Based on Peter Collingbourne's destructor patches.
Prior to this change, clang was considering ?1 to be the complete
destructor and the base destructor, which was wrong. This lead to
crashes when clang tried to emit two LLVM functions with the same name.
In this ABI, TUs with non-inline dtors might not emit a complete
destructor. They are emitted as inline thunks in TUs that need them,
and they always delegate to the base dtors of the complete class and its
virtual bases. This change uses the DeferredDecls machinery to emit
complete dtors as needed.
Currently in clang try body destructors can catch exceptions thrown by
virtual base destructors. In the Microsoft C++ ABI, clang may not have
the destructor definition, in which case clang won't wrap the virtual
virtual base destructor calls in a try-catch. Diagnosing this in user
code is TODO.
Finally, for classes that don't use virtual inheritance, MSVC always
calls the base destructor (?1) directly. This is a useful code size
optimization that avoids emitting lots of extra thunks or aliases.
Implementing it also means our existing tests continue to pass, and is
consistent with MSVC's output.
We can do the same for Itanium by tweaking GetAddrOfCXXDestructor, but
it will require further testing.
Reviewers: rjmccall
CC: cfe-commits
Differential Revision: http://llvm-reviews.chandlerc.com/D1066
llvm-svn: 186828
A constructor for an abstract class does not call constructors for virtual
base classes, so it is not an error if no initializer is present for the
virtual base and the virtual base cannot be default initialized.
Also provide a (disabled by default, for now) warning for the case where a
virtual base class's initializer is ignored in an abstract class's constructor,
and address a defect in DR257 where it was not carried through to C++11's rules
for implicit deletion of special member functions.
Based on a patch by Maurice Bos.
llvm-svn: 186803
optimize, to follow the permissions granted in N3664. Under those rules, only
calls generated by new-expressions and delete-expressions are permitted to be
optimized, and direct calls to ::operator new and ::operator delete must be
treated as normal calls.
llvm-svn: 186799
The functionality is equivalent to the GCC attribute. Variables of tagged
types will be warned about as unused if they are not used in any way
except for possible (even non-trivial) ctors/dtors called. Useful for tagging
classes like std::string (which is not part of this commit).
llvm-svn: 186765
Every #include is surrounded by #if 0 in order to comment it out, which adds
lines. That is fixed up right after, but that all can be inside #if part
that is not processed, so fix up also after every end of a conditional part.
llvm-svn: 186763
A class with a field of non-POD-for-layout type is not POD-for-layout.
This computation should not depend on whether the field is of POD type
in the language sense.
Fixes PR16537.
Patch by Josh Magee.
llvm-svn: 186741
This is the same way GenericSelectionExpr works, and it's generally a
more consistent approach.
A large part of this patch is devoted to caching the value of the condition
of a ChooseExpr; it's needed to avoid threading an ASTContext into
IgnoreParens().
Fixes <rdar://problem/14438917>.
llvm-svn: 186738
This patch essentially removes all the FIXMEs following calls to DeduceTemplateArguments() that want to keep track of deduction failure info.
llvm-svn: 186730
Summary: In ARC mode, clang emits a warning if the result of an 'init' method is unused but miss cases where the method does not follows the Cocoa naming convention but is properly declared as an init family method.
CC: cfe-commits, eli.friedman
Differential Revision: http://llvm-reviews.chandlerc.com/D1163
llvm-svn: 186718
Canonical types are unchanged. The type printer had to be changed to
avoid printing any non-default implicit calling convention as well as
the calling convention attribute.
Reviewers: rjmccall
Differential Revision: http://llvm-reviews.chandlerc.com/D1132
llvm-svn: 186714
When we see a pack, and replace it with a template argument which is
also a pack, we want to use the pack pattern, not the expanded pack.
The caller should take care of expanding the pack afterwards.
Fixes PR16646.
llvm-svn: 186713
and add a new option --driver-mode= to control it explicitly.
The CCCIsCXX and CCCIsCPP flags were non-overlapping, i.e. there
are currently really three modes that Clang can run in: gcc, g++
or cpp, so it makes sense to represent them as an enum.
Having a command line flag to control it helps testing.
llvm-svn: 186605
MSBuild writes response files as UTF-16 little endian with a byte order
mark. With this change, clang will be able to read them, although we
still can't parse any of their flags.
Adds a UTF-16-LE response file with a BOM for testing.
Differential Revision: http://llvm-reviews.chandlerc.com/D1137
llvm-svn: 186603
r186331).
Original commit log:
If we friend a declaration twice, that should not make it visible to
name lookup in the surrounding context. Slightly rework how we handle
friend declarations to inherit the visibility of the prior
declaration, rather than setting a friend declaration to be visible
whenever there was a prior declaration.
llvm-svn: 186546
cl.exe treats wide bitfields as an error. This patch causes them to be
an error if IsMsStruct is true, as it is in straight C.
Patch by Warren Hunt!
Reviewers: eli.friedman
Differential Revision: http://llvm-reviews.chandlerc.com/D1125
llvm-svn: 186536
Previously, we would simply abort the path when we saw a default member
initialization; now, we actually attempt to evaluate it. Like default
arguments, the contents of these expressions are not actually part of the
current function, so we fall back to constant evaluation.
llvm-svn: 186521
Previously, SValBuilder knew how to evaluate StringLiterals, but couldn't
handle an array-to-pointer decay for constant values. Additionally,
RegionStore was being too strict about loading from an array, refusing to
return a 'char' value from a 'const char' array. Both of these have been
fixed.
llvm-svn: 186520
Previously, the use of a std::initializer_list (actually, a
CXXStdInitializerListExpr) would cause the analyzer to give up on the rest
of the path. Now, it just uses an opaque symbolic value for the
initializer_list and continues on.
At some point in the future we can add proper support for initializer_list,
with access to the elements in the InitListExpr.
<rdar://problem/14340207>
llvm-svn: 186519
Summary:
This patch enables ExprEndgine to reason about temporary object destructors.
However, these destructor calls are never inlined, since this feature is still
broken. Still, this is sufficient to properly handle noreturn temporary
destructors and close bug #15599. I have also enabled the cfg-temporary-dtors
analyzer option by default.
Reviewers: jordan_rose
CC: cfe-commits
Differential Revision: http://llvm-reviews.chandlerc.com/D1131
llvm-svn: 186498
_Complex load/store didn't have their alignment set properly, which was visible when GCC's torture tests use volatile _Complex.
Update some existing tests to check for alignment, and add a new test which also has over-aligned volatile _Complex (since the imaginary part shouldn't be overaligned, only the real part).
llvm-svn: 186490
Sema needs to be able to accurately determine what will be
emitted as a constant initializer and what will not, so
we get accurate errors in C and accurate -Wglobal-constructors
warnings in C++. This makes Expr::isConstantInitializer match
CGExprConstant as closely as possible.
llvm-svn: 186464
Unfortunately I don't think there's a good way to validate branch targets on
release builds. Fortunately it's a minor part of this test (and based on
generic code) so I don't mind dropping it.
llvm-svn: 186398
This adds three overloaded intrinsics to Clang:
T __builtin_arm_ldrex(const volatile T *addr)
int __builtin_arm_strex(T val, volatile T *addr)
void __builtin_arm_clrex()
The intent is that these do what users would expect when given most sensible
types. Currently, "sensible" translates to ints, floats and pointers.
llvm-svn: 186394
Not completely sure this is right, but it's clearly better than what
we did before this commit (effectively dropping the attribute).
<rdar://problem/14413117>
llvm-svn: 186373
The record layout code didn't properly take into account that
an empty class at offset 0 can have an alignment greater than 1.
Patch by Andrea Di Biagio.
llvm-svn: 186370
Make sure we call BuildFieldReferenceExpr with the appropriate decl
when a member of an anonymous union is made public with a using decl.
Also, fix a crash on invalid field access into an anonymous union.
Fixes PR16630.
llvm-svn: 186367
recovery is not attempted with the fixit. Also move the associated test
case from FixIt/fixit.cpp to SemaCXX/member-expr.cpp since the fixit is
no longer automatically applied.
llvm-svn: 186342
This breaks the build of basic patterns with repeated friend
declarations. See the added test case in SemaCXX/friend.cpp or the test
case reported to the original commit log.
Original commit log:
If we friend a declaration twice, that should not make it visible to
name lookup in the surrounding context. Slightly rework how we handle
friend declarations to inherit the visibility of the prior
declaration, rather than setting a friend declaration to be visible
whenever there was a prior declaration.
llvm-svn: 186331
does not substitute a sizeof-pack expression.
The solution is proposed by Richard Smith.
Differential Revision: http://llvm-reviews.chandlerc.com/D869
llvm-svn: 186306
global allocation or deallocation function, that should not cause that global
allocation or deallocation function to become unavailable.
llvm-svn: 186270
This simplifies the core benefit of -flimit-debug-info by taking a more
systematic approach to avoid emitting debug info definitions for types
that only require declarations. The previous ad-hoc approach (3 cases
removed in this patch) had many holes.
The general approach (adding a bit to TagDecl and callback through
ASTConsumer) has been discussed with Richard Smith - though always open
to revision.
llvm-svn: 186262
Test coverage for non-dependent pack expansions doesn't demonstrate a
failure prior to this patch (a follow-up commit improving debug info
will cover this commit specifically) but covers a related hole in our
test coverage.
Reviewed by Richard Smith & Eli Friedman.
llvm-svn: 186261
decls. That can reenter deserialization and explode horribly by trying to merge
a declaration that we've not got very far through deserializing yet.
llvm-svn: 186236
They don't seem to be used for back references, presumably because a
function template is unlikely to reoccur, while a class template name
may reoccur as a type.
This fixes a mangling issue for llvm::hash_combine() in Hashing.h.
Reviewers: timurrrr
Differential Revision: http://llvm-reviews.chandlerc.com/D1078
llvm-svn: 186233
Original commit log:
If we friend a declaration twice, that should not make it visible to
name lookup in the surrounding context. Slightly rework how we handle
friend declarations to inherit the visibility of the prior
declaration, rather than setting a friend declaration to be visible
whenever there was a prior declaration.
llvm-svn: 186199
fix is.
Original commit log:
If we friend a declaration twice, that should not make it visible to
name lookup in the surrounding context. Slightly rework how we handle
friend declarations to inherit the visibility of the prior
declaration, rather than setting a friend declaration to be visible
whenever there was a prior declaration.
llvm-svn: 186185
Make sure we don't crash when checking whether an assignment operator
without any arguments is a special member. <rdar://problem/14397774>.
llvm-svn: 186137
Various pieces of code, like base initialization in Sema and RTTI IRGen,
don't properly ignore qualifiers on base classes. Instead of auditing the
whole codebase, just strip them off in the getter. (The type as written is
still available in the TypeSourceInfo for code that cares.)
Fixes PR16596.
llvm-svn: 186125
followed by an identifier, then diagnose an identifier as being a bogus part of
the declarator instead of tripping over it. Improves diagnostics for cases like
std::vector<const int *p> my_vec;
llvm-svn: 186061
& operator (ignoring any overloaded operator& for the type). The purpose of
this builtin is for use in std::addressof, to allow it to be made constexpr;
the existing implementation technique (reinterpret_cast to some reference type,
take address, reinterpert_cast back) does not permit this because
reinterpret_cast between reference types is not permitted in a constant
expression in C++11 onwards.
llvm-svn: 186053
lookup in the surrounding context. Slightly rework how we handle friend
declarations to inherit the visibility of the prior declaration, rather
than setting a friend declaration to be visible whenever there was a prior
declaration.
llvm-svn: 186040
migrate to 'copy attribute if Object
class implements NSCopying otherwise
assume implied 'strong'. Remove
lifetime qualifier on property as it has
moved to property's attribute. Added TODO
comment for future work by poking into
setter implementation.
llvm-svn: 186037
Previously, we asserted that whenever 'new' did not include a constructor
call, the type must be a non-record type. In C++11, however, uniform
initialization syntax (braces) allow 'new' to construct records with
list-initialization: "new Point{1, 2}".
Removing this assertion should be perfectly safe; the code here matches
what VisitDeclStmt does for regions allocated on the stack.
<rdar://problem/14403437>
llvm-svn: 186028
Specifically, handle the case where the block is in a default argument
in a class method. The mangling here follows what we do for lambdas.
llvm-svn: 185991
Compute mangling numbers for externally visible local variables and tags.
Change the mangler to consistently use discriminators where necessary.
Tweak the scheme we use to number decls which are not externally visible
to avoid unnecessary discriminators in common cases now that we request
them more consistently.
Fixes <rdar://problem/14204721>.
llvm-svn: 185986
method declaration into its implementation to
prevent a bogus warning about mismatched attributes.
then make sure the warning about missing call to super comes out
of the method implementation. // rdar://14251387
llvm-svn: 185974
Do not generate VLAs as complex variables any more, as they are now
correctly represented as breg+0 locations in the backend.
(Paired commit with LLVM: r185966)
rdar://problem/13658587
llvm-svn: 185967
list is the name of a class, not a namespace. Change the test as well - the previous
version did not test properly.
Fixes radar://14317928.
llvm-svn: 185898
Combined with typo correction's new ability to apply global/absolute nested
name specifiers to possible corrections, cases such as in PR12287 where the
desired function is being shadowed by a lexically closer function with the
same name but a different number of parameters will now include a FixIt.
On a side note, since the test for this change caused
test/SemaCXX/typo-correction.cpp to exceed the typo correction limit for
a single file, I've included a test case for exceeding the limit and added
some comments to both the original and part two of typo-correction.cpp
warning future editors of the files about the limit.
llvm-svn: 185881
Use UsualArithmeticConversions unconditionally in analysis of
comparisons and conditional operators: the method performs
the usual arithmetic conversions if both sides are arithmetic, and
usual unary conversions if they are not. This is just a cleanup
for conditional operators; for comparisons, it fixes the issue that
we would try to check isArithmetic() on an atomic type.
Also, fix GetExprRange() in SemaChecking.cpp so it deals with variables
of atomic type correctly.
Fixes PR15537.
llvm-svn: 185857
It would emit @llvm.memcpy with "-triple x86_64-(mingw32|win32)" and had been failing since Nick's r185735.
; Function Attrs: nounwind
declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture, i8* nocapture readonly, i64, i32, i1) #1
llvm-svn: 185796
Sema::MergeFunctionDecl attempts merging two decls even if the old decl
is invalid. This can lead to interesting circumstances where we
successfully merge the decls but the result makes no sense.
Take the following for example:
template <typename T>
int main(void);
int main(void);
Sema will not consider these to be overloads of the same name because
main can't be overloaded, which means that this must be a redeclaration.
In this case the templated decl is compatible with the non-templated
decl allowing the Sema::CheckFunctionDeclaration machinery to move on
and do bizarre things like setting the previous decl of a non-templated
decl to a templated decl!
The way I see it, we should just bail from MergeFunctionDecl if the old
decl is invalid.
This fixes PR16531.
llvm-svn: 185779
This boils down to us sending invalid function decls to
CheckFunctionDeclaration becauswe we did not consider that CheckMain
could cause the decl to be invalid. Instead, interogate the new decl's
main-validity and *then* send it over to get CheckFunctionDeclaration'd
if it was still valid after calling CheckMain.
llvm-svn: 185745
As it turns out, the NoFunction bit for local class mangling needed to be
propagated into more places. r185450 turned what used to be an incorrect
mangling into an assertion.
llvm-svn: 185713
Some versions of python will expand the glob used in the test, others wont,
causing the test to fail when run with LIT_USE_INTERNAL_SHELL=1.
llvm-svn: 185653
Adrian Prantl