When initializing `.base` in class initialization, use `partial Base` as
the destination type rather than `Base`. Treat `partial Base` as not
being abstract even when `Base` is.
Allow conversion from a `partial T` initializer to a `T` initializer.
Store the vptr while performing the conversion. Do not store the vptr
when performing a `partial T` initialization, only when performing a
non-partial `T` initialization.
Stop using "performed builtin conversion" as a proxy for whether we
created an initializing expression with a correctly-set storage
argument. That isn't correct in the case where the builtin conversion
creates a new initializing expression without setting its storage, such
as by creating an `AsCompatible` wrapper around an existing initializing
expression.
Instead look at whether the storage argument is a `TemporaryStorage`,
and only overwrite in that case, otherwise assuming that the storage
argument has been set correctly.
This fixes a miscompile that was already visible in our lowering tests!
Introduces `Context` and `SoftContext` messages, which can be introduced
through a `ContextBuilder`:
- The `Context` messages come before the diagnostic in the output.
- The first `Context` message steals the diagnostic level from the main
diagnostic, and turns the main diagnostic into a Note attached to the
context.
- A `SoftContext` message works similarly, but if it's preceeded by a
`Context` or `SoftContext` message, then it is dropped. This can be used
as a default/backup scope when nothing more interesting is provided up
the stack, such as in `TryEvalBlockForSpecific`.
The `ContextBuilder` is provided to a callback through
`Diagnostics::ContextScope`, an RAII type `AnnotationScope` but for
context messages.
This allows a high level operation to provide a context message like
"failed to identify facet type {0}" which will then be used as the error
if a diagnostic is produced during identification, with the latter
diagnostic attached as a note to explain why the contextual operation
failed.
In particular, this allows monomorphization errors (such as an array
bound being negative) to be attached to a higher lever operation instead
of being top-level diagnostics themselves, with the monomorphization
site being a note. This inverts the source code locations that appear in
the diagnostic, so that the top-level diagnostic points to the "user
code" which causes the monomorphization.
This is presented as an alternative strategy to #6753, which plumbed
diagnoser callbacks around to achieve the same goals.
We replace the diagnoser callbacks in type completion and operators with
ContextScope callbacks instead, which now provide better diagnostics for
monomorphization errors. Other callers to MakeSpecific do not yet have
ContextScopes introduced in order to turn monomorphization errors into
more interesting diagnostics.
While convert has the option to avoid diagnostics, when that flag is
false, ErrorInst results must also produce a diagnostic. Otherwise we
end up with errors in the semir but not error provided to the user.
The new diagnostics reveal that a number of tests for abstract types
were passing incorrectly. They had errors in the semir but no
diagnostics. A TODO is added in convert to allow an abstract conversion
target type when not initializing.
This is needed to model things like the category of `x` in the body of
`fn Foo(F:! Core.Form, x:? F)`, where the category of `x` is determined
by the concrete value of `F` (see #5389 for the design of `:?`
bindings).
This will be used in a follow-up PR.
The primary change in this PR is to split the `Initializing` expression
category into separate `ReprInitializing` and `InPlaceInitializing`
categories, depending on whether initialization uses the types
initializing representation, or is guaranteed to be in place. It also
rationalizes and documents the SemIR-level semantics of those categories
(including where #5545's "ephemeral entire reference" category will
fit), and introduces two new inst kinds to close gaps exposed in the
process.
Some additional secondary changes:
- Consistently format the storage arguments of initializers with `to`,
regardless of whether initialization is in-place, and document the `to`
notation.
- Rename some inst kinds and functions, and restructure some of the
code, for clarity and consistency with the new documentation.
- Resolve a TODO to handle more category conversions in
`CategoryConverter`, in order to make it easier to reason about category
conversions.
See #6588 and the review history of this PR for background.
---------
Co-authored-by: Richard Smith <richard@metafoo.co.uk>
Identifying a facet type takes both a self and facet type as a pair, and
then encode the self into the IdentifiedFacetType. This makes a
constraint that requires some _other_ type implements an interface
visible in the IdentifiedFacetType. And it will help to enable facet
types with `where T impls Z` for `T` that is not `.Self` in the future.
IdentifiedFacetTypes are now stored in a CanonicalValueStore instead of
a RelationalValueStore as they key is the combination of self and
(declared) facet type together now.
When the self-type is a facet value (has type FacetType) this is most
straightforward. But when it's a type we need to construct a FacetValue
to construct a specific for a require decl, to replace the generic
binding of the symbolic `Self`, which has type FacetType. To do so, we
make a FacetValue with an empty FacetType (equivalent to TypeType). This
prevents any looking for witnesses through the FacetType, which matches
what you can get from a type directly, requiring witnesses to come from
finding an `impl` decl.
Add additional InstNamer logic for such empty facet types so they print
as `<typename>.type.facet` if possible instead of as just `facet_value`.
Also clarify and enforce that `ConversionTarget::init_id` is used only
as storage for in-place initialization, and correspondingly rename it to
`storage_id`.
`ReturnTypeInfo` is built around the assumption that a function call
results in exactly one initializing expression, but with `ref` returns
there may be zero, and in the future composite return forms will enable
there to be more than one. This change removes some usages of
`ReturnTypeInfo`, and restructures the calling code to be prepared for
multiple initializing returns.
Adds Inst::IsOneOf which takes a variadic generic parameter pack of
kinds to check against. Also add forwarding functions to TypeStore and
InstStore. Convert uses of the regex `Is<.*\|\|` to IsOneOf.
This is based on #6522
The main changes here are:
- Introducing `InitForm` and `RefForm` to represent initializing and
reference forms (the two return forms currently supported by the
parser).
- Introducing the `FormType` singleton inst to represent their type
(i.e. `Core.Form`).
- Emitting an inst representing a function's declared return form as
part of handling the function signature.
The return form inst is currently ignored. Subsequent PRs will expose it
in `SemIR::Function` and use it to determine the form of call
expressions.
The fallthrough-based approach was unwieldy and error-prone, and
inherently couldn't support category conversions whose steps don't
follow the fixed order of the `switch` statement.
I'm doing this because I figured it'd be an incremental improvement for
all the operator lookups that we do. Even to the extent that we've
discussed witness caching, I think it'll still apply. It does add one
more step to adding new interfaces (before, you'd just write the string,
now you add it to the def file and reference it).
I'll claim it makes GetClangOperatorKind a lot friendlier to read/edit,
nevermind removing the string comparisons. :)
This separates the return type from the return pattern, and replaces the
return pattern with a block of return patterns. This is a step toward
support for `ref` returns (where there's no corresponding return
pattern) and compund-form returns (where there may be multiple return
patterns).
Avoid using a large switch that needs to be manually extended when
adding a new kind of instruction. Instead, the expression category for
an instruction is now specified when defining the `InstKind`.
In passing, add a distinct expression category value for patterns. This
isn't used for much except some error checking at the moment, but it
keeps the number of instructions that we need to manually classify as
`NotExpr` despite having a type very low.
Give TupleLiteral and StructLiteral a constant value, if their contents
have constant values. Their constant values are TupleValue and
StructValue respectively. This supports their ability to convert to a
constant type (or facet type).
This way when deduce finds a TupleLiteral as the argument to a
_symbolic_ facet type, it can also find a constant value to use for that
argument. This allows deduction to move onto step two, where it can
substitute into the symbolic parameter from previous deduced arguments,
and then perform the conversion from the TupleValue to the desired facet
type.
Allow `PerformBuiltinConversion()` to convert from a canonical
TupleValue or StructValue to `type` instead of only from literals. Then,
also support conversion from a symbolic binding of type TupleType or
StructType to `type`.
If the value representation of `T` is a copy representation, but it
copies all of the bits of `T`'s object representation, then it's OK to
use that as the value representation of `MaybeUnformed(T)` too.
This fixes the behavior of interop with nullable pointers, which are
represented as an adapter of `MaybeUnformed(T*)`, and need to be passed
to and returned from functions on the Carbon / C++ boundary as `T*`s.
As a byproduct, the only test that exercised the "address of a temporary
object" diagnostic now trigers the "address of a non-reference
expression" diagnostic. We could restore it by using a type that doesn't
support `value_of_initializer`, but it seems better to remove the
diagnostic altogether: not only does it simplify the code, I'd also
argue "non-reference expression" is more accurate as a user-facing
description of the operand.
#6289 absentmindedly added fields in more places, and this is undoing
that plus further fixes.
This does some cleanup of types with relation to singletons. For
`TypeType` and `ErrorInst`, they're always complete due to a
`SetComplete` call in `file.cpp`. For `CppVoidType`, it's intended to be
incomplete by construction, and so a `TypeId` should be okay. The intent
though on not generally providing these had been that `GetSingletonType`
needs to be called to get a type to be marked as complete.
In the case of `AutoType`, removing `TypeId`does change a small printing
detail. I think that's old legacy that's just been carried forward.
Otherwise, for both `InstType` and `AutoType`, I've added
`GetSingletonType` calls where they were used in order to ensure
completeness is applied correctly. These calls cause small SemIR
permutations.
This causes `AutoType` to be seen by lowering, so I'm adding a
placeholder for it. Also merging two functions that look like they're
identical in intent -- not sure why they're separate.
---------
Co-authored-by: Dana Jansens <danakj@orodu.net>
`const` doesn't mean much on the type of a value expression; it's valid
to remove it because we can't perform modifications to a const value
regardless.
We already allowed most of this, but only as part of adapter conversion
rather than in general, and we didn't previously allow it when the
source of the conversion was a reference expression.
---------
Co-authored-by: josh11b <15258583+josh11b@users.noreply.github.com>
The SymbolicBindingType refers to the type value that will be
substituted in for the BindSymbolicName, but holds onto the EntityNameId
from the BindSymbolicName instead of (or in addition to, for now) the
instruction.
The EntityNameId will be used to look in the ScopeStack to find the
witnesses either from the BindSymbolicName instruction, or other
instructions that specify `impls` constraints against the EntityName.
This will allow us to have the `T` in `I(T)` resolve to a `.Self`
reference in the type so that we get type equality with the binding's
type: `T:! I(.Self)`.
Previously it performed two kinds of operations, with a boolean
parameter to control whether it would unwrap FacetValue or not. This
made the function hard to explain as "canonicalization".
Now the contract of GetCanonicalFacetOrTypeValue is as follows:
1. For a facet value expression, it returns the canonical value of the
facet value.
2. For a `<facet value> as type` it returns the canonical value of the
`<facet value>`.
3. For other type expressions, it returns the canonical value of the
type.
1 and 2 together collapse together two representations of a facet value
(as a FacetType or as a TypeType) into a single canonical value, which
is important for constant comparison of facet values where the `as type`
is not meant to change the result. This is the case in impl lookups and
`.Self` comparisons.
The step of unwrapping `FacetValue` is only useful in the constant
evaluation of `LookupImplWitness` and is used to collapse *symbolic*
queries on `FacetValue(T)` and on `T` down to a single canonical value,
since they produce the same result later when `T` is replaced with a
facet value or type that can provide a concrete witness. This is now
extensively documented in the constant evaluation of
`LookupImplWitness`.
This change came out of a request/discussion in #6115 (see comment
https://github.com/carbon-language/carbon-lang/pull/6115#discussion_r2383696576).
When doing impl lookup with a constraint facet type including the
builtin `TypeCanAggregateDestroy`, we look at the type to see if it
satisfies it. However if the type is a facet value, we need to look at
the FacetType to see if the eventual concrete type is going to satisfy
it.
Note that we can do this check up front in the `LookupImplWitness()`
function without creating a symbolic instruction to be modified by
future specifics with a more precise type for the facet value, because
the result of `TypeCanAggregateDestroy` does not actually provide a
witness, so we don't need the final specific type.
This was noticed by removing the "shortcut" in convert for converting a
`FacetAccessType(<symbolic binding>)` to `typeof(<symbolic binding>)`.
By removing the shortcut, we go into impl lookup when checking `impl`
decls containing `TypeCanAggregateDestroy` via deduce.
This makes convert more consistent, it always makes a FacetAccessType
for a facet value, rather than only doing so after lookup returns. The
intention for this is that FacetAccessType will evaluate to
SymbolicBindingType in the future, so this will expose that constant
value to impl lookup instead of the original facet value, which will
avoid impl lookup having to deal with `.Self` or `BindSymbolicName`
specifically.
If a `BindSymbolicName` is converted to `type` and then to its exact
`FacetType`, we get a `FacetValue` wrapping the `BindSymbolicName` but
providing no different information: it has the same witnesses and
`FacetType` as the original `BindSymbolicName`. Yet it is a different
constant value, creating multiple canonical forms with the same meaning.
Now we make that `FacetValue` with the same `FacetType` as the
`BindSymbolicName` it wraps evaluate back to the `BindSymbolicName`,
making it the unique canonical form.
This makes the "shortcut" in convert for avoiding impl lookup when
converting from `FacetAccessType` to `FacetType` in this exact scenario
work the same as doing the full impl lookup.
---------
Co-authored-by: Richard Smith <richard@metafoo.co.uk>
Add `Dependent` value and initializing representations for types whose
representations are unknown because they are dependent. When generating
SemIR in such cases, use a worst-case initializing representation that
both provides a destination address and also propagates a potential
result value.
Use this to fix incorrect lowering and lowering crashes for specific
functions involving generic types that don't use a copy value
representation.
In lowering, be careful to distinguish between whether the initializing
representation for the generic return type uses a return slot (which
affects whether the SemIR declaration and call have one) and whether the
initializing representation for the specific return type uses a return
slot (which affects whether the LLVM IR declaration and call have one).
This is a bit of an experiment to see if there's a reasonable way to
write a shared enum type, rather than writing per-case wrappers for
things like `HasTypeQualifiers` or the printing. I think it's a bit
borderline complexity right now, but I'm not sure I can reduce it much
further.
This changes from things like `Internal::EnumClassName##RawEnum` to
`Internal::EnumClassName##Data::RawEnum` so that the enum entries can
have back references to bit shifts without needing to know the
containing type name. Because I'm trying to reduce duplication between
mask and non-mask enums, I did this to non-mask enums too.
This was motivated by #6035 adding another enum mask (which will grow
more entries, and is intended to switch if this is accepted), but I'm
not using that PR as a base here because I didn't want the merge
dependency.
Instead of hardcoding which types are copyable, add a `Core.Copy`
interface to perform copying. Move almost all the current copy support
to that interface. Some remaining pieces are still using builtin logic
after this PR:
* For tuples and structs, builtin logic is used to perform elementwise
copies. This also supports copying *adapters of* tuples and structs,
which seems like it may not be desirable, especially for non-extending
adapters. A `Copy` impl is provided for tuples of at most 2 elements, so
that `Core.Copy` constraints are satisfied, but we can't implement this
generally until we have variadics support, and don't yet have a
mechanism to generalize this to structs.
* For `enum` types imported from C++, builtin logic is used to perform a
copy. This is temporary until we have a mechanism to identify these
types from an impl in the prelude.
One lowering test in `toolchain/lower/testdata/class/generic.carbon` is
disabled for now, as it causes a crash in the lowering code due to an
ABI mismatch between the call signature in the lowered declaration of a
specific function and the call that is generated in the specific callee.
Fixing this is a little involved, and will be done in a separate PR.
---------
Co-authored-by: Geoff Romer <gromer@google.com>
Attach the cleanup to the `Temporary` instruction instead of to the
`TemporaryStorage` instruction. We create `TemporaryStorage`
instructions speculatively when creating an initializing expression, and
may overwrite those instructions with other instructions if it turns out
that a temporary is not required. Instead, wait until we finalize the
temporary and create a `Temporary` instruction to register the cleanup.
We already allowed this for reference expressions; this extends the
support to also cover value expressions. This requires a little more
work because the value representation of `T` and `MaybeUnformed(T)`
don't necessarily match in general.
* Treat `MaybeUnformed` and `partial` as qualifiers, like `const`.
* Allow pointer conversions to add qualifiers.
* Allow unsafe pointer conversions to remove qualifiers.
* Allow conversions on non-reference expressions to drop `const`.
* Allow unsafe conversions on any expression to drop `const`.
* Allow unsafe conversions on non-initializing expressions to drop
`partial`. For initializing expressions, we should initialize the
vptr when dropping `partial`; this is not yet supported so we reject.
* Allow conversions on reference expressions to add `MaybeUnformed`.
* Allow unsafe conversions on reference expressions to drop
`MaybeUnformed`. For non-reference expressions, additional work is
required, because the value / initializing representation may not
match between `T` and `MaybeUnformed(T)`, so those are rejected for
now.
Following the direction of #5913, add support for parsing an `unsafe as`
operator. For now, we allow one additional conversion using `unsafe as`
beyond the conversions supported by `as`: we permit pointer conversions
that remove qualifiers, such as `const T*` -> `T*`.
This is necessary if the source type is an adapter, as we would not
otherwise be able to determine what type it adapts and hence could be
converted to.
This addresses/avoids the duplicate import of vtables.
I went through a few iterations/etc along the way and left them in the
commit
history for the PR in case any of them are useful to illustrate how I
got here,
or worth revisiting.
Essentially I ended up with a circularity in importing - importing the
class
imported the vtable_decl which imported the virtual functions - and then
pending
specifics of the virtual functions needed the self specific of the
enclosing
class which wasn't ready yet.
Adding ImportRef to the vtable_decl to break the cycle caused me trouble
when
naming the vtable_decl instructions - so I tried making the functions in
the
vtable unloaded ImportRefs instead. That worked, but meant that
importing a
class still was doing O(number of vtable entries) even if the vtable
wasn't
used.
So I revisited the lazy vtable_decl - figured out how to make the naming
work
(when building the vtable_ptr, even though the vtable_decl doesn't have
to be
loaded for the vtable_ptr, I force it to be loaded anyway, to load the
vtable so
it's usable by lowering, etc). And then I could go back to the old
non-lazy
loaded vtable entries (using some loaded ImportRefs in the cases where
we needed
them/had already adopted them).
Then thinking about the VtablePtr instruction, went back/forth on
exactly what
it needed - went from VtablePtr's member being a VtableDecl InstId, to a
ClassId, then back to a VtableId as it was before this patch.
Naming the instructions has one oddity, that the VtableDecl and
VtablePtr
instructions seem to need to add the pending name for the VtableId -
despite not
using the VtableId in their own name - should the inst namer be doing
this work
for parameters of instructions rather than requiring the inst to do it
deliberately? (or am I holding it wrong in some way?)
---------
Co-authored-by: Richard Smith <richard@metafoo.co.uk>
* Rename the type.
* Change lowering to lower FloatLiteralType values as the placeholder
`{}` value we use for literals instead of as an LLVM f64.
* Change eval to convert the type as part of a floating point
conversion, so that lowering can lower converted constants properly.
For now we still represent a value of FloatLiteralType as a
double-precision APFloat. (That will need to change so that we can
losslessly convert literals to f80 / f128 values, and so that we can
convert literals to f32 values without double-rounding.)
When initializing a C++ thunk parameter:
* If we have an initializing expression, materialize a temporary and
pass its address.
* If we have a reference expression, pass its address directly.
* If we have a value expression with a pointer value representation,
pass the pointer.
* Otherwise, create a new temporary and initialize it with a copy of the
argument, and pass its address.
Richard Smith
Dana Jansens
Boaz Brickner
David BlaikieandRichard Smith