When we are deserializing the lexical decls of a DeclContext from PCH, notify the PCHReader to hold off passing Decls
to the consumer until the DeclContext is fully prepared.
Before, due to recursive loading, we could be in a situation where we would try to deserialize the decls of a DeclContext which was already doing that, and bad things would happen. In the specific case I encountered, the lexical decls would
form a cycle and we would enter infinite loop territory.
Refactor the way PCHReader tracks whether we are in recursive loading.
-Replace CurrentlyLoadingTypeOrDecl with a counting scheme (NumCurrentElementsDeserializing)
-Provide outside access to the mechanism by adding methods StartedDeserializing/FinishedDeserializing
to ExternalASTSource.
Chris Lattner [Fri, 30 Jul 2010 04:02:24 +0000 (04:02 +0000)]
fix PR5179 and correctly fix PR5831 to not miscompile.
The X86-64 ABI code didn't handle the case when a struct
would get classified and turn up as "NoClass INTEGER" for
example. This is perfectly possible when the first slot
is all padding (e.g. due to empty base classes). In this
situation, the first 8-byte doesn't take a register at all,
only the second 8-byte does.
This fixes this by enhancing the x86-64 abi stuff to allow
and handle this case, reverts the broken fix for PR5831,
and enhances the target independent stuff to be able to
handle an argument value in registers being accessed at an
offset from the memory value.
This is the last x86-64 calling convention related miscompile
that I'm aware of.
Sebastian Redl [Fri, 30 Jul 2010 00:29:29 +0000 (00:29 +0000)]
Make macro weirdness in chained PCH work. This required changing the way PCHReader and PCHWriter are initialized to correctly pick up all initializer. On the upside, this means that there is far less repetition in the dependent PCH now.
Daniel Dunbar [Thu, 29 Jul 2010 22:57:21 +0000 (22:57 +0000)]
Revert "Re-apply fixed version of 108749, correctly conditionalizing the new
sections on", this change uncovered a possible linker bug which resulted in the
wrong messages getting dispatched. Backing this out while we investigate...
Chris Lattner [Thu, 29 Jul 2010 18:39:32 +0000 (18:39 +0000)]
Implement the clang-side of detection for when to pass as
<2 x float> instead of double. This works but can't be turned
on until I teach codegen to pass <2 x float> as one XMM register
instead of two.
Merge PCHWriterDecl.cpp's isRequiredDecl and CodeGenModule::MayDeferGeneration into a new function,
DeclIsRequiredFunctionOrFileScopedVar.
This is essentially a CodeGen predicate that is also needed by the PCH mechanism to determine whether a decl
needs to be deserialized during PCH loading for codegen purposes.
Since this logic is shared by CodeGen and the PCH mechanism, move it to the ASTContext,
thus CodeGenModule's GetLinkageForFunction/GetLinkageForVariable and the GVALinkage enum is moved out of CodeGen.
This fixes current (and avoids future) codegen-from-PCH bugs.
Chris Lattner [Thu, 29 Jul 2010 18:13:09 +0000 (18:13 +0000)]
fix rdar://8251384, another case where we could access beyond the
end of a struct. This improves the case when the struct being passed
contains 3 floats, either due to a struct or array of 3 things. Before
we'd generate this IR for the testcase:
Chris Lattner [Thu, 29 Jul 2010 17:49:08 +0000 (17:49 +0000)]
start setting up infrastructure for passing multi-floats
as <2 x float> instead of as double. The backend isn't ready
yet, but infrastructure in the frontend can come up.
Chris Lattner [Thu, 29 Jul 2010 17:40:35 +0000 (17:40 +0000)]
rename Get8ByteTypeAtOffset -> GetINTEGERTypeAtOffset to
make it clear that this function should only return a type
that the codegen will classify the same as an INTEGER type.
Chris Lattner [Thu, 29 Jul 2010 17:04:54 +0000 (17:04 +0000)]
fix PR7742 / rdar://8250764, a miscompilation of struct
return where the struct has a base but no fields. This
was because the x86-64 abi logic was checking the wrong
predicate in one place.
This was introduced in r91874, which was a fix for PR5831,
which lacked a CHECK line, so I verified and added it.
This patch refactors much of the common code in ClassTemplateDecl and
FunctionTemplateDecl into a common base class RedeclarableTemplateDecl
together with support functions in a template class RedeclarableTemplate.
The patch also includes similar refactoring for these classes' PCH
reader and writer implementations.
Douglas Gregor [Thu, 29 Jul 2010 16:05:45 +0000 (16:05 +0000)]
When taking the address of a value of Objective-C object type (e.g.,
one because we're referencing a variable of type NSString &), the
resulting type is an ObjCObjectPointerType.
Douglas Gregor [Thu, 29 Jul 2010 15:18:02 +0000 (15:18 +0000)]
Allow a looser form of compatibility checking (which ignores
qualifiers) when checking a K&R function definition against a previous
prototype. Fixes <rdar://problem/8193107>.
Chris Lattner [Thu, 29 Jul 2010 06:26:06 +0000 (06:26 +0000)]
Kill off the 'coerce' ABI passing form. Now 'direct' and 'extend' always
have a "coerce to" type which often matches the default lowering of Clang
type to LLVM IR type, but the coerce case can be handled by making them
not be the same.
This simplifies things and fixes issues where X86-64 abi lowering would
return coerce after making preferred types exactly match up. This caused
us to compile:
define <4 x float> @foo(<4 x float> %X.coerce) nounwind {
entry:
%retval = alloca <4 x float>, align 16 ; <<4 x float>*> [#uses=2]
%coerce = alloca <4 x float>, align 16 ; <<4 x float>*> [#uses=2]
%X.addr = alloca <4 x float>, align 16 ; <<4 x float>*> [#uses=3]
store <4 x float> %X.coerce, <4 x float>* %coerce
%X = load <4 x float>* %coerce ; <<4 x float>> [#uses=1]
store <4 x float> %X, <4 x float>* %X.addr
%tmp = load <4 x float>* %X.addr ; <<4 x float>> [#uses=1]
%tmp1 = load <4 x float>* %X.addr ; <<4 x float>> [#uses=1]
%add = fadd <4 x float> %tmp, %tmp1 ; <<4 x float>> [#uses=1]
store <4 x float> %add, <4 x float>* %retval
%0 = load <4 x float>* %retval ; <<4 x float>> [#uses=1]
ret <4 x float> %0
}
Now we get:
define <4 x float> @foo(<4 x float> %X) nounwind {
entry:
%X.addr = alloca <4 x float>, align 16 ; <<4 x float>*> [#uses=3]
store <4 x float> %X, <4 x float>* %X.addr
%tmp = load <4 x float>* %X.addr ; <<4 x float>> [#uses=1]
%tmp1 = load <4 x float>* %X.addr ; <<4 x float>> [#uses=1]
%add = fadd <4 x float> %tmp, %tmp1 ; <<4 x float>> [#uses=1]
ret <4 x float> %add
}
Chris Lattner [Thu, 29 Jul 2010 04:41:05 +0000 (04:41 +0000)]
sink preferred type stuff lower. It's possible that this might
improve codegen for vaarg or something, because its codepath is
getting preferred types now.
Chris Lattner [Thu, 29 Jul 2010 02:31:05 +0000 (02:31 +0000)]
dissolve some more complexity: make the x86-64 abi lowering code
compute its own preferred types instead of having CGT compute
them then pass them (circuituously) down into ABIInfo.
Chris Lattner [Thu, 29 Jul 2010 02:16:43 +0000 (02:16 +0000)]
now that ABIInfo depends on CGT, it has trivial access to such
things as TargetData, ASTContext, LLVMContext etc. Stop passing
them through so many APIs.
Check for an invalid SourceLocation in clang_getCursor(). This avoids a possible assertion failure in SourceManager in the call to Lexer::GetBeginningOfToken(). Fixes <rdar://problem/8244873>.
Augment RegionStore::BindStruct() to bind symbolicated struct values. This fixes a false path issue reported in <rdar://problem/8243408> and also spurs another cause where the idempotent operations checker fires.
Douglas Gregor [Wed, 28 Jul 2010 23:59:57 +0000 (23:59 +0000)]
Don't set out-of-line template specialization/definition information
for AST nodes that aren't actually out-of-line (i.e., require a
nested-name-specifier). Fixes <rdar://problem/8204126>.
Chris Lattner [Wed, 28 Jul 2010 23:47:21 +0000 (23:47 +0000)]
pass argument vectors in a type that corresponds to the user type if
possible. This improves the example to pass <4 x float> instead of
<2 x double> but we still get awful code, and still don't get the
return value right.
Chris Lattner [Wed, 28 Jul 2010 22:15:08 +0000 (22:15 +0000)]
fix PR7714 by not referencing off the end of a struct when passed by value in
x86-64 abi. This also improves codegen as well. Some refactoring is needed of
this code.
Douglas Gregor [Wed, 28 Jul 2010 21:50:18 +0000 (21:50 +0000)]
When performing code completion for a case statement in a switch whose
condition is not of enumeration type, provide code-completion results
containing all values of integral or enumeral type.
Sebastian Redl [Wed, 28 Jul 2010 21:07:02 +0000 (21:07 +0000)]
Add a test case for tentative definitions in chained PCH. Fix a bug that completely messed up source locations and thus caused a crash whenever a diagnostic was emitted in chained PCH files.
Douglas Gregor [Wed, 28 Jul 2010 18:22:12 +0000 (18:22 +0000)]
The grammar for GNU typeof in C requires an expression to be
parenthesized, unlike in C++, e.g.,
C has: typeof ( expression)
C++ has: typeof unary-expression
So, once we've parsed a parenthesized expression after typeof, we
should only go on to parse the postfix expression suffix if we're in
C++. Fixes <rdar://problem/8237491>.
Add proper callbacks for DeclStmt -- we weren't recursing on
the decls. This was just an oversight before; one we didn't
catch because lots of information in a DeclStmt was also being
traversed (redundantly) elsewhere.
Once DeclStmt was cleaned up, I could clean up some of the
redundant traversals found elswhere as well -- in particular,
traversing the declarations inside a function as part of the
function callback (instead of as part of the CompoundExpr
callback that constitutes the body of the function). The old
way was really weird, and led to some parts of local variable
declarations (but not all) being visited twice. That is now
resolved. I also was able to simplify the traversers for
IfStmt/WhileStmt/etc, which used to have redundant calls to
work around the fact DeclStmt wasn't working properly.
While in the area, I fixed up a few more recursion-ordering
issues. I try to hold to the principle that
RecursiveASTVisitor visits objects in the source code in the
same order they're typed. So the return-type of a variable
comes before the variable-name. This still isn't perfect, but
we're closer to that.
Daniel Dunbar [Wed, 28 Jul 2010 15:40:33 +0000 (15:40 +0000)]
Preprocessor: Add support for '#pragma clang __debug crash' and '#pragma clang
__debug overflow_stack'.
- For testing crash reporting stuff... you'd think I could just use some C++
code but Doug keeps fixing stuff!
Douglas Gregor [Wed, 28 Jul 2010 15:14:14 +0000 (15:14 +0000)]
Enable expression transformations in the current-instantiation
rebuilder, i.e., remove a silly short-sighted hack from long
ago. Thanks to Abramo Bagnara for the test case/bug report!
Douglas Gregor [Wed, 28 Jul 2010 14:49:07 +0000 (14:49 +0000)]
When a nested-name-specifier refers into a current instantiation that has
dependent bases, construct a dependent nested-name-specifier rather
than complaining that the name could not be found within the current
instantiation itself. Fixes PR7725.
John McCall [Wed, 28 Jul 2010 01:07:35 +0000 (01:07 +0000)]
When creating a jump destination, its scope should be the scope of the
enclosing normal cleanup, not the top of the EH stack. I'm *really*
surprised this hasn't been causing more problems.
Tom Care [Tue, 27 Jul 2010 23:30:21 +0000 (23:30 +0000)]
Added some false positive checking to UnreachableCodeChecker
- Allowed reporting of dead macros
- Added path walking function to search for false positives in conditional statements
- Updated some affected tests
- Added some false positive test cases