Method implemented in class's implementation may implement
one declared in class's extension and not one declared
in class's superclass. This supresses a bogus warning on
method type mismatch.
Fixes //rdar: // 8530080
When dealing with an assignment with LHS being a property reference
expression, the entire assignment tree is rewritten into a property
setter messaging. This includes rewriting the RHS.
Do not attempt to rewrite RHS again. Never rewrite a rewritten text!
Fixes //rdar: //8527018.
Douglas Gregor [Fri, 8 Oct 2010 20:56:19 +0000 (20:56 +0000)]
Revert the hack Chris Lattner added in r97981 to work around
brokenness in the designated-initializer ASTs. The ASTs were fixed by
Alp Toker's patch (r116098) for PR6955. Fixes PR6537.
Douglas Gregor [Fri, 8 Oct 2010 20:44:28 +0000 (20:44 +0000)]
This patch fixes multiple issues in clang's designated init builder and
completes support for C1X anonymous struct/union init features:
* Indexed anonymous member initializers should not be expanded. Doing so makes
little sense and would cause unresolvable semantic ambiguity in valid code
(regression introduced by r69153).
* Subobject initialization of (possibly nested) anonymous members are now
referred to with paths relative to the naming record context, eliminating the
synthesis of incorrect implicit InitListExprs that caused CodeGen to assert.
* Field lookup was missing a null check in IdentifierInfo comparison which
caused lookup for a known (already resolved) field to match the first unnamed
data member it encountered leading to silent miscompilation.
* Subobject paths are no longer built using the general purpose
Sema::BuildAnonymousStructUnionMemberPath(). If any corner cases crop up, we
will now assert earlier in Sema instead of passing invalid InitListExprs
through to CodeGen.
Daniel Dunbar [Fri, 8 Oct 2010 19:30:33 +0000 (19:30 +0000)]
libclang: Disable LLVM pretty stack trace functionality, which inadvertently
sets up signal handlers it shouldn't when we are being used a shared library.
John McCall [Fri, 8 Oct 2010 18:24:19 +0000 (18:24 +0000)]
__attribute__((aligned(n))) directly specifies the alignment of a declaration
unless it's a non-packed field, in which case it can only increase the
alignment. [[align]] effectively works the same way for well-formed code
(because it's ill-formed for [[align]] to decrease alignment ever).
Douglas Gregor [Fri, 8 Oct 2010 04:03:57 +0000 (04:03 +0000)]
Fix three related, wily issues with the recompilation of precompiled
preambles:
- When we rebuild a precompiled preamble, make sure to disable
skipping anything in the main file; we may have had leftover
preamble-skipping values in the lexer, which leads to very empty
preamble. This is a correctness issue.
- When we rebuild a precompiled preamble, clear out any prior state
in the Diagnostic object. Otherwise, we might think that there
were errors when we were building the preamble itself, and
therefore reject the resulting preamble. This is mainly a
performance issue.
- Don't remove old remappings when digging out the remapping for the
main file. Having the old mappings around does not hurt in the
common case (later remappings will just overwrite them), and is
important when we fail to find a preamble: we don't want to have
removed the remapping, because we'll need it later.
John McCall [Fri, 8 Oct 2010 02:01:28 +0000 (02:01 +0000)]
Track the location of the context requiring an implicit conversion and use it
to white-list conversions required by system headers. rdar://problem/8232669
Fix an infinite loop, caused by unintended syntax bug (the 'break;' after 'default:' was intended to break out of the while loop).
Fixes rdar://8518859&8520617.
Start and end location of a property-dot syntax expression
must match start and end location of the expression
as expected by the rewriter client. Fixes // rdar: // 8520727
Daniel Dunbar [Thu, 7 Oct 2010 15:00:30 +0000 (15:00 +0000)]
Driver: When clang is built with a VENDOR set, include the base LLVM version in
the version information, to help prevent user confusion about vendor version
numbers vs. LLVM version numbers.
Tom Care [Wed, 6 Oct 2010 23:02:25 +0000 (23:02 +0000)]
UnreachableCodeChecker cleanup and improvements
- Fixed some iterator style issues
- Don't process blocks that have been visited already
- Fixed a case where a unreachable block cycle was not reported
- Minor test case changes
- Added one test case from flow-sensitive version of the check. More coming.
Douglas Gregor [Wed, 6 Oct 2010 16:00:31 +0000 (16:00 +0000)]
Reject the allocation of variably-modified types in C++ 'new'
expressions. Fixes PR8209 in the narrowest way possible. I'm still
considering whether I want to implement the extension that permits the
use of VLA types in a 'new' expression.
Remove hard CRLF end-of-line markers from two files. These should be added
automatically when syncing due to the auto-props selecting svn:eol-style of
'native'.
John McCall [Wed, 6 Oct 2010 01:40:51 +0000 (01:40 +0000)]
Teach the driver to pass -fexceptions in Objective-C inputs on ARM.
I don't know when this stopped happening, but this seems to be the
right place to do it.
John McCall [Tue, 5 Oct 2010 22:36:42 +0000 (22:36 +0000)]
When instantiating a new-expression, force a rebuild if there were default
arguments in either the placement or constructor arguments. This is
important if the default arguments refer to a declaration or create a
temporary.
Method declaration and its implementation must match in all their types.
Previously, compiler warned only if it was unsafe if types
did not match. Fixes // rdar: //7933061
John McCall [Tue, 5 Oct 2010 20:48:15 +0000 (20:48 +0000)]
Teach PopCleanupBlock to correctly handle the possibility of branching through
a EH-only cleanup as part of a fallthrough branch-through. That this happens
for this test case is actually a separate bug.
Douglas Gregor [Tue, 5 Oct 2010 18:37:06 +0000 (18:37 +0000)]
Fix a marvelous chained AST writing bug, where we end up with the
following amusing sequence:
- AST writing schedules writing a type X* that it had never seen
before
- AST writing starts writing another declaration, ends up
deserializing X* from a prior AST file. Now we have two type IDs for
the same type!
- AST writer tries to write X*. It only has the lower-numbered ID
from the the prior AST file, so references to the higher-numbered ID
that was scheduled for writing go off into lalaland.
To fix this, keep the higher-numbered ID so we end up writing the type
twice. Since this issue occurs so rarely, and type records are
generally rather small, I deemed this better than the alternative: to
keep a separate mapping from the higher-numbered IDs to the
lower-numbered IDs, which we would end up having to check whenever we
want to deserialize any type.
Fix a block rewriter bug where copy/dispose entries in
block descriptor for outer block was missing even though
the block was importing objects into its inner blocks.
//rdar://84995992
Douglas Gregor [Tue, 5 Oct 2010 15:41:24 +0000 (15:41 +0000)]
Register the __builtin_va_list_type node when we parse it, rather than
waiting until we think we need it: we didn't catch all of the places
where we actually needed it, and we probably wouldn't ever. Fixes a
C++ PCH crasher.
Douglas Gregor [Tue, 5 Oct 2010 14:55:45 +0000 (14:55 +0000)]
Register the __builtin_va_list_type node when we parse it, rather than
waiting until we think we need it: we didn't catch all of the places
where we actually needed it, and we probably wouldn't ever. Fixes a
C++ PCH crasher.
In Sema's TryRefInitWithConversionFunction, suppress user conversions for the overload candidates.
Fixes an infinite recursion in overload resolution for rdar://8499524.
Many thanks to Doug!
John McCall [Tue, 5 Oct 2010 02:33:56 +0000 (02:33 +0000)]
If we're resolving all outstanding fixups, and there are multiple fixups
for the same destination, then we must potentially rewrite the initial branch
of every fixup. Without this patch, a short-circuit check meant to prevent
a switch case from being redundantly added was preventing later fixups from
being processed. Fixes PR8175 (again).
John McCall [Mon, 4 Oct 2010 23:42:51 +0000 (23:42 +0000)]
In the fragile ObjC ABI, save the caught exception to the side if there are
both @catches and a @finally, because the second call to @objc_exception_try_enter
will clobber the exception slot. Fixes rdar://problem/8440970.
Douglas Gregor [Sat, 2 Oct 2010 23:49:58 +0000 (23:49 +0000)]
When providing a block literal as a code completion for a
function/method argument, include the parameter name and always
include parentheses (even for zero-parameter blocks). Otherwise, the
block literal placeholder '^' can look very weird.
Douglas Gregor [Sat, 2 Oct 2010 22:49:11 +0000 (22:49 +0000)]
Introduce a new libclang function, clang_getCursorDisplayName(), which
produces a simple "display" name that captures the
arguments/parameters for a function, function template, class
template, or class template specialization.
Chris Lattner [Sat, 2 Oct 2010 21:32:59 +0000 (21:32 +0000)]
the mmx intrinsic for pshufw should map to the IR intrinsic, not
to a shufflevector. Otherwise it doesn't turn into a pshufw.
This bug was introduced in the mmx rewrite.
Douglas Gregor [Sat, 2 Oct 2010 21:06:43 +0000 (21:06 +0000)]
When we insert a category (or class extension) into an interface, mark
the interface as having changed since it was originally
serialized. This ensures that we see class extensions/categories in
chained PCH files.
Douglas Gregor [Sat, 2 Oct 2010 19:51:13 +0000 (19:51 +0000)]
Provide proper source location and range information for C++ base
specifier cursors in libclang. FIXME -=2, fixes the rest of
<rdar://problem/8274883>.
Chris Lattner [Sat, 2 Oct 2010 00:09:12 +0000 (00:09 +0000)]
when expanding a builtin, if the argument is required to be a constant,
force it to be a constant instead of emitting with EmitScalarExpr. In
-ftrapv mode, they are not the same.
projects / clang / log