Reimplement the witness matching logic used for generic requirements
so that it properly models the expectations required of the witness,
then captures the results in the AST. The new approach has a number of
advantages over the existing hacks:
* The constraint solver no longer requires hacks to try to tangle
together the innermost archetypes from the requirement with the
outer archetypes of the context of the protocol
conformance. Instead, we create a synthetic set of archetypes that
describes the requirement as it should be matched against
witnesses. This eliminates the infamous 'SelfTypeVar' hack.
* The type checker no longer records substitutions involving a weird
mix of archetypes from different contexts (see above), so it's
actually plausible to reason about the substitutions of a witness. A
new `Witness` class contains the declaration, substitutions, and all
other information required to interpret the witness.
* SILGen now uses the substitution information for witnesses when
building witness thunks, rather than computing all of it from
scratch. ``substSelfTypeIntoProtocolRequirementType()` is now gone
(absorbed into the type checker, and improved from there), and the
witness-thunk emission code is simpler. A few other bits of SILGen
got simpler because the substitutions can now be trusted.
* Witness matching and thunk generation involving generic requirements
and nested generics now works, based on some work @slavapestov was
already doing in this area.
* The AST verifier can now verify the archetypes that occur in witness substitutions.
* Although it's not in this commit, the `Witness` structure is
suitable for complete (de-)serialization, unlike the weird mix of
archetypes previously present.
Fixes rdar://problem/24079818 and cleans up an area that's been messy
and poorly understood for a very, very long time.
Today, loads and stores are treated as having @unowned(unsafe) ownership
semantics. This leaves the user to specify ownership changes on the loaded or
stored value independently of the load/store by inserting ARC operations. With
the change to Semantic SIL, this will no longer be true. Instead loads, stores
have ownership semantics that one must reason about such as copy, take, and
trivial.
This change moves us closer to that world by eliminating the default
OwnershipQualification argument from create{Load,Store}. This means that the
compiler developer cannot ignore reasoning about the ownership semantics of the
memory operation that they are creating.
Operationally, this is a NFC change since I have just gone through the compiler
and updated all places where we create loads, stores to pass in the former
default argument ({Load,Store}OwnershipQualifier::Unqualified), to
SILBuilder::create{Load,Store}(...). For now, one can just do that in situations
where one needs to create loads/stores, but over time, I am going to tighten the
semantics up via the verifier.
rdar://28685236
When AnyHashable was added, SILGen gained support for lowering
AnyHashableErasureExpr, however we forgot to also add support
for AnyHashable parameter and result conversions to
FunctionConversionExpr.
Fixes <https://bugs.swift.org/browse/SR-2603>.
Previously IRGen was using a heuristic to assign the argument number
to the $error variable that was not generally correct for optimizaed
code. This patch inserts a debug_value instruction in SILGen and thus
assigns the ArgNo together with all the other function arguments. This
is more robust and faster than than the old heuristic.
<rdar://problem/28748323>
Fixes rdar://problem/28873860, where we would miscompile when lightweight generic classes were extended to conform to Swift protocols because we tried to emit parameters for the class's generic parameters for the witness entry points. Prevent this by lowering the witness into a pseudogeneric function in SILGen, and teaching IRGen to do the right thing for a witness with pseudogeneric parameters.
Sema produces a weird mix of requirement and witness archetypes when
it builds witness substitutions. Instead of hacking around this in
SILGen, just build new archetypes, since we already have to use an
ArchetypeBuilder to get the correct generic signature on the interface
type.
Then, we just have to map the witnessSubs provided by Sema to use our
new archetypes.
This unblocks some work on generic inlining.
Fixes <rdar://problem/28765006>.
We don't want the machine calling conventions for closure invocation functions to necessarily be tied to the convention for normal thin functions or methods. NFC yet; for now, 'closure' follows the same behavior as the 'method' convention, but as part of partial_apply simplification it will be a requirement that partial_apply takes a @convention(closure) function and a box and produces a @convention(thick) function from them.
radar rdar://problem/28434323
SILGen has no reason to insert shadow copies for inout parameters any more. They cannot be captured. We still emit these copies. Sometimes deshadowing removes them, but sometimes it does not.
In this PR we just avoid emitting the copies and remove the deshadowing pass.
This PR chery-picked some of @dduan work and built on top of it.
Sugared GenericTypeParamTypes point to GenericTypeParamDecls,
allowing the name of the parameter as written by the user to be
recovered. Canonical GenericTypeParamTypes on the other hand
only store a depth and index, without referencing the original
declaration.
When printing SIL, we wish to output the original generic parameter
names, even though SIL only uses canonical types. Previously,
we used to accomplish this by mapping the generic parameter to an
archetype and printing the name of the archetype. This was not
adequate if multiple generic parameters mapped to the same
archetype, or if a generic parameter was mapped to a concrete type.
The new approach preserves the original sugared types in the
GenericEnvironment, adding a new GenericEnvironment::getSugaredType()
method.
There are also some other assorted simplifications made possible
by this.
Unfortunately this makes GenericEnvironments use a bit more memory,
however I have more improvements coming that will offset the gains,
in addition to making substitution lists smaller also.
The behaviour of ilist has changed in LLVM. It is no longer permissible to
dereference the `end()` value. Add a check to ensure that we do not
accidentally dereference the iterator.
This lets us get to the goal of +0 guaranteed closure contexts. NFC yet, just add the under-the-hood ability for partial_apply instructions producing callee-guaranteed closures to be parsed, printed, and serialized.
This particular flag should only be used in rare cases where we don't
want to know about failures, but instead want to get some
partially-formed type. Only very specific parts of the type checker
need this (associated type inference), and code completion relies on
it for slightly-better results.
It's the same thing as for alloc_ref: the optional [tail_elems ...] attribute specify the tail elements to allocate.
For details see docs/SIL.rst
This feature is needed so that we can allocate a MangedBuffer with alloc_ref_dynamic.
The ManagedBuffer.create() function uses the dynamic self type to create the buffer instance.
Backed off of the implementation with a shared body block, which got
messy really fast with multiple patterns containing address-only values
to pass in, as it led to separated stack alloc/cleanup requirements and
passing stack alloc’d values as block args, aliasing of addresses for
stack cleanup verifying, etc. A mess.
This now does the conservative thing of repeating the body in each
matching pattern block.
The existing code assumed that the _bridgeToObjectiveC witness would never come from a protocol extension. Generating the correct set of substitutions is a bit harder than it ought to be, due to inconsistencies in how Sema represents generic witnesses. Fixes rdar://problem/28279269.
The commit was reverted because of a regression in the
Prototypes/CollectionTransformers test. I believe the root
cause was an escape analysis bug, which is fixed in my
previous commit.
id-as-Any lets you pass Optional to an ObjC API that takes `nonnull id`, and also lets you bridge containers of `Optional` to `NSArray` etc. When this occurs, we can unwrap the value and bridge it so that inhabited optionals still pass into ObjC in the expected way, but we need something to represent `none` other than the `nil` pointer. Cocoa provides `NSNull` as the canonical "null for containers" object, which is the least bad of many possible answers. If we happen to have the rare nested optional `T??`, there is no precedented analog for these in Cocoa, so just generate a unique sentinel object to preserve the `nil`-ness depth so we at least don't lose information round-tripping across the ObjC-Swift bridge.
Making Optional conform to _ObjectiveCBridgeable is more or less enough to make this all work, though there are a few additional edge case things that need to be fixed up. We don't want to accept `AnyObject??` as an @objc-compatible type, so special-case Optional in `getForeignRepresentable`.
Implements SR-0140 (rdar://problem/27905315).
Those builtins are: allocWithTailElems_<n>, getTailAddr and projectTailElems
Also rename the "gep" builtin, which indexes raw bytes, to "gepRaw" and add a new "gep" builtin to index in a typed array.
If the thunk's type otherwise did not involve type parameters, we
would still pass around the generic parameters from the caller's
context, which is wasteful.
This would manifest as crashes in IRGen when bridging a block
taking another block in generic context.
- When emitting a re-abstraction thunk, make it pseudogeneric if
its parent function is pseudogeneric. This ensures we don't
try to get runtime type metadata when it doesn't exist.
- Mangle pseudogeneric-ness of a reabstraction thunk correctly.
Otherwise we could emit thunks with different signatures under
the same mangling.
- Only set the pseudogeneric attribute if the thunk has a generic
signature, since otherwise the mangling is no longer unique.
Fixes <rdar://problem/27718566>.
id-as-Any lets you pass Optional to an ObjC API that takes `nonnull id`, and also lets you bridge containers of `Optional` to `NSArray` etc. When this occurs, we can unwrap the value and bridge it so that inhabited optionals still pass into ObjC in the expected way, but we need something to represent `none` other than the `nil` pointer. Cocoa provides `NSNull` as the canonical "null for containers" object, which is the least bad of many possible answers. If we happen to have the rare nested optional `T??`, there is no precedented analog for these in Cocoa, so just generate a unique sentinel object to preserve the `nil`-ness depth so we at least don't lose information round-tripping across the ObjC-Swift bridge.
Making Optional conform to _ObjectiveCBridgeable is more or less enough to make this all work, though there are a few additional edge case things that need to be fixed up. We don't want to accept `AnyObject??` as an @objc-compatible type, so special-case Optional in `getForeignRepresentable`.
Implements SR-0140 (rdar://problem/27905315).
Those builtins are: allocWithTailElems_<n>, getTailAddr and projectTailElems
Also rename the "gep" builtin, which indexes raw bytes, to "gepRaw" and add a new "gep" builtin to index in a typed array.
We can't place counters on function decls with no bodies. The old
behavior was to map a fresh counter to a null AST node, which is highly
suspicious at best.
Actually bridging ObjCBool to Bool is overkill for this, but moreover
it caused problems for non-boolean types that took this code
path. Just go back to the previous logic of unwrapping multiple levels
of struct; this way we can also handle wrappers around integer types
(if we ever have any).
rdar://problem/27985744
