When binding an optional value, or function that returns an optional
value, if that value was produced from a decl that was declared an
IUO, create a disjunction.
After solving, make use of the disjunction choices in rewriting
expressions to force optionals where needed.
This is disabled for now, as it results in a source compatibility
issue without associated changes that actually start generating
Optional<T> in place of ImplicitlyUnwrappedOptional<T>. It's
complicated, but basically having two '??' (one returning T, one
returning T?) and creating a disjunction where the first (favored)
choice is ImplicitlyUnwrappedOptional<T> and second is T results in
our selecting the wrong '??' in some cases.
Replace ImplicitlyUnwrappedCoercionResult with the more general
ImplicitlyUnwrappedValue.
Add ImplicitlyUnwrappedDisjunctionChoice as a marker that indicates
that we've already created a disjunction from an
ImplicitlyUnwrappedValue, in order to avoid infinite recursion during
binding.
Also add support to buildDisjunctionForImplicitlyUnwrappedOptional for
functions returning optionals.
NFC.
This reverts commit 2f80af15ec.
I expected this to have no effect, but it results in one of the
expression type checker tests taking longer, so that test was disabled.
This commit also re-enables the test since it now passes again.
Instead of binding collection types directly let's try to
bind using temporary type variables substituted for element
types, that's going to ensure that subtype relationship is
always preserved.
Resolves: rdar://problem/35541153
These purportedly mark that we should stop attempting fixes for a given
constraint, but in fact the only code creating these is clearly
unreachable so these serve no purpose.
This allows us to eliminate some special casing around things like
parens, and allows the code to work for other code involving expressions
that are semantically neutral.
We now store an entire OverloadChoice in the unviable candidates
list (which is used for error recovery), just like we store them
for viable candidates.
The additional information isn't used, so NFC.
Consider different overload choices for the same location in evaluation
order, this makes overload resolution more predictable because it's going
to follow expression bottom-up, that prevents situations when some
expressions are considered ambigious because choices taken further up
equate the score, instead each level is given distinct weight
based on evaluation order.
Resolves: rdar://problem/31888810
Presence of some constraints (Subtype at least) requires a certain
contextual ranking of the type variables associated with them when
it comes to picking bindings, otherwise it might lead to no or
invalid solutions, because only a set of the bindings for the best
type variable is attempted.
Resolves: rdar://problem/22898292
Presence of some constraints (Subtype at least) requires a certain
contextual ranking of the type variables associated with them when
it comes to picking bindings, otherwise it might lead to no or
invalid solutions, because only a set of the bindings for the best
type variable is attempted.
Resolves: rdar://problem/22898292
For the moment, update the sole caller to not infer Any in cases where
neither argument was Any. This is in part because not all the cases are
handled here and the result of inferring Any will be to miss appropriate
bindings in certain cases. It is also in part because by inferring Any
we may end up deferring diagnostics until later in a function, which may
result in very hard to understand diagnostics for users.
This will be revisted once all the cases are properly handled here.
Currently edge related to the parameter bindings is contracted
without properly checking if newly created equivalence class has
the same inout & l-value requirements. This patch improves the
situation by disallowing contraction of the edges related to parameter
binding constraint where left-hand side has `inout` attribute set.
Such guarantees that parameter can get `inout` type assigned when
argument gets `l-value` type.
Resolves: rdar://problem/33429010
Change the potential binding ranking logic to prefer type variables
which don't have other type variables related to them, but have literal
bindings, over the ones that do have other type variables.
The naming here might not be perfect as it's not the only place that we
bind types to typevars in matchTypes, but it is a step towards cleaning
things up a bit. I'll revisit naming after I get a better idea what
other common code there may be that can get factored out.
When dealing with explicit coercions apply their conversions
early to avoid searching for solutions with incorrect types,
and therefore prune search space.
Move disjunction selection logic one level up from the `solveSimplified`
which allows to simplify its logic and avoid collecting disjunctions
multiple times for each solver step.
Sort constraint components/buckets based on how many disjunctions
they have, that helps to prune some of the branches with incorrect
solutions early which limits overall depth of the search.
Resolves: SR-4714
If we allow the right-hand type to be bound first, it can artificially
limit the options for the left-hand side, so attempt to delay binding of
the right-hand side types until after we've chosen something for the
left-hand side.
The test here will not fail without this change, but will fail without
this change if the ConstraintGraph.cpp changes from @xedin's edge
contraction patch (https://github.com/apple/swift/pull/11118) are
applied.
While shrinking we have to allocate containers for the reduced domains
for some of the candidates, it's currently done using permanent arena
of the `ASTContext` allocator. This patch changes candidate solver to
use arena associated with the parent constraint system, which significantly
limits lifetime of domain containers.
