Douglas Gregor [Thu, 5 Jan 2012 01:11:47 +0000 (01:11 +0000)]
When we're performing name lookup for a tag, we still allow ourselves
to see hidden declarations because every tag lookup is effectively a
redeclaration lookup. For example, image that
struct foo;
is declared in a submodule that is known but hasn't been imported. If
someone later writes
struct foo *foo_p;
then "struct foo" is either a reference or a redeclaration. To keep
the redeclaration chains sound, we treat it like a redeclaration for
name-lookup purposes.
Fixes a code gen bug for setter code for a property of
c++ object reference type with trivial copy constructor.
This causes an assert crash and bad code gen. when assert
is off. // rdar://6137845
Douglas Gregor [Thu, 5 Jan 2012 00:04:05 +0000 (00:04 +0000)]
When generating includes for all of the headers we found in an
umbrella directory, skip includes for any headers that are part of an
unavailable module.
Anna Zaks [Wed, 4 Jan 2012 23:54:01 +0000 (23:54 +0000)]
[analyzer] Be less pessimistic about invalidation of global variables
as a result of a call.
Problem:
Global variables, which come in from system libraries should not be
invalidated by all calls. Also, non-system globals should not be
invalidated by system calls.
Solution:
The following solution to invalidation of globals seems flexible enough
for taint (does not invalidate stdin) and should not lead to too
many false positives. We split globals into 3 classes:
* immutable - values are preserved by calls (unless the specific
global is passed in as a parameter):
A : Most system globals and const scalars
* invalidated by functions defined in system headers:
B: errno
* invalidated by all other functions (note, these functions may in
turn contain system calls):
B: errno
C: all other globals (which are not in A nor B)
Douglas Gregor [Wed, 4 Jan 2012 23:32:19 +0000 (23:32 +0000)]
Store the submodules of a module in source order, as they are stored
in the module map. This provides a bit more predictability for the
user, as well as eliminating the need to sort the submodules when
serializing them.
objc: When issuing warning for missing synthesis for
properties in classes declared with objc_suppress_autosynthesis
attribute, pinpoint location of the said class in a note.
Eli Friedman [Wed, 4 Jan 2012 23:13:47 +0000 (23:13 +0000)]
Add an APValue representation for the difference between two address-of-label expressions. Add support to Evaluate and CGExprConstant for generating/handling them. Remove the special-case for such differences in Expr::isConstantInitializer.
With that done, remove a bunch of buggy code from CGExprConstant for handling scalar expressions which is no longer necessary.
Douglas Gregor [Wed, 4 Jan 2012 21:16:09 +0000 (21:16 +0000)]
Add __has_feature(modules) to indicate when modules are available (in
any language variant), and restrict __has_feature(objc_modules) to
mean that we also have the Objective-C @import syntax. I anticipate
__has_feature(cxx_modules) and/or __has_feature(c_modules) for when we
nail down the module syntax for C/C++.
Douglas Gregor [Wed, 4 Jan 2012 17:13:46 +0000 (17:13 +0000)]
Implement declaration merging for non-template functions from
different modules. This implementation is a first approximation of
what we want, using only the function type to determine
equivalence. Later, we'll want to deal with some of the more subtle
issues, including:
- C allows a prototyped declaration and a non-prototyped declaration
to be merged, which we should support
- We may want to ignore the return type when merging, then
complain if the return types differ. Or, we may want to leave it
as it us, so that we only complain if overload resolution
eventually fails.
- C++ non-static member functions need to consider cv-qualifiers
and ref-qualifiers.
- Function templates need to consider the template parameters and
return type.
- Function template specializations will have special rules.
- We can now (accidentally!) end up overloading in C, even without
the "overloadable" attribute, and will need to detect this at some
point.
The actual detection of "is this an overload?" is implemented by
Sema::IsOverload(), which will need to be moved into the AST library
for re-use here. That will be a future refactor.
Douglas Gregor [Wed, 4 Jan 2012 16:44:10 +0000 (16:44 +0000)]
Implement declaration merging for typedefs loaded from disjoint
modules, so long as the typedefs refer to the same underlying
type. This ensures that the typedefs end up in the same redeclaration
chain.
To test this, fix name lookup for C/Objective-C to properly deal with
multiple declarations with the same name in the same scope.
Eli Friedman [Wed, 4 Jan 2012 04:41:38 +0000 (04:41 +0000)]
Add an explicit LambdaExprContext to Declarator, to parallel BlockLiteralContext. Use it to ensure semantic analysis of types isn't confused by the lack of a type specifier.
Eli Friedman [Wed, 4 Jan 2012 02:40:39 +0000 (02:40 +0000)]
Stub out the Sema interface for lambda expressions, and change the parser to use it. Unconditionally error on lambda expressions because they don't work in any meaningful way yet.
In non-gc, non-arc mode, property of 'Class' type
variety is treated as a 'void *'. No need to issue
warning reserved for objc object properties.
// rdar://10565506
Douglas Gregor [Tue, 3 Jan 2012 23:34:23 +0000 (23:34 +0000)]
Minor tweak to name lookup for C/Objective-C: after the first name, still consider whether this is a redeclaration lookup when determining whether to look for the visible declaration
Douglas Gregor [Tue, 3 Jan 2012 23:26:26 +0000 (23:26 +0000)]
Test "merging" of typedef types across distinct modules. At present,
the AST reader doesn't actually perform a merge, because name lookup
knows how to merge identical typedefs together.
As part of this, teach C/Objective-C name lookup to return multiple
results in all cases, rather than first digging through the attributes
to see if the value is overloadable. This way, we'll catch ambiguous
lookups in C/Objective-C.
Ted Kremenek [Tue, 3 Jan 2012 23:18:57 +0000 (23:18 +0000)]
Add initial version of checker to check if virtual member functions are called transitively
from C++ constructors or destructors. Checker by Lei Zhang with a few tweaks by Ted Kremenek.
David Chisnall [Tue, 3 Jan 2012 23:18:17 +0000 (23:18 +0000)]
Initialise constant Objective-C string isa pointers with a weak reference to the class, so that they are usable in +load methods if the string class has been compiled with clang - if it's been compiled with GCC, the ABI makes this impossible. (GNU runtimes)
Douglas Gregor [Tue, 3 Jan 2012 19:32:59 +0000 (19:32 +0000)]
Eliminate the uglified keyword __import_module__ for importing
modules. This leaves us without an explicit syntax for importing
modules in C/C++, because such a syntax needs to be discussed
first. In Objective-C/Objective-C++, the @import syntax is used to
import modules.
Note that, under -fmodules, C/C++ programs can import modules via the
#include mechanism when a module map is in place for that header. This
allows us to work with modules in C/C++ without committing to a syntax.
Douglas Gregor [Tue, 3 Jan 2012 17:31:38 +0000 (17:31 +0000)]
Don't attempt to merge a deserialized declaration with existing
declarations in the AST unless modules are enabled. This case doesn't
come up with precompiled headers, and it isn't cheap.
Douglas Gregor [Tue, 3 Jan 2012 17:27:13 +0000 (17:27 +0000)]
Factor the merging of declarations in the AST reader out to a separate
member function template, since the behavior is identical for
ObjCInterfaceDecl and ObjCProtocolDecl. It's expected that all
redeclarable entities will have the same behavior.
Douglas Gregor [Tue, 3 Jan 2012 15:21:29 +0000 (15:21 +0000)]
Rename the command-line option for mapping #include/#import over to
module imports from -fauto-module-import to -fmodules. The new name
will eventually be used to enable modules, and the #include/#import
mapping is a crucial part of the feature.
Richard Smith [Mon, 2 Jan 2012 18:14:06 +0000 (18:14 +0000)]
Add assertion to char32_t that the value is valid, as suggested by Jordy Rose.
Add a test that such characters don't make it through to StringLiteral objects
in error recovery.
Douglas Gregor [Mon, 2 Jan 2012 17:18:37 +0000 (17:18 +0000)]
Diagnose cases where the definition of a particular type is required,
is known (to Clang), but is not visible because the module has not yet
been imported.
Fix PR11685 by implementing -ffast-math and its various friends in the
Clang driver. This involves a bunch of silly option parsing code to try
to carefully emulate GCC's options. Currently, this takes a conservative
approach, and unless all of the unsafe optimizations are enabled, none
of them are. The fine grained control doesn't seem particularly useful.
If it ever becomes useful, we can add that to LLVM first, and then
expose it here.
This also fixes a few tiny bugs in the flag management around
-fhonor-infinities and -fhonor-nans; the flags now form proper sets both
for enabling and disabling, with the last flag winning.
I've also implemented a moderately terrifying GCC feature where
a language change is also provided by the '-ffast-math' flag by defining
the __FAST_MATH__ preprocessor macro. This feature is tracked and
serialized in the frontend but it isn't used yet. A subsequent patch
will add the preprocessor macro and tests for it.
I've manually tested that codegen appears to respect this, but I've not
dug in enough to see if there is an easy way to test codegen options w/o
relying on the particulars of LLVM's optimizations.
Douglas Gregor [Sun, 1 Jan 2012 21:47:52 +0000 (21:47 +0000)]
Implement declaration merging for Objective-C protocols across
multiple, disjoint modules. There is far too much duplicating with the
ObjCInterfaceDecl case here, which I'll eliminate shortly.
Douglas Gregor [Sun, 1 Jan 2012 20:30:41 +0000 (20:30 +0000)]
Wire up redeclaration chains for Objective-C protocols, so that both
forward declarations and definitions of an Objective-C protocol are
represented within a single chain of ObjCProtocolDecls.
Douglas Gregor [Sun, 1 Jan 2012 19:51:50 +0000 (19:51 +0000)]
Introduce the core infrastructure needed to model redeclaration chains
for Objective-C protocols, including:
- Using the first declaration as the canonical declaration
- Using the definition as the primary DeclContext
- Making sure that all declarations have a pointer to the definition
data, and that we know which declaration is the definition
- Serialization support for redeclaration chains and for adding
definitions to already-serialized declarations.
However, note that we're not taking advantage of much of this code
yet, because we're still re-using ObjCProtocolDecls.
Douglas Gregor [Sun, 1 Jan 2012 19:29:29 +0000 (19:29 +0000)]
Move the data that corresponds to the definition of a protocol into a
separately-allocated DefinitionData structure. Introduce various
functions that will help with the separation of declarations from
definitions (isThisDeclarationADefinition(), hasDefinition(),
getDefinition()).
Douglas Gregor [Sun, 1 Jan 2012 18:09:12 +0000 (18:09 +0000)]
Use declaresSameEntity() when comparing ObjCProtocolDecls, and
getCanonicalDecl() when putting ObjCProtocolDecls into a set. This is
groundwork for making ObjCProtocolDecl redeclarable.
Douglas Gregor [Sat, 31 Dec 2011 04:05:44 +0000 (04:05 +0000)]
Implement support for module requirements, which indicate the language
features needed for a particular module to be available. This allows
mixed-language modules, where certain headers only work under some
language variants (e.g., in C++, std.tuple might only be available in
C++11 mode).
Richard Smith [Fri, 30 Dec 2011 21:15:51 +0000 (21:15 +0000)]
Unrevert r147271, reverted in r147361.
Also temporarily remove the assumption from IR gen that we can emit IR for every
constant we can fold, since it isn't currently true in C++11, to fix PR11676.
Original comment from r147271:
constexpr: perform zero-initialization prior to / instead of performing a
constructor call when appropriate. Thanks to Eli for spotting this.
projects / clang / log