block variables on lhs need be ir-gen'ed after the
rhs when its 'forwarding' pointer may be modified
in rhs evaluation as result of call to Block_copy.
// rdar://9309454
Disable -Wnon-pod-memset for now while I try to reduce the false
positives still further.
The plan is to:
1) Create a more targeted warning for memset of memory pointing at
a type with virtual methods or bases where a vptr would be
overwritten.
2) Consider turning the above warning back on by default.
3) Evaluate whether any false positives in the existing warning can be
detected and white listed in the warning implementation.
4) If #3 lowers the noise floor enough, enable the full warning in -Wall
or -Wextra.
Comments or suggestions welcome. Even more welcome are specific test
cases which trigger the warning and shouldn't.
Charles Davis [Fri, 29 Apr 2011 15:50:52 +0000 (15:50 +0000)]
Remove comments about __int8 and friends from the mangler. Turns out we don't
actually have to implement them, since in modern versions of MSVC they're
aliases to the standard C types.
Relax the non-POD memset warning to use the less restrictive C++11
definition of POD. Specifically, this allows certain non-aggregate
types due to their data members being private.
The representation of C++11 POD testing is pretty gross. Any suggestions
for improvements there are welcome. Especially the name
'isCXX11PODType()' seems truly unfortunate.
Sebastian Redl [Fri, 29 Apr 2011 08:19:30 +0000 (08:19 +0000)]
Add a decl update when a static data member of a class template is instantiated in a different PCH than its containing class. Otherwise we get double definition errors. Fixes a Boost.MPL problem that affects Boost.Accumulators and probably a lot more of Boost.
Don't assume that the AST methods will only be invoked on C++ types.
Teaches isLiteralType and isTrivialType to behave plausibly and most
importantly not crash on normal RecordDecls.
Sadly I have no real way to test this. I stumbled onto it by
mis-implementing a warning.
Cut down unnecessary zero'ing when value-initializing arrays of C++ objects.
-C++ objects with user-declared constructor don't need zero'ing.
-We can zero-initialize arrays of C++ objects in "bulk" now, in which case don't zero-initialize each object again.
Use StringRef::substr() and unbounded StringRef::compare() instead of bounded version of StringRef::compare() because bounded version of StringRef::compare() is going to be removed.
Make the top-level driver ignore -fobjc-default-synthesize-properties while this feature undergoes more review and development. This is still available as a -cc1 option for testing.
Eliminates an assert in the strncpy/strncat checker caused by not validating a cast was successful. If the value of an argument was unknown, the cast would result in a NULL pointer which was later being dereferenced.
Douglas Gregor [Thu, 28 Apr 2011 17:56:11 +0000 (17:56 +0000)]
When determining whether two types are reference-compatible, check
non-CVR qualifiers as well as CVR qualifiers. For example, don't allow
a reference to an integer in address space 1 to bind to an integer in
address space 2.
Douglas Gregor [Thu, 28 Apr 2011 15:48:45 +0000 (15:48 +0000)]
libstdc++ 4.4 uses __is_signed as an identifier, while Clang treats it
as a keyword for the __is_signed type trait. Cope with this conflict
via some hackish recovery: if we see a declaration of the form
static const bool __is_signed
then we stop treating __is_signed as a keyword and instead treat it as
an identifier. It's ugly, but it's better than making the __is_signed
type trait conditional on some language flag. Fixes PR9804.
Silence more -Wnon-pod-memset given its current implementation. I may be
able to revert these based on a patch I'm working on, but no reason for
people to be spammed with warnings in the interim.
Support &__uuidof(type) as a non type template argument.
This idiom is used everywhere in MFC/COM code and as such this patch removes hundreds of errors when parsing MFC code with clang.
Example:
template <class T, const GUID* g = &__uuidof(T)>
class ComTemplate { };
John McCall [Thu, 28 Apr 2011 02:15:35 +0000 (02:15 +0000)]
When block-capturing a variable with a non-trivial destructor,
make sure to mark the destructor. This normally isn't required,
because the destructor should have been marked as part of the
declaration of the local, but it's necessary when the variable
is a parameter because it's the call sites that are responsible
for those destructors.
Douglas Gregor [Thu, 28 Apr 2011 00:56:09 +0000 (00:56 +0000)]
More cleanup of template argument deduction and its handling of
non-CVR qualifiers. We can now properly match address-space--qualified
references during template argument deduction.
John Wiegley [Thu, 28 Apr 2011 00:16:57 +0000 (00:16 +0000)]
Implementation of Embarcadero array type traits
Patch authored by John Wiegley.
These are array type traits used for parsing code that employs certain
features of the Embarcadero C++ compiler: __array_rank(T) and
__array_extent(T, Dim).
Douglas Gregor [Wed, 27 Apr 2011 23:34:22 +0000 (23:34 +0000)]
Clean up the handling of non-CVR qualifiers in template argument
deduction. The good news is that address spaces are a lot less broken
in this regard than I'd expected.
John Wiegley [Wed, 27 Apr 2011 23:09:49 +0000 (23:09 +0000)]
t/clang/type-traits
Patch authored by John Wiegley.
These type traits are used for parsing code that employs certain features of
the Embarcadero C++ compiler. Several of these constructs are also desired by
libc++, according to its project pages (such as __is_standard_layout).
Manuel Klimek [Wed, 27 Apr 2011 16:39:14 +0000 (16:39 +0000)]
This is the next step in building the standalone tools infrastructure:
This patch simplifies writing of standalone Clang tools. As an
example, we add clang-check, a tool that runs a syntax only frontend
action over a .cc file. When you integrate this into your favorite
editor, you get much faster feedback on your compilation errors, thus
reducing your feedback cycle especially when writing new code.
The tool depends on integration of an outstanding patch to CMake to
work which allows you to always have a current compile command
database in your cmake output directory when you set
CMAKE_EXPORT_COMPILE_COMMANDS.
More accurately model realloc() when the size argument is 0. realloc() with a size of 0 is equivalent to free(). The memory region should be marked as free and not used again.
Unit tests f2_realloc_0(), f6_realloc(), and f7_realloc() contributed by Marshall Clow <mclow.lists@gmail.com>. Thanks!
Add a warning (-Wnon-pod-memset) for calls to memset() with
a destination pointer that points to a non-POD type. This can flag such
horrible bugs as overwriting vptrs when a previously POD structure is
suddenly given a virtual method, or creating objects that crash on
practically any use by zero-ing out a member when its changed from
a const char* to a std::string, etc.
Douglas Gregor [Wed, 27 Apr 2011 06:18:01 +0000 (06:18 +0000)]
Introduce a new parser annotation token for primary expressions. When
ClassifyName() builds a primary expression, generate one of these
annotation tokens rather than jumping into the parser.
Douglas Gregor [Wed, 27 Apr 2011 05:41:15 +0000 (05:41 +0000)]
Simplify the parser's handling of Sema::ClassifyName() for types, by
creating a type-annotation token rather than jumping into the
declaration parsing.
Allow 'Environment::getSVal()' to allow an optional way for checkers to do a direct lookup to values bound to expressions, without
resulting to lazy logic. This is critical for the OSAtomicChecker that does a simulated load on any arbitrary expression.
Add support for Microsoft __interface keyword. An __interface class is basically a normal class containing just pure virtual functions. No urgency to enforce that restriction in clang for now, so make __interface an "class" alias.
Douglas Gregor [Wed, 27 Apr 2011 04:48:22 +0000 (04:48 +0000)]
Extend Sema::ClassifyName() to support C++, ironing out a few issues
in the classification of template names and using declarations. We now
properly typo-correct the leading identifiers in statements to types,
templates, values, etc. As an added bonus, this reduces the number of
lookups required for disambiguation.
Douglas Gregor [Wed, 27 Apr 2011 03:47:06 +0000 (03:47 +0000)]
Improve diagnostics for typo correction via Sema::ClassifyName(), by
looking at the context and the correction and using a custom
diagnostic. Also, enable some Fix-It tests that were somewhat lamely
disabled.
Douglas Gregor [Wed, 27 Apr 2011 00:01:52 +0000 (00:01 +0000)]
When comparing Objective-C pointers during overload resolution to
determine which is a better conversion to "void*", be sure to perform
the comparison using the safe-for-id
ASTContext::canAssignObjCInterfaces() rather than the asserts-with-id
ASTContext::canAssignObjCInterfaces().
Douglas Gregor [Tue, 26 Apr 2011 23:16:46 +0000 (23:16 +0000)]
When computing Objective-C pointer conversions in C++, retain
the qualifiers (e.g., GC qualifiers) on the type we're converting
from, rather than just blindly adopting the qualifiers of the type
we're converting to or dropping qualifiers altogether.
As an added bonus, properly diagnose GC qualifier mismatches to
eliminate a crash in the overload resolution failure diagnostics.
Manuel Klimek [Tue, 26 Apr 2011 21:50:03 +0000 (21:50 +0000)]
To be able to replay compilations we need to accurately remodel how
includes get resolved, especially when they are found relatively to
another include file. We also try to get it working for framework
includes, but that part of the code is untested, as I don't have a code
base that uses it.
Gcc pads the size of an array using the alignment of its elements.
The size of the array may not be aligned according to alignment of its elements if an alignment attribute is
specified in a typedef. Fixes rdar://8665729 & http://llvm.org/PR5637.
John McCall [Tue, 26 Apr 2011 20:42:42 +0000 (20:42 +0000)]
Make yet another placeholder type, this one marking that an expression is a bound
member function, i.e. something of the form 'x.f' where 'f' is a non-static
member function. Diagnose this in the general case. Some of the new diagnostics
are probably worse than the old ones, but we now get this right much more
universally, and there's certainly room for improvement in the diagnostics.
projects / clang / log