GCC's attribute 'target', in addition to being an optimization hint,
also allows function multiversioning. We currently have the former
implemented, this is the latter's implementation.
This works by enabling functions with the same name/signature to coexist,
so that they can all be emitted. Multiversion state is stored in the
FunctionDecl itself, and SemaDecl manages the definitions.
Note that it ends up having to permit redefinition of functions so
that they can all be emitted. Additionally, all versions of the function
must be emitted, so this also manages that.
Note that this includes some additional rules that GCC does not, since
defining something as a MultiVersion function after a usage has been made illegal.
The only 'history rewriting' that happens is if a function is emitted before
it has been converted to a multiversion'ed function, at which point its name
needs to be changed.
Function templates and virtual functions are NOT yet supported (not supported
in GCC either).
Additionally, constructors/destructors are disallowed, but the former is
planned.
llvm-svn: 322028
Attempting to recompute it are doomed to fail because the IDNS of a declaration
is not necessarily preserved across serialization and deserialization (in turn
because whether a friend declaration is visible depends on whether some prior
non-friend declaration exists).
llvm-svn: 321921
Extend the hashing to functions, which allows detection of function definition
mismatches across modules. This is a re-commit of r320230.
llvm-svn: 321395
whether they have an initializer.
We cannot distinguish between a declaration of a variable template
specialization and a definition of one that lacks an initializer without this,
and would previously mistake the latter for the former.
llvm-svn: 319605
The anonymous union did NOT save us storage, but instead behaved as if we added an additional integer data member to FunctionDecl.
For additional context, the anonymous union renders the bit fields as non-adjacent and prevents them from sharing the same 'memory location' (i.e. bit-storage) by requiring the anonymous union object to be appropriately aligned.
This was confirmed through discussion with Richard Smith in Albuquerque (ISO C++ Meeting)
https://reviews.llvm.org/rL316292
llvm-svn: 317984
for instantiating its definition.
We model the 'inline'ness as being instantiated with the static data member in
order to track whether the declaration has become a definition yet.
llvm-svn: 317147
In order to identify the copy deduction candidate, I considered two approaches:
- attempt to determine whether an implicit guide is a copy deduction candidate by checking certain properties of its subsituted parameter during overload-resolution.
- using one of the many bits (WillHaveBody) from FunctionDecl (that CXXDeductionGuideDecl inherits from) that are otherwise irrelevant for deduction guides
After some brittle gymnastics w the first strategy, I settled on the second, although to avoid confusion and to give that bit a better name, i turned it into a member of an anonymous union.
Given this identification 'bit', the tweak to overload resolution was a simple reordering of the deduction guide checks (in SemaOverload.cpp::isBetterOverloadCandidate), in-line with Jason Merrill's p0620r0 drafting which made it into the working paper. Concordant with that, I made sure the copy deduction candidate is always added.
References:
See https://bugs.llvm.org/show_bug.cgi?id=34970
See http://wg21.link/p0620r0
llvm-svn: 316292
This feature is not (yet) approved by the C++ committee, so this is liable to
be reverted or significantly modified based on committee feedback.
No functionality change intended for existing code (a new type must be defined
in namespace std to take advantage of this feature).
llvm-svn: 315662
This is breaking a build of https://github.com/abseil/abseil-cpp and so
likely not really NFC. Also reverted subsequent r314956/7.
I'll forward reproduction instructions to Richard.
llvm-svn: 315439
In its place, track on the canonical function declaration whether there is a
declaration with a body (and if so, which one). This brings function definition
handling in line with what we do in all other contexts, and is necessary to
allow us to merge declarations within multiple definitions of the same function
(eg, PR33924).
No functionality change intended.
llvm-svn: 314955
move constructor.
Previously user-defined reduction initializer was considered as an
assignment expression, not as initializer. Fixed this by treating the
initializer expression as an initializer.
llvm-svn: 312638
constructors when deciding whether classes should be passed indirectly.
This fixes ABI differences between Clang and GCC:
* Previously, Clang ignored the move constructor when making this
determination. It now takes the move constructor into account, per
https://github.com/itanium-cxx-abi/cxx-abi/pull/17 (this change may
seem recent, but the ABI change was agreed on the Itanium C++ ABI
list a long time ago).
* Previously, Clang's behavior when the copy constructor was deleted
was unstable -- depending on whether the lazy declaration of the
copy constructor had been triggered, you might get different behavior.
We now eagerly declare the copy constructor whenever its deletedness
is unclear, and ignore deleted copy/move constructors when looking for
a trivial such constructor.
This also fixes an ABI difference between Clang and MSVC:
* If the copy constructor would be implicitly deleted (but has not been
lazily declared yet), for instance because the class has an rvalue
reference member, we would pass it directly. We now pass such a class
indirectly, matching MSVC.
Based on a patch by Vassil Vassilev, which was based on a patch by Bernd
Schmidt, which was based on a patch by Reid Kleckner!
This is a re-commit of r310401, which was reverted in r310464 due to ARM
failures (which should now be fixed).
llvm-svn: 310983
constructors when deciding whether classes should be passed indirectly.
This fixes ABI differences between Clang and GCC:
* Previously, Clang ignored the move constructor when making this
determination. It now takes the move constructor into account, per
https://github.com/itanium-cxx-abi/cxx-abi/pull/17 (this change may
seem recent, but the ABI change was agreed on the Itanium C++ ABI
list a long time ago).
* Previously, Clang's behavior when the copy constructor was deleted
was unstable -- depending on whether the lazy declaration of the
copy constructor had been triggered, you might get different behavior.
We now eagerly declare the copy constructor whenever its deletedness
is unclear, and ignore deleted copy/move constructors when looking for
a trivial such constructor.
This also fixes an ABI difference between Clang and MSVC:
* If the copy constructor would be implicitly deleted (but has not been
lazily declared yet), for instance because the class has an rvalue
reference member, we would pass it directly. We now pass such a class
indirectly, matching MSVC.
llvm-svn: 310401
These cases occur frequently for declarations in the global module (above the
module-declaration) in a Modules TS module interface. When we merge a
definition from another module into such a module-private definition, ensure
that we transitively make everything lexically within that definition visible
to that translation unit.
llvm-svn: 307129
It was reverted in r305460 but the issue appears to only break our self-host
libcxx modules bot. Reapplying it will give us a chance to get a reproducer and
fix the issue.
llvm-svn: 306903
We use this when running a preprocessor-only action on an AST file in order to
avoid paying the runtime cost of loading the extra information.
llvm-svn: 306760
Summary:
In change 2ba19793512, the ASTReader logic for ObjC interfaces was modified to
preserve the first definition-data read, "merging" later definitions into it
rather than overwriting it (though this "merging" is, in practice, a no-op that
discards the later definition-data).
Unfortunately this change was only made to ObjC interfaces, not protocols; this
means that when (for example) loading a protocol that references an interface,
if both the protocol and interface are multiply defined (as can easily happen
if the same header is read from multiple contexts), an _inconsistent_ pair of
definitions is loaded: first-read for the interface and last-read for the
protocol.
This in turn causes very subtle downstream bugs in the Swift ClangImporter,
which filters the results of name lookups based on the owning module of a
definition; inconsistency between a pair of related definitions causes name
lookup failures at various stages of compilation.
To fix these downstream issues, this change replicates the logic applied to
interfaces in change 2ba19793512, but for ObjC protocols.
rdar://30851899
Reviewers: doug.gregor, rsmith
Reviewed By: doug.gregor
Subscribers: jordan_rose, cfe-commits
Differential Revision: https://reviews.llvm.org/D34741
llvm-svn: 306583
declarations that are owned but unconditionally visible.
This allows us to set declarations as visible even if they have a local owning
module, without losing information. In turn, that means that our Objective-C
support can keep on incorrectly assuming the "hidden" bit on the declaration is
the whole story with regard to name visibility. This will also be useful once
we support the C++ Modules TS export semantics.
Objective-C name visibility is still incorrect in any case where the "hidden"
bit is not the complete story: for instance, in Objective-C++ the set of
visible categories will be wrong during template instantiation, and with local
submodule visibility enabled it will be wrong when building modules. Fixing that
will require a major overhaul of how visibility is handled for Objective-C (and
particularly for categories).
llvm-svn: 306075
Currently, we load all template specialization if we have more than one module
attached and we touch anything around the template definition.
This patch registers the template specializations as lazily-loadable entities.
In some TUs it reduces the amount of deserializations by 1%.
llvm-svn: 305120
Summary:
If the first parameter of the function is the ImplicitParamDecl, codegen
automatically marks it as an implicit argument with `this` or `self`
pointer. Added internal kind of the ImplicitParamDecl to separate
'this', 'self', 'vtt' and other implicit parameters from other kind of
parameters.
Reviewers: rjmccall, aaron.ballman
Subscribers: cfe-commits
Differential Revision: https://reviews.llvm.org/D33735
llvm-svn: 305075
As discussed in D30793, we have some unsafe calls to isConsumerInterestedIn().
This patch implements Richard's suggestion (from the inline comment) that we
should track if we just deserialized an declaration. If we just deserialized,
we can skip the unsafe call because we know it's interesting. If we didn't just
deserialize the declaration, calling isConsumerInterestedIn() should be safe.
We tried to create a test case for this but we were not successful.
Patch by Raphael Isemann (D32499)!
llvm-svn: 303432
This patch implements the suggestion in D29753 that calling DeclMustBeEmitted in
the middle of deserialization should be avoided and that the actual check should
be deferred until it's safe to do so.
This patch fixes a crash when accessing the invalid redecl chains while trying
to evaluate the value of a const VarDecl that contains a function call.
Patch by Raphael Isemann (D30793)!
llvm-svn: 300110
Calculating the hash in Sema::ActOnTagFinishDefinition could happen before
all sub-Decls were parsed or processed, which would produce the wrong hash
value. Change to calculating the hash on the first use and storing the value
instead. Also, avoid using the macros that were only for Boolean fields and
use an explicit checker during the DefintionData merge. No functional change,
but was this blocking other ODRHash patches.
llvm-svn: 299989
Matching the function-homing support for modular codegen. Any type
implicitly (implicit template specializations) or explicitly defined in
a module is attached to that module's object file and omitted elsewhere
(only a declaration used if necessary for references).
llvm-svn: 299987
Some decls are created not where they are written, but in other module
files/users (implicit special members and function template implicit
specializations). To correctly identify them, use a bit next to the definition
to track the modular codegen property.
Discussed whether the module file bit could be omitted in favor of
reconstituting from the modular codegen decls list - best guess today is that
the efficiency improvement of not having to deserialize the whole list whenever
any function is queried by a module user is worth it for the small size
increase of this redundant (list + bit-on-def) representation.
Reviewers: rsmith
Differential Revision: https://reviews.llvm.org/D29901
llvm-svn: 299982
This enhances the AST to keep track of locations of the names in those ObjC property attributes, and reports them for indexing.
Patch by Nathan Hawes!
https://reviews.llvm.org/D30907
llvm-svn: 297972
This is a stopgap fix for PR31863, a regression introduced in r276159.
Consider this snippet:
struct FVector;
struct FVector {};
struct FBox {
FVector Min;
FBox(int);
};
namespace {
FBox InvalidBoundingBox(0);
}
While parsing the DECL_VAR for 'struct FBox', clang recursively read all the
dep decls until it finds the DECL_CXX_RECORD forward declaration for 'struct
FVector'. Then, it resumes all the way up back to DECL_VAR handling in
`ReadDeclRecord`, where it checks if `isConsumerInterestedIn` for the decl.
One of the condition for `isConsumerInterestedIn` to return false is if the
VarDecl is imported from a module `D->getImportedOwningModule()`, because it
will get emitted when we import the relevant module. However, before checking
if it comes from a module, clang checks if `Ctx.DeclMustBeEmitted(D)`, which
triggers the emission of 'struct FBox'. Since one of its fields is still
incomplete, it crashes.
Instead, check if `D->getImportedOwningModule()` is true before calling
`Ctx.DeclMustBeEmitted(D)`.
Differential Revision: https://reviews.llvm.org/D29753
rdar://problem/30173654
llvm-svn: 296656
Essentially, as a base class constructor does not construct virtual bases, such
a constructor for an abstract class does not need the corresponding base class
construction to be valid, and likewise for destructors.
This creates an awkward situation: clang will sometimes generate references to
the complete object and deleting destructors for an abstract class (it puts
them in the construction vtable for a derived class). But we can't generate a
"correct" version of these because we can't generate references to base class
constructors any more (if they're template specializations, say, we might not
have instantiated them and can't assume any other TU will emit a copy).
Fortunately, we don't need to, since no correct program can ever invoke them,
so instead emit symbols that just trap.
We should stop emitting references to these symbols, but still need to emit
definitions for compatibility.
llvm-svn: 296275
The goal of this is to fix a bug in modules where we'd merge
FunctionDecls that differed in their pass_object_size attributes. Since
we can overload on the presence of pass_object_size attributes, this
behavior is incorrect.
We don't represent `N` in `pass_object_size(N)` as part of
ExtParameterInfo, since it's an error to overload solely on the value of
N. This means that we have a bug if we have two modules that declare
functions that differ only in their pass_object_size attrs, like so:
// In module A, from a.h
void foo(char *__attribute__((pass_object_size(0))));
// In module B, from b.h
void foo(char *__attribute__((pass_object_size(1))));
// In module C, in main.c
#include "a.h"
#include "b.h"
At the moment, we'll merge the foo decls, when we should instead emit a
diagnostic about an invalid overload. We seem to have similar (silent)
behavior if we overload only on the return type of `foo` instead; I'll
try to find a good place to put a FIXME (or I'll just file a bug) soon.
This patch also fixes a bug where we'd not output the proper extended
parameter info for declarations with pass_object_size attrs.
llvm-svn: 296076
Reserve a spot for ODR hash in CXXRecordDecl and in its modules storage.
Default the hash value to 0 for all classes.
Differential Revision: https://reviews.llvm.org/D21675
llvm-svn: 295533
A slightly weaker form of ODR checking than previous attempts, but hopefully
won't break the modules build bot. Future work will be needed to catch all
cases.
When objects are imported for modules, there is a chance that a name collision
will cause an ODR violation. Previously, only a small number of such
violations were detected. This patch provides a stronger check based on
AST nodes.
The information needed to uniquely identify an object is taken from the AST and
put into a one-dimensional byte stream. This stream is then hashed to give
a value to represent the object, which is stored with the other object data
in the module.
When modules are loaded, and Decl's are merged, the hash values of the two
Decl's are compared. Only Decl's with matched hash values will be merged.
Mismatch hashes will generate a module error, and if possible, point to the
first difference between the two objects.
The transform from AST to byte stream is a modified depth first algorithm.
Due to references between some AST nodes, a pure depth first algorithm could
generate loops. For Stmt nodes, a straight depth first processing occurs.
For Type and Decl nodes, they are replaced with an index number and only on
first visit will these nodes be processed. As an optimization, boolean
values are saved and stored together in reverse order at the end of the
byte stream to lower the ammount of data that needs to be hashed.
Compile time impact was measured at 1.5-2.0% during module building, and
negligible during builds without module building.
Differential Revision: https://reviews.llvm.org/D21675
llvm-svn: 295421
Recommit r293585 that was reverted in r293611 with new fixes. The previous
issue was determined to be an overly aggressive AST visitor from forward
declared objects. The visitor will now only deeply visit certain Decl's and
only do a shallow information extraction from all other Decl's.
When objects are imported for modules, there is a chance that a name collision
will cause an ODR violation. Previously, only a small number of such
violations were detected. This patch provides a stronger check based on
AST nodes.
The information needed to uniquely identify an object is taken from the AST and
put into a one-dimensional byte stream. This stream is then hashed to give
a value to represent the object, which is stored with the other object data
in the module.
When modules are loaded, and Decl's are merged, the hash values of the two
Decl's are compared. Only Decl's with matched hash values will be merged.
Mismatch hashes will generate a module error, and if possible, point to the
first difference between the two objects.
The transform from AST to byte stream is a modified depth first algorithm.
Due to references between some AST nodes, a pure depth first algorithm could
generate loops. For Stmt nodes, a straight depth first processing occurs.
For Type and Decl nodes, they are replaced with an index number and only on
first visit will these nodes be processed. As an optimization, boolean
values are saved and stored together in reverse order at the end of the
byte stream to lower the ammount of data that needs to be hashed.
Compile time impact was measured at 1.5-2.0% during module building, and
negligible during builds without module building.
Differential Revision: https://reviews.llvm.org/D21675
llvm-svn: 295284
Erich Keane