Chandler Carruth [Mon, 15 Nov 2010 07:15:26 +0000 (07:15 +0000)]
Clean up some names and fix the handling of default sysroots on Windows and
other platforms where the textual default of '/' isn't the system's root
directory. We should probably still make the textual default platform specific,
but this should avoid the particularly bad problem with the previous state: we
applied a sysroot of '/' to '/usr/local/google' which added
'//usr/local/include' to the windows header search path, a share on another
machine named 'usr'. Oops.
Ted Kremenek [Sun, 14 Nov 2010 17:47:35 +0000 (17:47 +0000)]
"Fix" some unintentional fallout from converting
the Stmt* visitation in CursorVisitor to be
data-recursive.
Since AnnotationTokensWorker explicitly calls
CursorVisitor::VisitChildren(), it essentially
transforms the data-recursive algorithm in
CursorVisitor back into a non-data recursive one.
This is particularly bad because the data-recursive
algorithm uses more stack space per stack frame,
which can cause us to blow the stack in some cases.
"Fix" this by making the stack that AnnotationTokensWorker
runs in really huge. The real fix is to modify
AnnotationTokensWorker not to do the explicit
recursive call.
Ted Kremenek [Sun, 14 Nov 2010 17:04:23 +0000 (17:04 +0000)]
Clean up initialization of LangOptions object by splitting part of it into a parent class
'LangOptionsBase' that default initializes all flags to 0. Now the ctor of LangOptions
only initializes the flags that might be non-zero (significantly simplifying things).
This also appeases Valgrind on Mac OS X, which no longer reports that one of the flags
may be uninitialized.
Zhongxing Xu [Sun, 14 Nov 2010 15:23:50 +0000 (15:23 +0000)]
Revert r118991.
Elidable CXXConstructExpr should inhibit calling destructor for temporary
that is copied, not the one created. This is because eliding copy constructor
means that the object that was to be copied will be constructed directly in
memory the copy would be constructed in.
John McCall [Sun, 14 Nov 2010 09:40:28 +0000 (09:40 +0000)]
Always emit full loads from volatile complex l-values.
Return the result of a complex assignment with the original values,
not by performing a load from the l-value; this is the correct
semantics in C, although not in C++.
John McCall [Sun, 14 Nov 2010 08:17:51 +0000 (08:17 +0000)]
Add a few more complex-related cast kinds that arise due to arbitrary
implicit conversions; the last batch was specific to promotions.
I think this is the full set we need. I do think dividing the cast
kinds into floating and integral is probably a good idea.
Annotate a *lot* more C casts with useful cast kinds.
Block API patch to do copy ctor of copied-in cxx objects in
copy helper function and dtor of copied cxx objects
in dispose helper functions. __block variables
TBD next.
Douglas Gregor [Sat, 13 Nov 2010 19:36:57 +0000 (19:36 +0000)]
When we're type-checking the result of calling a conversion function
(while computing user conversion sequences), make sure that a result
of class type is a complete class type. Had we gone through
ActOnCallExpr, this would have happened when we built the CallExpr.
John McCall [Sat, 13 Nov 2010 05:51:15 +0000 (05:51 +0000)]
Pre-compute all possible usual-arithmetic-conversions results for the
promoted arithmetic types for which builtin operator candidates are
emitted. A few of these need further analysis.
Removes all the uses of UsualArithmeticConversionsType except the
core function in SemaExpr.cpp.
Ted Kremenek [Sat, 13 Nov 2010 05:04:52 +0000 (05:04 +0000)]
Teach IdempotentOperations::PathWasCompletelyAnalyzed to also consider items remaining in the
worklist that could have impacted the evaluation of a block.
Ted Kremenek [Sat, 13 Nov 2010 05:04:49 +0000 (05:04 +0000)]
Add GRWorkList::VisitItemsInWorkList() to allow a client to introspect the contents of a worklist.
This API required changing the BFS worklist to use a deque instead of a queue, but that is better
for performance reasons anyway.
Bob Wilson [Fri, 12 Nov 2010 17:24:52 +0000 (17:24 +0000)]
Increase VectorTypeBitfields::VecKind field from 2 to 3 bits.
With the addition of 2 enum values for Neon vectors, this field must now
hold 6 different values and so requires 3 bits. Make the NumElements field
one bit smaller to compensate.
John McCall [Fri, 12 Nov 2010 08:19:04 +0000 (08:19 +0000)]
Replace one hack with a different hack: strip out the ObjectType
parameters to the Transform*Type functions and instead call out
the specific cases where an object type and the unqualified lookup
results are important. Fixes an assert and failed compile on
a testcase from PR7248.
Douglas Gregor [Fri, 12 Nov 2010 07:15:47 +0000 (07:15 +0000)]
Make sure to always check the result of
SourceManager::getPresumedLoc(), so that we don't try to make use of
an invalid presumed location. Doing so can cause crashes.
Ted Kremenek [Fri, 12 Nov 2010 04:25:07 +0000 (04:25 +0000)]
Fix use of an uninitialized SourceLocation because DeclarationNameLoc failed to completely zero-initialize itself.
Now we explicitly memset all of its values.
This bug was uncovered by the 'Index/recursive-cxx-member-calls.cpp', which exhibited an assertion
on an i386 darwin build of clang. Adding this test case back since the assertion is now resolved.
Douglas Gregor [Fri, 12 Nov 2010 03:34:06 +0000 (03:34 +0000)]
When performing initialization of a copy of a temporary object, use
direct-initialization (rather than copy-initialization) to initialize
the temporary, allowing explicit constructors. Fixes PR8342.
Ted Kremenek [Thu, 11 Nov 2010 23:10:10 +0000 (23:10 +0000)]
RegionStore/BasicStore: do not return UndefinedVal for accesses to concrete addresses; instead return UnknownVal. This
leads it up to checkers (e.g., DereferenceChecker) to guard against illegal accesses (e.g., null dereferences).
Douglas Gregor [Thu, 11 Nov 2010 20:45:16 +0000 (20:45 +0000)]
Teach debug-info generation that SourceManager::getPresumedLoc() can
produce an invalid location even when given a valid location, if the
file system has changed underneath us. Recovery more gracefully.
Ted Kremenek [Thu, 11 Nov 2010 08:05:23 +0000 (08:05 +0000)]
Annotate tokens in a separate thread to avoid blowing out stack space. While the CursorVisitor
is gradually becoming more data recursive, AnnotateTokensVisitor does its own recursive call
within the visitor that can still blow out the stack. This can potentially be reworked to avoid this,
but for now just do token annotation on a separate thread.
Ted Kremenek [Thu, 11 Nov 2010 08:05:18 +0000 (08:05 +0000)]
Generalize data-recursive visitation in CursorVisitor to also handle MemberExprs
and CXXCallMemberExprs. This scheme is hopefully general enough to extend to the
rest of the visitor if necessary.
Douglas Gregor [Thu, 11 Nov 2010 00:39:14 +0000 (00:39 +0000)]
Improve ASTUnit's capture of diagnostics so that the
diagnostic-capturing client lives as long as the ASTUnit itself
does. Otherwise, we can end up with crashes when we get a diagnostic
outside of parsing/code completion. The circumstances under which this
happen are really hard to reproduce, because a file needs to change
from under us.
John McCall [Wed, 10 Nov 2010 23:38:19 +0000 (23:38 +0000)]
When -Wconversion computes the range of a type, it uses the (bit-)range
of the enumerators rather than the actual expressible range. This is
great when dealing with opaque *values* of that type, but when computing
the range of the type for purposes of converting *into* it, it produces
warnings in cases we don't care about (e.g. enum_t x = 500;). Divide
the logic into these two cases and use the more conservative range for
targets.
Bob Wilson [Wed, 10 Nov 2010 21:56:12 +0000 (21:56 +0000)]
Add a variant of GCC-style vector types for ARM NEON.
NEON vector types need to be mangled in a special way to comply with ARM's ABI,
similar to some of the AltiVec-specific vector types. This patch is mostly
just renaming a bunch of "AltiVecSpecific" things, since they will no longer
be specific to AltiVec. Besides that, it just adds the new "NeonVector" enum.
Douglas Gregor [Wed, 10 Nov 2010 19:44:59 +0000 (19:44 +0000)]
Instantiate class member template partial specialization declarations
in the order they occur within the class template, delaying
out-of-line member template partial specializations until after the
class has been fully instantiated. This fixes a regression introduced
by r118454 (itself a fix for PR8001).
John McCall [Wed, 10 Nov 2010 07:01:40 +0000 (07:01 +0000)]
Propagate the deprecated and unavailable attributes from a
@property declaration to the autogenerated methods. I'm uncertain
whether this should apply to attributes in general, but these are
a reasonable core.
Ted Kremenek [Wed, 10 Nov 2010 05:59:39 +0000 (05:59 +0000)]
Region-allocate all AttributeList objects from a factory object instead of manually managing them
using new/delete and OwningPtrs. After memory profiling Clang, I witnessed periodic leaks of these
objects; digging deeper into the code, it was clear that our management of these objects was a mess. The ownership rules were murky at best, and not always followed. Worse, there are plenty of error paths where we could screw up.
This patch introduces AttributeList::Factory, which is a factory class that creates AttributeList
objects and then blows them away all at once. While conceptually simple, most of the changes in
this patch just have to do with migrating over to the new interface. Most of the changes have resulted in some nice simplifications.
This new strategy currently holds on to all AttributeList objects during the lifetime of the Parser
object. This is easily tunable. If we desire to have more bound the lifetime of AttributeList
objects more precisely, we can have the AttributeList::Factory object (in Parser) push/pop its
underlying allocator as we enter/leave key methods in the Parser. This means that we get
simple memory management while still having the ability to finely control memory use if necessary.
Note that because AttributeList objects are now BumpPtrAllocated, we may reduce malloc() traffic
in many large files with attributes.
This fixes the leak reported in: <rdar://problem/8650003>
projects / clang / log