Most clients that will be considering same-type-to-concrete
constraints will need to look at all of the constraints within the
equivalence class together. Store the same-type-to-concrete
constraints on the EquivalenceClass itself, which fits this use case
better, and should reduce the storage requirements by making potential
archetypes two pointers smaller. NFC
It looks like the devirtualizer used to have problems computing
method types in the presence of generic substitutions, covariant
returns and other things, so it would bail if a perticular set
of pre-conditions was not met on the types of original method call
and the devirtualized method call.
I don't think any of this is necessary anymore. If this patch
introduces any regressions, we need to fix the root cause instead
of re-introducing this logic.
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.
This pass now canonicalizes results before lowering and handles all combinations
of direct and indirect multiple return values. The logic is much less ad-hoc and
more robust.
try_apply still isn't handled, but should be much easier now.
Add visitLoadInst, visitStoreInst, visitDebugValueInst, etc.
Previously some decls (TypeAliasDecl and ExtensionDecl) had bits
explicitly marking whether they've been validated, while other decls
just deduced this from hasInterfaceType. The doing the latter doesn't
work when the interface type can be computed before doing full
validation (such as protocols and associatedtypes, which have trivial
interface types), and so an explicit bit is adopted for all decls.
This makes the demangler about 10 times faster.
It also changes the lifetimes of nodes. Previously nodes were reference-counted.
Now the returned demangle node-tree is owned by the Demangler class and it’s lifetime ends with the lifetime of the Demangler.
Therefore the old (and already deprecated) global functions demangleSymbolAsNode and demangleTypeAsNode are no longer available.
Another change is that the demangling for reflection now only supports the new mangling (which should be no problem because
we are generating only new mangled names for reflection).
The compiler itself no longer uses this API but the debugger does,
in order to pretty-print option sets.
The normal way to test this would be to add an LLDB-side test that
uses a framework with versioned API notes. Unfortunately I can't
think of a straightforward way to test it Swift-side.
When we see a second same-type-to-concrete constraint on a particular
potential archetype, record it. Previously, we were checking it and
then updating the requirement source eagerly. That won't work with
proper recursion detection, and meant that we missed out on some
obvious redundant-same-type-constraint diagnostics.
The scheme here is to build up the equivalence classes without losing
any information, and then determine which information is redundant at
the end.
Introduce an equivalence-class abstraction that captures all of the
members of the equivalence class in a separate type that will maintain
the "truth" about the meaning of the equivalence class, rather than
having that information distributed amongst the potential archetypes
within the class.
For now, use it to capture the members of the equivalence classes, so
we have one SmallVector per equivalence class rather than N
SmallVectors.
Diagnose when a same-type constraint (to a concrete type) is made
redundant by another same-type constraint. Slightly improve the
diagnostic that handles collisions between two same-type constraints.
This is disabled by default but enabled under the frontend option
-propagate-constraints.
The idea here is to have a pass that enforces local consistency in our
constraint system, in order to reduce the domains of constraint
variables, speeding up the solving of the constraint system.
The initial focus will be on reducing the size of the disjunctions for
function overloads with the hope that it substantially improves the
performance of type checking many expressions.
