Ted Kremenek [Thu, 3 Dec 2009 03:27:11 +0000 (03:27 +0000)]
Add batch version of 'StoreManager::InvalidateRegion()' for invalidating multiple regions as once. After adopting this in the CFRefCount::EvalCall(), we see a reduction in analysis time of 1.5% when analyzing all of SQLite3.
Daniel Dunbar [Thu, 3 Dec 2009 01:54:28 +0000 (01:54 +0000)]
Switch clang_createTranslationUnitFromSourceFile to use ASTUnit::LoadFromCommandLine.
- This is much faster, as it avoids the overhead of dumping an AST file to disk and reloading it.
- For debugging purposes, there is a clang_setUseExternalASTGeneration hook which can be used to disable this.
On the Sketch Cocoa app, the speedup is pretty nice, especially when using a PCH file while scanning the source:
Wall time to c-index-test all files (no PCH):
Old: 23.4221
New: 12.3884
Wall time to c-index-test all files (with a PCH, and "local" mode):
Old: 10.9233
New: 1.9038
John McCall [Thu, 3 Dec 2009 00:58:24 +0000 (00:58 +0000)]
Stop stripping UnresolvedUsingDecls out of LookupResults that have other
results in them (which we were doing intentionally as a stopgap). Fix
an DeclContext lookup-table ordering problem which was causing UsingDecls to
show up incorrectly when looking for ordinary results. And oh hey
Clang-Code-Syntax passes now.
Ted Kremenek [Thu, 3 Dec 2009 00:46:16 +0000 (00:46 +0000)]
Add a heuristic to the dead stores checker to prune dead stores for variables annotated with '__block'. This is overly conservative, but now the analyzer doesn't report dead stores for variables that can be updated by a block call.
Mike Stump [Wed, 2 Dec 2009 23:46:02 +0000 (23:46 +0000)]
Cleanups on exceptional edges don't work at all, yet. This doesn't
catch very many of them and if we caught all of them, the errors would
be annoying. I'm working on this next. WIP.
Douglas Gregor [Wed, 2 Dec 2009 22:36:29 +0000 (22:36 +0000)]
Improve source location information for C++ member initializers in a
constructor, by keeping the DeclaratorInfo* rather than just the type
and a single location.
Daniel Dunbar [Wed, 2 Dec 2009 21:47:55 +0000 (21:47 +0000)]
Normalize CIndex/c-index-test/index-test link lines in the hopes it will fix
*something*.
- We really need to fix how LLVM's build systems manage
linking. Pretty-please-someone-else-do-this? :)
Mike Stump [Wed, 2 Dec 2009 19:53:57 +0000 (19:53 +0000)]
Pull the terminate handler up so that we can use it for the catch
parameter setup code and set up the catch parameter setup code to
protect that code with terminate.
Douglas Gregor [Wed, 2 Dec 2009 18:12:28 +0000 (18:12 +0000)]
Extend -remap-file=from;to to permit mapping from a non-existent
file. This is accomplished by introducing the notion of a "virtual"
file into the file manager, which provides a FileEntry* for a named
file whose size and modification time are known but which may not
exist on disk.
Added a cute little test that remaps both a .c file and a .h file it
includes to alternative files.
Douglas Gregor [Wed, 2 Dec 2009 09:21:34 +0000 (09:21 +0000)]
Extend the CIndex code-completion API to perform code completion
involving unsaved files, using the -remap-file= functionality recently
added to clang-cc.
John McCall [Wed, 2 Dec 2009 08:25:40 +0000 (08:25 +0000)]
Rip out the last remaining implicit use of OverloadedFunctionDecl in Sema:
LookupResult::getAsSingleDecl() is no more. Shift Sema::LookupSingleName to
return null on overloaded results.
Douglas Gregor [Wed, 2 Dec 2009 08:08:39 +0000 (08:08 +0000)]
Introduce a new clang-cc option
-remap-file=from;to
which takes the file "from" and transparently replaces its contents
with the contents of the file "to" from the source manager's
perspective. This is the moral equivalent of
cp from saved
cp to from
<call clang>
cp saved from
rm saved
John McCall [Wed, 2 Dec 2009 08:04:21 +0000 (08:04 +0000)]
Push overloaded function templates through the parser using a totally different
leaked data structure than before. This kills off the last remaining
explicit uses of OverloadedFunctionDecl in Sema.
Mike Stump [Wed, 2 Dec 2009 07:41:41 +0000 (07:41 +0000)]
More exception handling improvements... WIP.
Highlights include:
Add a helper to generate __cxa_free_exception and _ZSt9terminatev.
Add a region to handle EH object deallocation for ctor failures for throw.
Add a terminate handler for __cxa_end_catch.
A framework for adding cleanup actions for the exceptional edges only.
Douglas Gregor [Wed, 2 Dec 2009 06:49:09 +0000 (06:49 +0000)]
Extend the source manager with the ability to override the contents of
files with the contents of an arbitrary memory buffer. Use this new
functionality to drastically clean up the way in which we handle file
truncation for code-completion: all of the truncation/completion logic
is now encapsulated in the preprocessor where it belongs
(<rdar://problem/7434737>).
John McCall [Wed, 2 Dec 2009 03:53:29 +0000 (03:53 +0000)]
Stop trying to analyze class-hierarchies for dependently-scoped id-expressions;
there's nothing interesting we can say now that we're correctly not requiring
the qualifier to name a known base class in dependent contexts.
Require scope specifiers on member access expressions to name complete types
if they're not dependent; delay lookup when they are dependent.
Use more appropriate diagnostics when qualified implicit member access
expressions find declarations from unrelated classes.
Daniel Dunbar [Wed, 2 Dec 2009 03:23:45 +0000 (03:23 +0000)]
Add ASTUnit::LoadFromCommandLine, which creates an ASTUnit out of a list of
(clang/driver) command line arguments (including the source file).
- The arguments are expected to include the source file.
- The idea is that even though this is a somewhat odd API, its the form which
many tools can most easily use (for example, by interposing with the compiler).
Also, switch index-test's -ast-from-source to use this entry point, and provide
a -arg command line argument which can be used to test that the command line
arguments are handled correctly.
Daniel Dunbar [Tue, 1 Dec 2009 23:07:57 +0000 (23:07 +0000)]
In SourceManager::isBeforeInTranslationUnit, if we are trying to compare two source locations with no common ancestor in the include stack, determine order by assuming memory buffers preceed files, and then that FileIDs are created in order.
The later assumption is patently false, but this was already broken -- this situation is conceptually impossible, my feeling is we should fix SourceManager and friends to make it impossible in practice as well. However, we need to fix PR5662 and perhaps some other things involving memory buffers first. In the short term I'm pretty sure this is reliable.
Chris, Argiris, is this going to break anything that wasn't already broken?
John McCall [Tue, 1 Dec 2009 22:10:20 +0000 (22:10 +0000)]
Rework how we support C++ implicit member accesses. If we can resolve an
implicit member access to a specific declaration, go ahead and create
it as a DeclRefExpr or a MemberExpr (with implicit CXXThisExpr base) as
appropriate. Otherwise, create an UnresolvedMemberExpr or
DependentScopeMemberExpr with a null base expression.
By representing implicit accesses directly in the AST, we get the ability
to correctly delay the decision about whether it's actually an instance
member access or not until resolution is complete. This permits us
to correctly avoid diagnosing the 'problem' of 'MyType::foo()'
where the relationship to the type isn't really known until instantiation.
(objc2 nonfragile-abi specific). If the translation unit includes an implementation
of a subclass (direct or indirect) of a weak_import root class, emit a weak reference
for the root class's metaclass (should complete radar 6815425).
Douglas Gregor [Tue, 1 Dec 2009 17:35:23 +0000 (17:35 +0000)]
Funtion templates and function template specializations do not
override virtual functions. Also, eliminate a (now redundant) call to
AddOverriddenMethods.
Douglas Gregor [Tue, 1 Dec 2009 17:24:26 +0000 (17:24 +0000)]
Move the checking of overridden virtual functions into the code path
common to both parsing and template instantiation, so that we'll find
overridden virtuals for member functions of class templates when they
are instantiated.
Additionally, factor out the checking for pure virtual functions, so
that it will be executed both at parsing time and at template
instantiation time.
These changes fix PR5656 (for real), although one more tweak
w.r.t. member function templates will be coming along shortly.
Douglas Gregor [Tue, 1 Dec 2009 16:58:18 +0000 (16:58 +0000)]
Don't automatically assume that an id-expression refers to a
ValueDecl, because that isn't always the case in ill-formed
code. Diagnose a common mistake (forgetting to provide a template
argument list for a class template, PR5655) and dyn_cast so that we
handle the general problem of referring to a non-value declaration
gracefully.
projects / clang / log