this. In no particular order:
- Protocols always get an implicit generic parameter list, with a
single generic parameter 'Self' that conforms to the protocol itself.
- The 'Self' archetype type now knows which protocol it is
associated with (since we can no longer point it at the Self
associated type declaration).
- Protocol methods now get interface types (i.e., canonicalizable
dependent function types).
- The "all archetypes" list for a polymorphic function type does not
include the Self archetype nor its nested types, because they are
handled implicitly. This avoids the need to rework IRGen's handling
of archetypes for now.
- When (de-)serializing a XREF for a function type that has an
interface type, use the canonicalized interface type, which can be
meaningfully compared during deserialization (unlike the
PolymorphicFunctionType we'd otherwise be dealing with).
- Added a SIL-specific type attribute @sil_self, which extracts the
'Self' archetype of a protocol, because we can no longer refer to
the associated type "P.Self".
Swift SVN r9066
Rewrite ForEachStmt SILGen to use the Optional intrinsics with the Generator.next method to iterate through sequences, and kill off the Enumerator path in Sema. Cut over 'EnumeratorType.Element' requirements to instead require 'GeneratorType.Element' in the stdlib.
There are a couple of bugs remaining that need follow-up work. There appears to be a bug in nested enum layout (e.g. T??) that's causing test/Interpreter/enum to break; I'll investigate and fix. There's also a lingering type-checker bug with inferred associated types that causes them to fail requirement checks <rdar://problem/15172101>, which I think Doug needs to look into.
Swift SVN r9017
Chris and I want to move toward canonicalizing on more abstract aggregate operations (copy_addr) instead of on the component load/store/copy|destroy_value/retain|release operations, which is easier for early passes like inout deshadowing, NRVO, and move optimization to reason about. As a first step, replace the handful of places where SILGen currently used TypeLowering::emitCopyInto with simple CopyAddrInst insertions. This affects inout initializations and emitSemanticLoadInto, neither of which should disturb early passes that relied on the old behavior.
Swift SVN r8991
preceding copy_addr instruction when totally trivial. Adopt this in SILGen, eliminating
a couple dozen destroy_addr instructions from the stdlib and producing more canonical SIL.
Swift SVN r8968
When we walk a ClassDecl, generate its vtable, first pulling in decls from its ancestor classes, then overlaying overridden or new decls as we discover them.
Swift SVN r8947
Make ApplyInst and PartialApplyInst directly take substitutions for generic functions instead of trying to stage out substitutions separately. The legacy reasons for doing this are gone.
Swift SVN r8747
These are the terms sent out in the proposal last week and described in
StoredAndComputedVariables.rst.
variable
anything declared with 'var'
member variable
a variable inside a nominal type (may be an instance variable or not)
property
another term for "member variable"
computed variable
a variable with a custom getter or setter
stored variable
a variable with backing storage; any non-computed variable
These terms pre-exist in SIL and IRGen, so I only attempted to solidify
their definitions. Other than the use of "field" for "tuple element",
none of these should be exposed to users.
field
a tuple element, or
the underlying storage for a stored variable in a struct or class
physical
describes an entity whose value can be accessed directly
logical
describes an entity whose value must be accessed through some accessor
Swift SVN r8698
Doug pointed out that 'isObjC' incorrectly excludes C functions, for which we'll also need to be able to independently reference Swift and foreign entries.
Swift SVN r8669
Create a local box for byref arguments that can be closed over. The box receives the value of the byref at function entry, and the value in the box at function exit is written back to the byref. This will allow us to eliminate the no-capture restriction on byrefs, allowing them to be used in auto closures and other HOFs, and doing the right thing in the 99% case where closures are effectively nocapture without requiring user annotation. <rdar://problem/14732389>
There are a couple of loose ends:
- The type-checker can lose its restriction on byref captures. I'll do that next.
- Debug info needs to learn how to represent local byref boxes. Adrian is better qualified to decide how to do that.
- Byref shadow copies completely defeat the copy-on-write optimization for String, making append sequences ridiculously slow. Memory promotion needs to learn to eliminate shadow copies completely when it promotes away byref boxes in order to solve this regression.
Swift SVN r8630
Introduce an EnumCaseDecl for source fidelity to track the 'case' location and ordering of EnumElementDecls. Parse a comma-separated list of EnumElementDecls after a 'case' token.
Swift SVN r8509
Improve the type checker to create implicit DestructorDecls, tighten the
assertion in ImplicitReturnLocation::getImplicitReturnLoc(), and add a verifier
check that a class in a type checked AST always has exactly one destructor.
SILGen used to generate a destructor if the class does not have a
DestructorDecl. SILGen used to put the ClassDecl inside the SILLocation for
the destructor SIL code. This is not a very clean solution: in this case
ImplicitReturnLocation SILLocations contain ClassDecl, which is surprising.
rdar://14970972 Implicit destructors should have AST nodes
Swift SVN r8498
Now that all [objc] constructors have a suitable "init" selector, emit
an Objective-C thunk for the initializing constructor (not the
allocating constructor!) with that selector, and make sure it shows up
in the Objective-C metadata.
With this, we can write a Swift constructor "constructor()" to
override "-init"; see the change to ListMaker that exercises this.
Joe or John: I'd love a review of these.
Swift SVN r8373
AnyFunctionRef is a universal function reference that can wrap all AST nodes
that represent functions and exposes a common interface to them. Use it in two
places in SIL where CapturingExpr was used previously.
AnyFunctionRef allows further simplifications in other places, but these will
be done separately.
Swift SVN r8239
When enabled, this causes references to global variables in the
top_level_code function to be emitted as:
%0 = global_addr #x : $*Int64
%1 = mark_uninitialized %0 : $*Int64
where all local uses of "x" use the %1 value. This representation will
allow the definite initialization pass to know that "x" must be forced
to be initialized before the function returns or the value escapes (which
we don't have a representation for yet).
Swift SVN r8147
binding decls for global variables, enter the global_addr into the
VarLocs mapping so that the global_addrs are CSE'd in the generated
SIL code. For now, this simply reduces the size of the generated
SIL (by reducing some common subexpressions) with no other
functionality change.
Swift SVN r8146
ObjC methods are already tagged with a special calling convention and have special IRGen handling to keep the _cmd argument abstracted away from SIL. We can use the CC to also abstract away the detail that Swift methods pass 'self' last but ObjC methods pass 'self' first. This eliminates a weird special case from SIL's perspective, and also means that 'partial_apply' can work on objc methods correctly without becoming significantly more complex.
Swift SVN r8091
ConstructorDecl::getBody() and DestructorDecl::getBody() return 'BraceStmt *'.
After changing the AST representation for functions, FuncDecl::getBody() will
return 'BraceStmt *' and FuncDecl::getFuncExpr() will be gone.
Swift SVN r8050
global variables used by functions in the capture list as well.
SILGen and other things that don't care about these (i.e., all
current current clients) filter the list to get what they want.
This is needed for future definite init improvements, and unblocked
by Doug's patch in r8039 (thanks! :)
No functionality change.
Swift SVN r8045
