Douglas Gregor [Thu, 22 Dec 2011 01:48:48 +0000 (01:48 +0000)]
When deserializing an Objective-C class, check whether we have another
declaration of that same class that either came from some other module
or occurred in the translation unit loading the module. In this case,
we need to merge the two redeclaration chains immediately so that all
such declarations have the same canonical declaration in the resulting
AST (even though they don't in the module files we've imported).
Focusing on Objective-C classes until I'm happy with the design, then
I'll both (1) extend this notion to other kinds of declarations, and
(2) optimize away this extra checking when we're not dealing with
modules. For now, doing this checking for PCH files/preambles gives us
better testing coverage.
Ted Kremenek [Thu, 22 Dec 2011 01:30:46 +0000 (01:30 +0000)]
Enhance AnalysisDeclContext::getReferencedBlockVars() to understand PseudoObjExprs. It turns out
that the information collected by this method is a super set of the captured variables in BlockDecl.
In Lexer::getCharAndSizeSlow[NoWarn] make sure we don't go over the end of the buffer
when the end of the buffer is immediately after an escaped newline.
For SourceManager::isBeforeInTranslationUnit(), have it consider macro arg expanded
token locations as coming before the closing ')' of a function macro expansion.
Douglas Gregor [Wed, 21 Dec 2011 15:12:03 +0000 (15:12 +0000)]
Implement a trivial, obvious optimization for deserialization of
redeclaration chains: only ever have the reader search for
redeclarations of the first (canonical) declaration, since we only
ever record redeclaration ranges for the that declaration. Searching
for redeclarations of non-canonical declarations will never find
anything, so it's a complete waste of time.
Richard Smith [Wed, 21 Dec 2011 02:55:12 +0000 (02:55 +0000)]
C++11 half of r147023: In C++11, additionally eagerly instantiate:
- constexpr function template instantiations
- variables of reference type
- constexpr variables
Eli Friedman [Wed, 21 Dec 2011 00:43:02 +0000 (00:43 +0000)]
Fix a case where Expr::isConstantInitializer would return true for an expression we can't support. In a slightly amusing twist, the case in question was already in the clang regression tests marked as a valid construct. <rdar://problem/10020074>
Richard Smith [Wed, 21 Dec 2011 00:25:33 +0000 (00:25 +0000)]
C++ constant expression handling: eagerly instantiate static const integral data
members of class templates so that their values can be used in ICEs. This
required reverting r105465, to get such instantiated members to be included in
serialized ASTs.
Anna Zaks [Tue, 20 Dec 2011 22:35:30 +0000 (22:35 +0000)]
[analyzer] Do not invalidate arguments when the parameter's
type is a pointer to const. (radar://10595327)
The regions corresponding to the pointer and reference arguments to
a function get invalidated by the calls since a function call can
possibly modify the pointed to data. With this change, we are not going
to invalidate the data if the argument is a pointer to const. This
change makes the analyzer more optimistic in reporting errors.
(Support for C, C++ and Obj C)
Douglas Gregor [Tue, 20 Dec 2011 22:06:13 +0000 (22:06 +0000)]
When we make a previously-deserialized module definition visible,
notify the AST deserialization listener so that the AST writer knows
that it can write the macro definition.
Douglas Gregor [Tue, 20 Dec 2011 18:11:52 +0000 (18:11 +0000)]
When performing name lookup for a redeclaration, ignore module
visibility restrictions. This ensures that all declarations of the
same entity end up in the same redeclaration chain, even if some of
those declarations aren't visible. While this may seem unfortunate to
some---why can't two C modules have different functions named
'f'?---it's an acknowedgment that a module does not introduce a new
"namespace" of names.
As part of this, stop merging the 'module-private' bit from previous
declarations to later declarations, because we want each declaration
in a module to stand on its own because this can effect, for example,
submodule visibility.
Note that this notion of names that are invisible to normal name
lookup but are available for redeclaration lookups is how we should
implement friend declarations and extern declarations within local
function scopes. I'm not tackling that problem now.
Craig Topper [Tue, 20 Dec 2011 09:55:26 +0000 (09:55 +0000)]
Add AVX2 intrinsics for pavg, pblend, and pcmp instructions. Also remove unneeded builtins for SSE pcmp. Change SSE pcmpeqq and pcmpgtq to not use builtins and just use vector == and >.
Bob Wilson [Tue, 20 Dec 2011 06:16:48 +0000 (06:16 +0000)]
Relax type checking for a few Neon intrinsics. <rdar://problem/10538555>
Not long ago, I tightened up the type checking for pointer arguments of
Neon intrinsics to match the specifications provided by ARM. One consequence
was that it became impossible to access the unaligned versions of a few
Neon load/store operations. Since there are just a few of these intrinsics
where it makes a difference, I think it's better to relax the type checking
than to either introduce new non-standard unaligned intrinsics or to disallow
intrinsics for the unaligned operations.
Richard Smith [Tue, 20 Dec 2011 02:08:33 +0000 (02:08 +0000)]
Unlike in C++03, a constant-expression is not an unevaluated operand in C++11.
Split out a new ExpressionEvaluationContext flag for this case, and don't treat
it as unevaluated in C++11. This fixes some crash-on-invalids where we would
allow references to class members in potentially-evaluated constant expressions
in static member functions, and also fixes half of PR10177.
The fix to PR10177 exposed a case where template instantiation failed to provide
a source location for a diagnostic, so TreeTransform has been tweaked to supply
source locations when transforming a type. The source location is still not very
good, but MarkDeclarationsReferencedInType would need to operate on a TypeLoc to
improve it further.
Also fix MarkDeclarationReferenced in C++98 mode to trigger instantiation for
static data members of class templates which are used in constant expressions.
This fixes a link-time problem, but we still incorrectly treat the member as
non-constant. The rest of the fix for that issue is blocked on PCH support for
early-instantiated static data members, which will be added in a subsequent
patch.
Douglas Gregor [Tue, 20 Dec 2011 00:28:52 +0000 (00:28 +0000)]
Detect when mapping a #include/#import over to a submodule ends up
hitting a submodule that was never actually created, e.g., because
that header wasn't parsed. In such cases, complain (because the
module's umbrella headers don't cover everything) and fall back to
including the header.
Later, we'll add a warning at module-build time to catch all such
cases. However, this fallback is important to eliminate assertions in
the ASTWriter when this happens.
Richard Smith [Mon, 19 Dec 2011 22:12:41 +0000 (22:12 +0000)]
Evaluation support for ExprWithCleanups. We won't evaluate any expression which
actually requires non-trivial cleanups, so no cleanups need to be performed.
Chris Lattner [Mon, 19 Dec 2011 21:16:08 +0000 (21:16 +0000)]
Fix PR5279 - don't sliently drop alignment information on stores of types that have alignment less than the natural alignment of the type when it comes from a typedef.
Douglas Gregor [Mon, 19 Dec 2011 21:09:25 +0000 (21:09 +0000)]
Eliminate the first->last redeclaration map from the AST file
format. It's no longer being used, now that we have a new
implementation of redeclaration chains.
Douglas Gregor [Mon, 19 Dec 2011 19:00:47 +0000 (19:00 +0000)]
Once we have fully deserialized a redeclaration chain for something
with a definition pointer (e.g., C++ and Objective-C classes), zip
through the redeclaration chain to make sure that all of the
declarations point to the definition data.
As part of this, realized again why the first redeclaration of an
entity in a file is important, and brought back that idea.
Douglas Gregor [Mon, 19 Dec 2011 18:19:24 +0000 (18:19 +0000)]
Re-implement (de-)serialization of redeclaration chains for
redeclaration templates (RedeclarableTemplateDecl), similarly to the
way (de-)serialization is implemented for Redeclarable<T>. In the
process, found a simpler formulation for handling redeclaration
chains and implemented that in both places.
The new test establishes that we're building the redeclaration chains
properly. However, the FIXME indicates where we're tickling a
different bug that has to do with us not setting the DefinitionData
pointer properly in redeclarations that we detected after the
definition itself was deserialized. The (separable) fix for that bug
is forthcoming.
Douglas Gregor [Mon, 19 Dec 2011 16:14:14 +0000 (16:14 +0000)]
The submodule offset map can introduce "empty" remapping entries for
imported modules that don't introduce any new entities of a particular
kind. Allow these entries to be replaced with entries for another
loaded module.
In the included test case, selectors exhibit this behavior.
Douglas Gregor [Mon, 19 Dec 2011 15:27:36 +0000 (15:27 +0000)]
Optimize serialized representation of redeclarable declarations for
which there are no redeclarations. This reduced by size of the PCH
file for Cocoa.h by ~650k: ~536k of that was in the new
LOCAL_REDECLARATIONS table, which went from a ridiculous 540k down to
an acceptable 3.5k, while the rest was due to the more compact
abbreviated representation of redeclarable declaration kinds (which no
longer need to store the 'first' declaration ID).
Rafael Espindola [Mon, 19 Dec 2011 14:41:01 +0000 (14:41 +0000)]
Extend the fix for PR9614 to handle inline asm in the outer decl and
the inner decl being a builtin. This is needed to support the glibc headers
in fedora 16 (2.14).
Richard Smith [Mon, 19 Dec 2011 06:19:21 +0000 (06:19 +0000)]
constexpr handling improvements. Produce detailed diagnostics when a 'constexpr'
variable is initialized by a non-constant expression, and pass in the variable
being declared so that earlier-initialized fields' values can be used.
Rearrange VarDecl init evaluation to make this possible, and in so doing fix a
long-standing issue in our C++ constant expression handling, where we would
mishandle cases like:
extern const int a;
const int n = a;
const int a = 5;
int arr[n];
Here, n is not initialized by a constant expression, so can't be used in an ICE,
even though the initialization expression would be an ICE if it appeared later
in the TU. This requires computing whether the initializer is an ICE eagerly,
and saving that information in PCH files.
Richard Smith [Sun, 18 Dec 2011 02:33:09 +0000 (02:33 +0000)]
PR11604: don't allow floating-literal-to-integer casts in ICEs if the (truncated)
floating literal value does not fit into the destination type. Such casts have
undefined behavior at translation time; treating them as non-ICE matches the
behavior of modern gcc versions.
Douglas Gregor [Sat, 17 Dec 2011 23:38:30 +0000 (23:38 +0000)]
Completely re-implement (de-)serialization of declaration
chains. The previous implementation relied heavily on the declaration
chain being stored as a (circular) linked list on disk, as it is in
memory. However, when deserializing from multiple modules, the
different chains could get mixed up, leading to broken declaration chains.
The new solution keeps track of the first and last declarations in the
chain for each module file. When we load a declaration, we search all
of the module files for redeclarations of that declaration, then
splice together all of the lists into a coherent whole (along with any
redeclarations that were actually parsed).
As a drive-by fix, (de-)serialize the redeclaration chains of
TypedefNameDecls, which had somehow gotten missed previously. Add a
test of this serialization.
This new scheme creates a redeclaration table that is fairly large in
the PCH file (on the order of 400k for Cocoa.h's 12MB PCH file). The
table is mmap'd in and searched via a binary search, but it's still
quite large. A future tweak will eliminate entries for declarations
that have no redeclarations anywhere, and should
drastically reduce the size of this table.
Chandler Carruth [Sat, 17 Dec 2011 23:10:01 +0000 (23:10 +0000)]
Split the Windows toolchain definition into its own file. This is
especially nice as the Windows toolchain needs the windows header files,
and has lots of platform specific hooks in it.
To facilitate the split, hoist a bunch of file-level static helpers into
class-level static helpers. Spiff up their doxygen comments while there
as they're now more likely to be looked up via docs.
Hopefully, this will be followed by further breaking apart of the
toolchain definitions. Most of the large and complex ones should likely
live on their own. I'm looking at you Darwin. ;]
Ted Kremenek [Sat, 17 Dec 2011 05:26:04 +0000 (05:26 +0000)]
Refactor 'TextDiagnostic' to have a parent class 'DiagnosticRenderer' which handles
the policy of how diagnostics are lowered/rendered, while TextDiagnostic handles
the actual pretty-printing.
This is a first part of reworking SerializedDiagnosticPrinter to use the same
inclusion-stack/macro-expansion logic as TextDiagnostic.
[PCH] In ASTReader::FinishedDeserializing fully load the interesting decls,
including deserializing their bodies, so that any other declarations that
get referenced in the body will be fully deserialized by the time we pass them to the consumer.
Could not reduce to a test case unfortunately. rdar://10587158.
In Parser::SkipUntil do not stop at '@' unconditionally.
Stopping at '@' was originally intended to avoid skipping an '@' at the @interface context
when doing parser recovery, but we should not stop at all '@' tokens because they may be part
of expressions (e.g. in @"string", @selector(), etc.), so in most cases we will want to skip them.
This commit caused 'test/Parser/method-def-in-class.m' to fail for the cases where we tried to
recover from unmatched angle bracket but IMO it is not a big deal to not have good recovery
from such broken code and the way we did recovery would not always work anyway (e.g. if there was '@'
in an expression).
The case that rdar://7029784 is about still passes.
After late parsing an objc method, make sure there are no leftover cached tokens,
because the memory associated with them is going to get released.
We also don't want them to affect later parsing.
(We do the same for C++ inline methods.)
The underlying cause for the leftover tokens is going to be addressed in the
next commit.
Couldn't get a test case for the crash though. rdar://10583033.
projects / clang / log