Ted Kremenek [Tue, 20 Oct 2009 21:39:41 +0000 (21:39 +0000)]
Add destructor and cleanup code to LocationContext (fixing some leaks). Along the way, have
AnalysisManager periodically cleanup its AnalysisContextManager and LocationContextManager objects,
as they don't need to forever retain all the CFGs ever created when analyzing a file.
Changed text of warning in my last patch.
Related to radar:
<rdar://problem/7308503> clang should disallow the trailing semicolon in method definitions
Issue warning if method body starts with a semicolon.
Fixes
<rdar://problem/7308503> clang should disallow the trailing semicolon in method definitions
Steve Naroff [Tue, 20 Oct 2009 14:46:24 +0000 (14:46 +0000)]
- Extend clang_createIndex() to support PCH and diagnostic 'filtering'. This seems cleaner to me without sacrificing much flexibility.
- Remove clang_wantOnlyLocalDeclarations().
- Remove 'displayDiagnostics' arguments to clang_createTranslationUnitFromSourceFile() and clang_createTranslationUnit().
- Have clang_createTranslationUnitFromSourceFile() strip the '-o <outfile>' command line arguments if they exist. Document this semantic in the header. Also verify we have a valid ASTUnit before telling it to 'unlinkTemporaryFile()'.
Eli Friedman [Tue, 20 Oct 2009 08:27:19 +0000 (08:27 +0000)]
Remove default argument for ImpCastExprToType. Add appropriate argument
to all callers. Switch a few other users of CK_Unknown to proper cast
kinds.
Note that there are still some situations where we end up with
CK_Unknown; they're pretty easy to find with grep. There
are still a few missing conversion kinds, specifically
pointer/int/float->bool and the various combinations of real/complex
float/int->real/complex float/int.
Ted Kremenek [Tue, 20 Oct 2009 06:31:34 +0000 (06:31 +0000)]
Call 'clear()' in ~RopePieceBTreeLeaf(), decrementing the reference
counts of the bufffers referened by the RopePieces in
RopePieceBTreeLeaf. This (I believe) corrently fixes the leak I meant
to fix in r84601 (which ended up causing an overrelease).
Douglas Gregor [Tue, 20 Oct 2009 05:58:46 +0000 (05:58 +0000)]
Handle substitutions into the "first qualifier in scope" of a
qualified member access expression (e.g., t->U::member) when that
first qualifier refers to a template parameters.
Ted Kremenek [Tue, 20 Oct 2009 05:49:17 +0000 (05:49 +0000)]
BumpVectorContext: Use 'unsigned' integer type with PointerIntUnion instead of bool to keep it clear that we are reasoning about an unsigned integer with a single bit.
Chris Lattner [Tue, 20 Oct 2009 05:36:05 +0000 (05:36 +0000)]
Implement PR5242: don't desugar a type more than once in a diagnostic. This
implements a framework that allows us to use information about previously
substituted values to simplify subsequent ones. Maybe this would be useful
for C++'y stuff, who knows. We now get:
t.c:4:21: error: invalid operands to binary expression ('size_t' (aka 'unsigned long *') and 'size_t')
return (size_t) 0 + (size_t) 0;
~~~~~~~~~~ ^ ~~~~~~~~~~
on the testcase. Note that size_t is only aka'd once.
Chris Lattner [Tue, 20 Oct 2009 05:25:22 +0000 (05:25 +0000)]
teach FormatDiagnostic to aggregate previously formatted arguments and
pass them down into the ArgToStringFn implementation. This allows
redundancy across operands to a diagnostic to be eliminated.
Ted Kremenek [Tue, 20 Oct 2009 05:25:11 +0000 (05:25 +0000)]
Fix a reference count imbalance in RewriteRope::MakeRopeString().
This was causing a ton of memory to be leaked when using HTML
diagnostics with the static analyzer (on large files with many errors).
Mike Stump [Tue, 20 Oct 2009 02:12:22 +0000 (02:12 +0000)]
Refine the type of the first parameter to block invoke functions.
WIP. I have yet to find the magic incantation to get the structure
type to be defined. If someone has a pointer, love to hear it.
Ted Kremenek [Mon, 19 Oct 2009 22:15:09 +0000 (22:15 +0000)]
Really fix <rdar://problem/7312058> by adding a 'displayDiagnostics' option to
clang_createTranslationUnit() and clang_createTranslationUnitFromSourceFile(). The user can now
specify if the diagnostics from Clang are printed to stderr or are silenced completely. We can
obviously evolve this API to be more general in the future.
Note: Added a FIXME since I wasn't certain what was the best way to redirect to something analogous
to '/dev/null' on Windows.
Ted Kremenek [Mon, 19 Oct 2009 21:44:57 +0000 (21:44 +0000)]
The constructor for ASTUnit now takes a DiagnosticClient*, allowing uses of ASTUnit to specify
alternate DiagnosticClients. To match this API, ASTUnit::LoadFromPCHFile() now takes a corresponding
DiagnosticClient* argument as well. The DiagnosticClient object is destroyed when the ASTUnit object
is destroyed.
The CIndex library now uses this API to create a 'IgnoreDiagnosticsClient' that simply silences
diagnostics when using the clang_createTranslationUnitFromSourceFile() function. This fixes
<rdar://problem/7312058>. This API can change in the future as we add more flexibility for clients.
Ted Kremenek [Mon, 19 Oct 2009 21:17:25 +0000 (21:17 +0000)]
Re-order includes so that the logic involving '#ifdef LLVM_ON_WIN32' appears after the main #includes. The ultimate solution is to just use LLVM-portable methods in llvm/System.
Steve Naroff [Mon, 19 Oct 2009 14:34:22 +0000 (14:34 +0000)]
Move Diagnostic/DiagClient/FileManager from Indexer => ASTUnit.
Removing this shared data should enable clang_createTranslationUnit/clang_createTranslationUnitFromSourceFile to be run from multiple threads (related to <rdar://problem/7303432>).
Daniel Dunbar [Mon, 19 Oct 2009 09:11:21 +0000 (09:11 +0000)]
Workaround a bug exposed by the FileCheckify of message-length.c, the caret end
column computation isn't correct and could exceed the line length, which
resulted in a buffer overflow later.
- Chris, is there a better way for this code to compute the final column used
by the caret?
Daniel Dunbar [Mon, 19 Oct 2009 01:21:12 +0000 (01:21 +0000)]
Update NamedDecl::getName() to work for empty names.
- I'm not sure this is ideal, but otherwise clients must be overly careful when handling decl's which can have empty names.
Anders Carlsson [Sun, 18 Oct 2009 23:09:21 +0000 (23:09 +0000)]
When binding a reference to a temporary, it's important that other temporaries created as on the RHS are destroyed before emitting the dtor for the temporary.
Nate Begeman [Sun, 18 Oct 2009 20:10:40 +0000 (20:10 +0000)]
Support IRgen of OpenCL vector initializers, ensuring the resulting IR is in a
form that LLVM code generators can turn into efficient code. For example,
int4 a, b, c;
a = (int4)(b.yzw, a.x)
The AuroraUX toolchain has conflicting wchar_t between the system stdlib.h header and the clang stddef.h header where clang was defining as int where we use long.
John McCall [Sun, 18 Oct 2009 09:09:24 +0000 (09:09 +0000)]
When performing template-substitution into a type, don't just replace the
TemplateTypeParmType with the substituted type directly; instead, replace it
with a SubstTemplateTypeParmType which will note that the type was originally
written as a template type parameter. This makes it reasonable to preserve
source information even through template substitution.
Also define the new SubstTemplateTypeParmType class, obviously.
For consistency with current behavior, we stringize these types as if they
were the underlying type. I'm not sure this is the right thing to do.
At any rate, I paled at adding yet another clause to the don't-desugar 'if'
statement, so I extracted a function to do it. The new function also does
The Right Thing more often, I think: e.g. if we have a chain of typedefs
leading to a vector type, we will now desugar all but the last one.
Daniel Dunbar [Sun, 18 Oct 2009 02:09:17 +0000 (02:09 +0000)]
Simplify HandleFormatAttr.
- I have this crazy dream that one day someone will invent a miraculous tool so
that developers, instead of hand optimizing their source code to obscure its
intent and decrease its maleability, will instead write what they mean, and
this strange and wonderful tool -- which I imagine would be called something
fancy sounding like "an optimizing compiler" -- will make their code fast
*for* them. With all the saved time, developers could maybe even focus on
making the magic "optimizing compiler" better!!
- No intended functionality change, all though I expect the universe to mock me
for snarkiness.
John McCall [Sun, 18 Oct 2009 01:05:36 +0000 (01:05 +0000)]
Clone the full Type hierarchy into the TypeLoc hierarchy. Normalize
TypeLoc class names to be $(Type classname)Loc. Rewrite the visitor.
Provide skeleton implementations for all the new TypeLocs.
Handle all cases in PCH. Handle a few more cases when inserting
location information in SemaType.
It should be extremely straightforward to add new location information
to existing TypeLoc objects now.
Douglas Gregor [Sat, 17 Oct 2009 21:40:42 +0000 (21:40 +0000)]
When type-checking a C++ "new" expression, don't type-check the actual
initialization if any of the constructor/initialization arguments are
type-dependent. Fixes PR5224.
projects / clang / log