Add the `@differentiable` function conversion pipeline:
- New expressions that convert between `@differentiable`,
`@differentiable(linear)`, and non-`@differentiable` functions:
- `DifferentiableFunction`
- `LinearFunction`
- `DifferentiableFunctionExtractOriginal`
- `LinearFunctionExtractOriginal`
- `LinearToDifferentiableFunction`
- All the AST handling (e.g. printing) necessary for those expressions.
- SILGen for those expressions.
- CSApply code that inserts these expressions to implicitly convert between
the various function types.
- Sema tests for the implicit conversions.
- SILGen tests for the SILGen of these expressions.
Resolves TF-833.
When wrapping a function which is supposed to capture the caller’s location, there’s always a risk that the wrapper won’t capture the information the wrapped function wants; for instance, you might pass `(…, line, column)` where the callee expected `(…, column, line)`.
This commit emits a warning when a call passes an explicit argument to something that has a default argument, and that explicit argument is itself a parameter with a default argument, and both parameters use magic identifiers, but they use *different* magic identifiers. This is partially in support of concise #file, but applies to all magic identifiers.
Fixes rdar://problem/58588633.
* WIP implementation
* Cleanup implementation
* Install backedge rather than storing array reference
* Add diagnostics
* Add missing parameter to ResultFinderForTypeContext constructor
* Fix tests for correct fix-it language
* Change to solution without backedge, change lookup behavior
* Improve diagnostics for weak captures and captures under different names
* Remove ghosts of implementations past
* Address review comments
* Reorder member variable initialization
* Fix typos
* Exclude value types from explicit self requirements
* Add tests
* Add implementation for AST lookup
* Add tests
* Begin addressing review comments
* Re-enable AST scope lookup
* Add fixme
* Pull fix-its into a separate function
* Remove capturedSelfContext tracking from type property initializers
* Add const specifiers to arguments
* Address review comments
* Fix string literals
* Refactor implicit self diagnostics
* Add comment
* Remove trailing whitespace
* Add tests for capture list across multiple lines
* Add additional test
* Fix typo
* Remove use of ?: to fix linux build
* Remove second use of ?:
* Rework logic for finding nested self contexts
Replaces `ComponentIdentTypeRepr::getIdentifier()` and `getIdLoc()` with `getNameRef()` and `getNameLoc()`, which use `DeclName` and `DeclNameRef` respectively.
This commit changes how we represent caller-side
default arguments within the AST. Instead of
directly inserting them into the call-site, use
a DefaultArgumentExpr to refer to them indirectly.
The main goal of this change is to make it such
that the expression type-checker no longer cares
about the difference between caller-side and
callee-side default arguments. In particular, it
no longer cares about whether a caller-side
default argument is well-formed when type-checking
an apply. This is important because any
conversions introduced by the default argument
shouldn't affect the score of the resulting
solution.
Instead, caller-side defaults are now lazily
type-checked when we want to emit them in SILGen.
This is done through introducing a request, and
adjusting the logic in SILGen to be more lenient
with ErrorExprs. Caller-side defaults in primary
files are still also currently checked as a part
of the declaration by `checkDefaultArguments`.
Resolves SR-11085.
Resolves rdar://problem/56144412.
Introduce the notion of "one-way" binding constraints of the form
$T0 one-way bind to $T1
which treats the type variables $T0 and $T1 as independent up until
the point where $T1 simplifies down to a concrete type, at which point
$T0 will be bound to that concrete type. $T0 won't be bound in any
other way, so type information ends up being propagated right-to-left,
only. This allows a constraint system to be broken up in more
components that are solved independently. Specifically, the connected
components algorithm now proceeds as follows:
1. Compute connected components, excluding one-way constraints from
consideration.
2. Compute a directed graph amongst the components using only the
one-way constraints, where an edge A -> B indicates that the type
variables in component A need to be solved before those in component
B.
3. Using the directed graph, compute the set of components that need
to be solved before a given component.
To utilize this, implement a new kind of solver step that handles the
propagation of partial solutions across one-way constraints. This
introduces a new kind of "split" within a connected component, where
we collect each combination of partial solutions for the input
components and (separately) try to solve the constraints in this
component. Any correct solution from any of these attempts will then
be recorded as a (partial) solution for this component.
For example, consider:
let _: Int8 = b ? Builtin.one_way(int8Or16(17)) :
Builtin.one_way(int8Or16(42\
))
where int8Or16 is overloaded with types `(Int8) -> Int8` and
`(Int16) -> Int16`. There are two one-way components (`int8Or16(17)`)
and (`int8Or16(42)`), each of which can produce a value of type `Int8`
or `Int16`. Those two components will be solved independently, and the
partial solutions for each will be fed into the component that
evaluates the ternary operator. There are four ways to attempt that
evaluation:
```
[Int8, Int8]
[Int8, Int16]
[Int16, Int8]
[Int16, Int16]
To test this, introduce a new expression builtin `Builtin.one_way(x)` that
introduces a one-way expression constraint binding the result of the
expression 'x'. The builtin is meant to be used for testing purposes,
and the one-way constraint expression itself can be synthesized by the
type checker to introduce one-way constraints later on.
Of these two, there are only two (partial) solutions that can work at
all, because the types in the ternary operator need a common
supertype:
[Int8, Int8]
[Int16, Int16]
Therefore, these are the partial solutions that will be considered the
results of the component containing the ternary expression. Note that
only one of them meets the final constraint (convertibility to
`Int8`), so the expression is well-formed.
Part of rdar://problem/50150793.
* Diagnose missing expr in non-void single-expr func. [52025782]
After SE-0255, compiling
```
func foo() -> Int {
return
}
```
would result in a diagnostic without source location:
```
<unknown>:0: error: cannot convert return expression of type '()' to
```
Now, it results in
```
filename.swift:8:5: error: non-void function should return a value
return
^
```
as it did prior to SE-0255.
To achieve that, during type checking for statements, when the
StmtChecker visits return statements, check whether we are within a
non-void-returning, single-expression function and that that single
expression is an implicit empty tuple. (An empty implicit tuple is
added as the result of a resultless return statement that appears as the
only element in a single-expression function.)
To facilitate that, `hasSingleExpressionBody` and
`getSingleExpressionBody` was added to `AnyFunctionRef` and added
`getSingleExpressionBody` to `AbstractClosureExpr` (which already had
`hasSingleExpressionBody`).
rdar://problem/52025782
If a function builder contains a buildIf function, then "if" statements
will be supported by passing an optional of the "then" branch.
"if" statements with an "else" statement are unsupported at present.
Instead of building ArgumentShuffleExprs, lets just build a TupleExpr,
with explicit representation of collected varargs and default
arguments.
This isn't quite as elegant as it should be, because when re-typechecking,
SanitizeExpr needs to restore the 'old' parameter list by stripping out
the nodes inserted by type checking. However that hackery is all isolated
in one place and will go away soon.
Note that there's a minor change the generated SIL. Caller default
arguments (#file, #line, etc) are no longer delayed and are instead
evaluated in their usual argument position. I don't believe this actually
results in an observable change in behavior, but if it turns out to be
a problem, we can pretty easily change it back to the old behavior with a
bit of extra work.
TupleShuffleExpr could not express the full range of tuple conversions that
were accepted by the constraint solver; in particular, while it could re-order
elements or introduce and eliminate labels, it could not convert the tuple
element types to their supertypes.
This was the source of the annoying "cannot express tuple conversion"
diagnostic.
Replace TupleShuffleExpr with DestructureTupleExpr, which evaluates a
source expression of tuple type and binds its elements to OpaqueValueExprs.
The DestructureTupleExpr's result expression can then produce an arbitrary
value written in terms of these OpaqueValueExprs, as long as each
OpaqueValueExpr is used exactly once.
This is sufficient to express conversions such as (Int, Float) => (Int?, Any),
as well as the various cases that were already supported, such as
(x: Int, y: Float) => (y: Float, x: Int).
https://bugs.swift.org/browse/SR-2672, rdar://problem/12340004
Before extending TupleShuffleExpr to represent all tuple
conversions allowed by the constraint solver, remove the
parts of TupleShuffleExpr that are no longer needed; this is
support for default arguments, varargs, and scalar-to-tuple and
tuple-to-scalar conversions.
Right now we use TupleShuffleExpr for two completely different things:
- Tuple conversions, where elements can be re-ordered and labels can be
introduced/eliminated
- Complex argument lists, involving default arguments or varargs
The first case does not allow default arguments or varargs, and the
second case does not allow re-ordering or introduction/elimination
of labels. Furthermore, the first case has a representation limitation
that prevents us from expressing tuple conversions that change the
type of tuple elements.
For all these reasons, it is better if we use two separate Expr kinds
for these purposes. For now, just make an identical copy of
TupleShuffleExpr and call it ArgumentShuffleExpr. In CSApply, use
ArgumentShuffleExpr when forming the arguments to a call, and keep
using TupleShuffleExpr for tuple conversions. Each usage of
TupleShuffleExpr has been audited to see if it should instead look at
ArgumentShuffleExpr.
In sequent commits I plan on redesigning TupleShuffleExpr to correctly
represent all tuple conversions without any unnecessary baggage.
Longer term, we actually want to change the representation of CallExpr
to directly store an argument list; then instead of a single child
expression that must be a ParenExpr, TupleExpr or ArgumentShuffleExpr,
all CallExprs will have a uniform representation and ArgumentShuffleExpr
will go away altogether. This should reduce memory usage and radically
simplify parts of SILGen.
For context, String, Nil, Bool, and Int already behave this way.
Note: Swift can compile against 80 or 64 bit floats as the builtin
literal type. Thus, it was necessary to capture this bit somehow in the
FloatLiteralExpr. This was done as another Type field capturing this
info.
