Daniel Dunbar [Tue, 13 Mar 2012 18:21:17 +0000 (18:21 +0000)]
[Basic] Stop using a SmallVector<> for Diagnostic. This drops Clang binary size
by ~%.3/~100k in my build -- simply by eliminating the horrible code bloat coming
from the .clear() of the SmallVector<FixItHint>, which does a std::~string, etc.
- My understanding is we don't ever emit arbitrary numbers of fixits, so I just
moved us to using a statically sized array like we do for arguments and
ranges.
Richard Smith [Tue, 13 Mar 2012 07:21:50 +0000 (07:21 +0000)]
Implement [temp.param]p5: the top-level cv-qualifiers on a non-type template
parameter's declaration are ignored when determining the parameter's type.
Richard Smith [Tue, 13 Mar 2012 03:12:56 +0000 (03:12 +0000)]
Fix PR10447: lazily building name lookup tables for DeclContexts was broken.
The deferred lookup table building step couldn't accurately tell which Decls
should be included in the lookup table, and consequently built different tables
in some cases.
Fix this by removing lazy building of DeclContext name lookup tables. In
practice, the laziness was frequently not worthwhile in C++, because we
performed lookup into most DeclContexts. In C, it had a bit more value,
since there is no qualified lookup.
In the place of lazy lookup table building, we simply don't build lookup tables
for function DeclContexts at all. Such name lookup tables are not useful, since
they don't capture the scoping information required to correctly perform name
lookup in a function scope.
The resulting performance delta is within the noise on my testing, but appears
to be a very slight win for C++ and a very slight loss for C. The C performance
can probably be recovered (if it is a measurable problem) by avoiding building
the lookup table for the translation unit.
[Sema] Prefer to use ObjCInterfaceDecl's protocol_begin()/protocol_end() iterators instead of
ObjCInterfaceDecl::getReferencedProtocols(), because the iterators are safe to use
even if the caller did not check that the interface is a definition.
John McCall [Tue, 13 Mar 2012 00:37:01 +0000 (00:37 +0000)]
Alternate fix to PR12248: put Sema in charge of special-casing
the diagnostic for assigning to a copied block capture. This has
the pleasant side-effect of letting us special-case the diagnostic
for assigning to a copied lambda capture as well, without introducing
a new non-modifiable enumerator for it.
Bill Wendling [Mon, 12 Mar 2012 21:22:35 +0000 (21:22 +0000)]
Have clang pay attention to the LIBRARY_PATH environment variable.
The LIBRARY_PATH environment variable should be honored by clang. Have the
driver pass the directories to the linker.
<rdar://problem/9743567> and PR10296.
Douglas Gregor [Mon, 12 Mar 2012 21:09:16 +0000 (21:09 +0000)]
C++11 [temp.deduct.call]p6 tweak: when given a set of overlaoded
functions that includes an explicit template argument list, perform
an inner deduction against each of the function templates in that list
and, if successful, use the result of that deduction for the outer
template argument deduction. Fixes PR11713.
[Sema] -Add an invalid objc category to the DeclContext so that it can be
serialized
-Don't add methods of invalid objc containers to the global method pool.
This protects us from trying to serialize a method whose container was not
serialized.
Richard Smith [Mon, 12 Mar 2012 07:56:15 +0000 (07:56 +0000)]
Fix parsing of type-specifier-seq's. Types are syntactically allowed to be
defined here, but not semantically, so
new struct S {};
is always ill-formed, even if there is a struct S in scope.
We also had a couple of bugs in ParseOptionalTypeSpecifier caused by it being
under-loved (due to it only being used in a few places) so merge it into
ParseDeclarationSpecifiers with a new DeclSpecContext. To avoid regressing, this
required improving ParseDeclarationSpecifiers' diagnostics in some cases. This
also required teaching ParseSpecifierQualifierList about constexpr... which
incidentally fixes an issue where we'd allow the constexpr specifier in other
bad places.
NAKAMURA Takumi [Sun, 11 Mar 2012 08:04:40 +0000 (08:04 +0000)]
Update library dependencies to add clangEdit on c-index-test and c-arcmt-test.
To link with -static -lclang, linker tries to seek not libclang.so, clang.dll nor libclang.dll.a, but libclang.a. USEDLIBS should have correct dependencies for -static.
(In contrast, USEDLIBS=libclang.so might be enough w/o -static)
FYI, cygwin build (in buildbot) is using -static, due to avoiding weirdness of extremely slower startup lag of clang.exe.
llvm::SwitchInst
Renamed methods caseBegin, caseEnd and caseDefault with case_begin, case_end, and case_default.
Added some notes relative to case iterators.
Douglas Gregor [Sun, 11 Mar 2012 03:29:50 +0000 (03:29 +0000)]
When template argument deduction is ignoring qualifiers, perform deep
structural comparison of non-dependent types. Otherwise, we end up
rejecting cases where the non-dependent types don't match due to
qualifiers in, e.g., a pointee type. Fixes PR12132.
Douglas Gregor [Sat, 10 Mar 2012 23:52:41 +0000 (23:52 +0000)]
When determining whether an identifier followed by a '<' in a member
access expression is the start of a template-id, ignore function
templates found in the context of the entire postfix-expression. Fixes
PR11856.
Aaron Ballman [Sat, 10 Mar 2012 22:21:14 +0000 (22:21 +0000)]
No longer defining GNUC mode when compiling for Microsoft compatibility. This allows people's cross-platform compiler-specific macros to work properly.
John McCall [Sat, 10 Mar 2012 09:33:50 +0000 (09:33 +0000)]
Remove BlockDeclRefExpr and introduce a bit on DeclRefExpr to
track whether the referenced declaration comes from an enclosing
local context. I'm amenable to suggestions about the exact meaning
of this bit.
Nick Lewycky [Sat, 10 Mar 2012 07:47:07 +0000 (07:47 +0000)]
Slightly tweak this condition. "isTransparentContext()" was checking whether an
enum is scoped or not, which is not relevant here. Instead, phrase the loop in
the same terms that the standard uses, instead of this awkward set of
conditions that is *nearly* equal.
Nick Lewycky [Sat, 10 Mar 2012 07:45:33 +0000 (07:45 +0000)]
Could not find this in C99. Perhaps this rule comes from a DR, but in any case
please annotate it with a note explaining why this wrong-seeming behaviour is
correct.
Remove the part of test/Sema/pragma-pack-2.c that has irregular "#pragma pack"s.
Before r151307 this part worked without compiler errors but now it only verifies
that our handling of irregular pragmas is broken compared to gcc, it has no
practical usefulness; it creates invalid structs that cannot be used for 'offsetof' testing.
If we later decide to handle irregular pragmas without compiler errors we can
put back this part.
John McCall [Sat, 10 Mar 2012 03:05:10 +0000 (03:05 +0000)]
Unify the BlockDeclRefExpr and DeclRefExpr paths so that
we correctly emit loads of BlockDeclRefExprs even when they
don't qualify as ODR-uses. I think I'm adequately convinced
that BlockDeclRefExpr can die.
Ted Kremenek [Sat, 10 Mar 2012 01:34:17 +0000 (01:34 +0000)]
[analyzer] fix regression in analyzer of NOT actually aborting on Stmts it doesn't understand. We registered
as aborted, but didn't treat such cases as sinks in the ExplodedGraph.
Along the way, add basic support for CXXCatchStmt, expanding the set of code we actually analyze (hopefully correctly).
Daniel Dunbar [Sat, 10 Mar 2012 01:03:58 +0000 (01:03 +0000)]
IRgen/ABI/x86_64: Avoid passing small structs using byval sometimes.
- We do this when it is easy to determine that the backend will pass them on
the stack properly by itself.
Currently LLVM codegen is really bad in some cases with byval, for example, on
the test case here (which is derived from Sema code, which likes to pass
SourceLocations around)::
struct s47 { unsigned a; };
void f47(int,int,int,int,int,int,struct s47);
void test47(int a, struct s47 b) { f47(a, a, a, a, a, a, b); }
Richard Smith [Sat, 10 Mar 2012 00:28:11 +0000 (00:28 +0000)]
Assign APValues by swapping from a temporary. Removes a bunch of unnecessary
copy-construction, which Daniel Dunbar reports as giving a 0.75% speedup on
403.gcc/combine.c. The performance differences on my constexpr torture tests
are below the noise floor.
Richard Smith [Fri, 9 Mar 2012 22:27:51 +0000 (22:27 +0000)]
Improve diagnostics for UCNs referring to control characters and members of the
basic source character set in C++98. Add -Wc++98-compat diagnostics for same in
literals in C++11. Extend such support to cover string literals as well as
character literals, and mark N2170 as done.
This seems too minor to warrant a release note to me. Let me know if you disagree.
Anna Zaks [Fri, 9 Mar 2012 21:14:01 +0000 (21:14 +0000)]
[analyzer] Add support for NoRedundancy inlining mode.
We do not reanalyze a function, which has already been analyzed as an
inlined callee. As per PRELIMINARY testing, this gives over
50% run time reduction on some benchmarks without decreasing of the
number of bugs found.
Daniel Dunbar [Fri, 9 Mar 2012 20:41:19 +0000 (20:41 +0000)]
Revert r152430 "[AST] Mark ASTContext::getTypeInfo and some friends as
LLVM_READONLY.", getTypeInfo() is totally not READONLY, which I should have
probably noticed given that I made it so mere hours ago. Oops.
When an error made a record member invalid, the record would stay as "isBeingDefined" and
not "completeDefinition". Even easily recoverable errors ended up propagating records in
such "beingDefined" state, for example:
struct A {
~A() const; // expected-error {{'const' qualifier is not allowed on a destructor}}
};
struct B : A {}; // A & B would stay as "not complete definition" and "being defined".
This weird state was impending lookups in the records and hitting assertion in the ASTWriter.
Daniel Dunbar [Fri, 9 Mar 2012 19:35:29 +0000 (19:35 +0000)]
[AST/etc] Mark {getSourceRange(),getStartLoc(),getEndLoc()} as LLVM_READONLY.
- The theory here is that we have these functions sprinkled in all over the
place. This should allow the optimizer to at least realize it can still do
load CSE across these calls.
- I blindly marked all instances as such, even though the optimizer can infer
this attribute in some instances (some of the inline ones) as that was easier
and also, when given the choice between thinking and not thinking, I prefer
the latter.
You might think this is mere frivolity, but actually this is good for a .7 -
1.1% speedup on 403.gcc/combine.c, JSC/Interpreter.cpp,
OGF/NSBezierPath-OAExtensions.m.
lldb support: under debugger support flag, when sending message
to forward class, and assigning to an 'id' type var, message
sends default to 'id'. // rdar"//10988847
Daniel Dunbar [Fri, 9 Mar 2012 18:35:03 +0000 (18:35 +0000)]
[AST/Sema/libclang] Replace getSourceRange().getBegin() with getLocStart().
- getSourceRange().getBegin() is about as awesome a pattern as .copy().size().
I already killed the hot paths so this doesn't seem to impact performance on my
tests-of-the-day, but it is a much more sensible (and shorter) pattern.
Daniel Dunbar [Fri, 9 Mar 2012 15:39:19 +0000 (15:39 +0000)]
[AST] Reimplement Stmt::getLoc{Start,End} to dispatch to subclass overloads.
- getSourceRange() can be very expensive, we should try to avoid it if at all possible.
In conjunction with the previous commit I measured a ~2% speedup on 403.gcc/combine.c and a 3% speedup on OmniGroupFrameworks/NSBezierPath-OAExtensions.m.
Daniel Dunbar [Fri, 9 Mar 2012 15:39:15 +0000 (15:39 +0000)]
[AST] Add {DeclRefExpr,MemberExpr,ImplicitCastExpr}::{getLocStart,getLocEnd} methods.
- There are probably a lot more of these worth adding, but these are a start at hitting some of the exprs for which getSourceRange().getBegin() is a poor substitute for getLocStart().
projects / clang / log