This also fixes passing a value expression to a forwarding reference,
since we currently deduce a `const T&&` parameter in that case.
We were accidentally looking at the type of the thunk parameter (which
is never an rvalue reference) rather than the type of the callee
parameter.
Previously, we picked a single Carbon parameter pattern for each C++
parameter pattern. This doesn't work well in cases where the Carbon
semantics and the C++ semantics are not perfectly aligned. In
particular, when a parameter is passed by value in C++, that might mean
either pass-by-move (which in Carbon would best be modeled by a `var`
pattern, as no other form of parameter would perform a move) or
pass-by-copy (which in Carbon would best be modeled by a value
parameter, as a `var` parameter would force an extra copy).
After this change, we compute a passing mode for each parameter based on
the implicit conversion sequence from the argument to the parameter as
determined by C++ overload resolution, and use that to determine the
Carbon pattern corresponding to each C++ parameter. This results in
potentially generating multiple different thunks for the same C++
function if it's called in different ways, but we already did that to
handle default arguments and list-initialization. The passing modes are
included in the thunk mangling.
Add a new value store for clang decl signatures, which capture the
information about parameter passing mode as well as the other existing
information about different ways that a C++ function might be imported
to Carbon.
Most of the rules for computing passing modes are the same as before:
const references use pass by value, non-const lvalue references use
pass-by-ref, non-const rvalue references use pass-by-var. But for C++
non-reference parameters, pick between pass-by-value and pass-by-var
based on whether the implicit conversion sequence was effectively
performing a copy. Prefer pass-by-value if either would work and they'd
do the same thing. We still use pass-by-value for const references, even
when the argument is an lvalue and we could pass a reference; we may
want to change this in future.
For virtual functions, we try to pick a worst-case passing mode, as we
can only pick a single signature for what goes in the vtable. Calls to
virtual functions will still use a thunk to C++, allowing variance in
the calling convention at call sites. We don't allow variance in the
overriders as we don't implement support for thunks for virtual
functions yet. We currently use pass-by-value for const reference
parameters here, but that should probably change at some point.
Assisted-by: Gemini via Antigravity
Fixes mangling collisions when two thunks with the same name (eg, `Op`)
are created in the same context, which in turn would lead to LLVM
verifier failures and miscompiles.
To support this, add a new value store to track a little more
information about thunks beyond what's in the `Function`.
This lets us stop eliding it in textual semir tests with dump ranges.
Previously it would always get elided, even though it was part of the
range being dumped, and was referred to by other instructions in the
dump range.
Since each `.Self` is unique (can change its type if not its value) in a
facet type, having each one distinct by location also aids
understanding.
See
[here](https://docs.google.com/document/d/1rWcueFwIfZox6GKVGxiUG4cBzjrZ6djXiIDGyJDtrE4/edit?tab=t.0)
for the design doc.
This also removes the default value of the `result_type_inst_id`
parameter of `HandleAction`, moves it before the action in the parameter
list, and documents it. This solves two problems:
- The default made it easy to forget, leading to unnecessary
`TypeOfInst` instructions.
- When it was present, putting it after the fairly "bulky" action
argument tended to make the callsite harder to read.
When importing a C++ function with an rvalue reference parameter, we
previously produced a Carbon value parameter. This would lead to the
toolchain believing it could pass the address of a non-expiring object
to the function, which would lead to a use-after-move.
Instead, we now map non-const rvalue reference parameters to Carbon
`var` parameters. This forces the object passed into C++ to be unique
and owned by the call. While that's not an exact match for C++ rvalue
reference parameters, given that it provides "always move" not
"conditionally move", it's the closest match we have at the moment.
When creating the C++ thunk, make the parameters references if the
corresponding callee parameters are `ref`s.
When creating the Carbon thunk, tag the call arguments as `ref` if the
corresponding callee parameters are `ref`s.
The FunctionDecl created for calling the Carbon thunk now takes a `self`
parameter for non-static methods, and the C++ thunk now passes an extra
argument for that `self` parameter when needed.
The CXXMethodDecl thunk created for calling methods now sets the storage
class appropriate depending on whether the method is static or not.
To reduce the number of parameters being passed around to thunk-building
functions, added a `FunctionInfo` struct and pass that around instead.
Add import/ and export/ under function/. Move most top-level tests to a
new basics/ with subdirectories for `import` directives and `inline
Cpp`. Add subdirectory for primitive type handling. Move all `reverse/`
tests to somewhere else, typically under an `export/` directory.
I split two test files up: constexpr.carbon got split into var/ and
function/ pieces, and reverse/simple.carbon was inlined into
namespace/export.carbon. The rest are just simple renames.
Instead of allowing lower to pick whatever type layout it desires,
compute the layouts of types as part of completing the type, and make
lower build types that match that representation.
For now we assume that all pointers are 64-bit, since we don't have
access to target information. We allow tail padding reuse for structs
and tuple types (and by extension, for classes, since they use structs
as their object representation), but not for arrays.
In order to build matching LLVM types, we create LLVM packed structs
where necessary, and we insert inter-field padding on the end of the
previous field so that GEP indexes still always match Carbon's
ElementIndexes.
We don't yet use the computed alignment much in LLVM IR generation -- in
particular, `alloca`s, `load`s, and `store`s should probably use the
computed type alignment, but don't.
Assisted-by: Gemini via Antigravity
Instead of exporting a class or namespace each time a new C++ name
lookup discovers it, track that we have exported the entity on its name
scope, and if a new name lookup finds the same entity, produce the same
clang declaration.
Also declare them `inline` since we're putting the `always_inline`
attribute on them. Use the `internal_linkage` attribute rather than
`SC_Static` since it's a more precise mechanism and matches what we do
for static member functions in reverse interop (where `SC_Static` means
something else and would not give the function internal linkage).
This is already allowed as a builtin conversion, but the impl allows the
generics system to know about it, so that conversions like
`Optional(T*)` to `Optional(const T*)` are allowed. This in turn allows
a C++ `T*` to be implicitly converted to a C++ `const T*` in Carbon
code.
Instead of recursing back into Convert, make CppThunkRef conversion just
add an extra step to category conversion, performing a copy conversion
followed by an ephemeral reference binding conversion.
This is only fixing the decision about *whether* to produce a witness.
Implementation of the witness is still a TODO, though where a body is
generated, it should also precisely reflect where one _needs_ to be
generated.
Note the tests:
- toolchain/lower/testdata/function/generic/import_core_witness.carbon
- toolchain/lower/testdata/function/generic/import_unused_def.carbon
These tests can probably be produced _without_ Core.Destroy, but I found
the essence of them while trying to build //examples with Core.Destroy
and a simpler minimization wasn't striking me.
Assisted-by: Google Antigravity with Gemini
---------
Co-authored-by: jonmeow <jperkins@google.com>
This will be used for const-evaling functions. Splitting into a separate
commit since it touches a lot of test files, and a couple fail_todo
tests are no longer failing.
Instead of injecting code to declare an `operator new`, generate AST for
it directly. In order to use this, directly generate a `CXXNewExpr`
rather than asking Clang to build one.
This is less of a hack, and doesn't visibly leak an `operator new`
declaration that inline C++ code or template instantiations might see.
It also avoids generating a warning in C++26 and later that the
`constexpr` declaration of `operator new` is used but not defined.
Assisted-by: Gemini 3.1 Pro via Antigravity
When a `var` is not explicitly given an initializer, initialize it in
one of two ways:
* If its type implements the new interface `Core.Default`, call
`Core.Default.Op` to initialize it.
* Otherwise, if its type implements `UnformedInit`, leave it in an
unformed state. For now, this is always an uninitialized state, but that
will change in the future.
* If neither of those apply, the `var` declaration is ill-formed.
This is a step towards implementing leads decision #6739 and proposals
#257 and #5913.
Assisted-by: Gemini 3.1 Pro via Antigravity
---------
Co-authored-by: Geoff Romer <gromer@google.com>
Previously we forced a temporary materialization, resulting in it being
treated as an ephemeral reference expression. This change allows
```carbon
var x: Class = {} as Class;
```
even when `Class` is not copyable.
In C++ overload resolution, when mapping a Carbon value expression into
a C++ argument, produce a const-qualified argument where possible. This
has two effects:
* Overload resolution does not consider non-const-qualified member
functions to be viable for a prvalue self any more. This is desirable
since such functions are not actually callable with a prvalue self, and
permits overload resolution to pick a const-qualified overload instead.
* Overload resolution does not allow a Carbon value expression to be
passed to a C++ `T&&` parameter any more. This is desirable since it's
not correct to move from a value expression. Previously we allowed this
and moved from the value!
This case is redundant: when deducing against a runtime parameter
pattern, the type is all that matters, and the type is added to the
deduction earlier. Additionally deducing the same argument against
parameter's subpattern just creates duplicate work, because the
subpattern has the same type.
Add an `IntFitsIn` interface with a custom witness, such that `T impls
IntFitsIn(U)` if `T` is an integer type all of whose values fit
losslessly into the integer type `U`. Use it to constrain implicit
conversions between integer types.
So far, this has not been extended to the
`CppCompat.[U]{Long32,LongLong64}` types, only to `Core.Int(N)` and
`Core.UInt(N)`.
Assisted-by: Gemini 3 Pro via Antigravity
Treat the initial sequence ofarguments in a call to a C++ function up to
and including the last argument that is a type or template as being the
explicit template arguments for the call, rather than rejecting them
because they can't be converted to the parameter types.
Implements the current direction on leads issue #6768, except that no
syntax for explicitly annotating an argument as being a template
argument is provided.
---------
Co-authored-by: Carbon Infra Bot <carbon-external-infra@google.com>
This is a step toward removing the index from `InitForm`, so that equal
form values always have equal representations.
---------
Co-authored-by: Jon Ross-Perkins <jperkins@google.com>
This is iterating on how `Destroy.Op` generates, to start adding body
capabilities. This changes the way the signature is created, and adds a
`CoreWitness` function kind so that mangling can prevent name
collisions. The result is that what _was_ `DestroyOp` is now
`Core.Destroy.Op` or, as can be seen in
toolchain/lower/testdata/interop/cpp/nullptr.carbon,
`_COp.<hash>:core.Destroy.Core` where `:core` is indicating that it's a
core witness (taking a note from `:thunk`).
Assisted-by: Google Antigravity with Gemini 3 Flash
Defer creating the CppContext until we have all of its components, so
that we know they're not null. Don't track the action on the context,
since it's not a reliable way of getting back to the compiler invocation
on failure. Don't flush the diagnostics emitter from the emitter
destructor since the derived class emitter will already have been
destroyed at that point. Distinguish between clang setup failing and
clang merely producing errors, and don't connect the check context to
clang if clang setup failed.
---------
Co-authored-by: David Blaikie <dblaikie@gmail.com>
This allows us to capture the location at which a type literal was used,
even in the cases where we don't otherwise need to create a new
instruction to represent the type such as for `char` or `str`.
The logic used to build the underlying type is now marked as desugaring.
For cases such as `iN`, this causes the call to `Core.Int` to no longer
be added as a dedicated IR instruction, and instead its constant value
is used directly as the value of the `type_literal`. This results in
this being on balance a reduction in the size of the IR.
This also fixes a crash in C++ interop when using a `char` literal as a
template argument. The crash was caused by the template argument not
having an associated location when mapping to a C++ location. See
changes to check/testdata/interop/cpp/template/type_param.carbon for an
example that used to crash before this change.
Update alias handling to allow an alias to point at any type literal,
reinstating support for aliases for type literals such as `bool` and
`i32` that had previously worked but stopped working when we
transitioned those types to being defined in the prelude. See changes to
toolchain/check/testdata/alias/builtins.carbon.
All the test changes other than the two mentioned above are mechanical
autoupdate changes switching to the new instruction.
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.
Implementation of unused pattern bindings #2022, continued.
Whereas previous PR #6460 took care of parsing, and PR #6479 prepared
the stage by using _ in some test cases, this PR has the the actual
implementation, using a simple dataflow analysis.
---------
Co-authored-by: Burak Emir <bqe@google.com>
Co-authored-by: jonmeow <jperkins@google.com>
Currently each interface has a `Self` facet internally that becomes a
binding to every entity inside the interface: associated constants,
functions, and require decls. Each of these has to be independently
generic as a result. This makes is challenging in extended name lookup
to move into an extended scope of an interface, as we have a specific
for the interface, but the names within require a different specific
that includes a `Self` facet value.
We generalize this relationship by adding a second generic to Interface,
called `generic_with_self`. When we want to work with entities inside
the interface, we move from the interface-without-specific to the
interface-with-self specific by adding a Self to the specific. This is
done independently of any particular entity inside the Interface, as
those entities are now all members of the interface-with-self generic.
Associated constants no longer need a generic of their own, as they do
not have separate generic bindings. Functions retain a generic, but if
the function has no generic arguments, it will have no bindings of its
own now.
Require decls retain a generic so that their specific can be
instantiated separately from the interface. Requiring the interface to
be complete does not require the types in a require decl to be complete
unless it is modified by `extend`. So we allow them to be completed
later by keeping them in a separate generic.
Named constraints look like interfaces and gain the additional inner
generic-with-self, with the same relationship to require decls.
This removes the need for name lookup to perform Substitution of a Self
facet into the extended scope instruction. Instead, the
`SpecificConstant` instruction inserted by a `require` decl is part of
the interface-with-self generic. When looking through a FacetType for
extended scopes, for each interface, we push the scope with the specific
for the interface-with-self. Then the constant value of the
`SpecificConstant` is correctly modified by the provided self
automatically through applying that specific.
This TODO had been written before C++ types were generating destroy
implementations, which is resolved now.
Assisted-by: Google Antigravity with Gemini 3 Flash
This is related to #6727, but is generally a necessary fix even without
that issue. I'm not adding a specific test of #6727 because it should
also be covered by the tests in #6726.
Assisted-by: Google Antigravity with Gemini 3 Flash
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>
Richard Smith
Dana Jansens
Geoff Romer
Nicholas Bishop
Jon Ross-Perkinsandjonmeow
