John McCall [Wed, 22 Jun 2011 02:32:12 +0000 (02:32 +0000)]
Emit @finally blocks completely lazily instead of forcing their
existence by always threading an edge from the catchall. Not doing
this was previously causing a crash in the very extreme case where
neither the normal cleanup nor the EH catchall was actually reachable:
we would delete the catchall entry block, which would cause us to
delete the entry block of the finally cleanup as well because the
cleanup logic would merge the blocks, which in turn triggered an assert
because later blocks in the finally would still be using values from the
entry. Laziness turns out to be the most elegant solution to the problem.
Sean Hunt [Wed, 22 Jun 2011 02:25:26 +0000 (02:25 +0000)]
Avoid making assumption that this is either a CXXMethodDecl or a
FunctionTemplateDecl. I'm not quite sure what else it could be, though,
and would appreciate some insight.
Sean Hunt [Wed, 22 Jun 2011 01:05:16 +0000 (01:05 +0000)]
Extend the deleted overload candidate note to cover all cases. It
probably shouldn't cover the implicit member case, but it needs to until
more descriptive diagnostics are implemented.
John McCall [Wed, 22 Jun 2011 00:53:57 +0000 (00:53 +0000)]
It is possible to request the nonfragile ABI with -fobjc-abi-version=2;
respect that when deciding whether -objc-exceptions implies the
-fexceptions -cc1 option.
Sean Hunt [Tue, 21 Jun 2011 23:42:56 +0000 (23:42 +0000)]
Attempt to reapply this patch for caching copy assignment operator
lookup. Previously, it was breaking self-host, but it's been a week and
a half and I can't reproduce, so I need to see if it's still failing.
Chandler Carruth [Tue, 21 Jun 2011 23:04:18 +0000 (23:04 +0000)]
Revert r133526 which re-orders the suggestions for -Wparentheses on ?:
operators. There is a consistent design of having the "silence the
warning" option first.
Bob Wilson [Tue, 21 Jun 2011 21:53:08 +0000 (21:53 +0000)]
Make InitHeaderSearch::AddPath and HeaderSearchOptions::AddPath consistent
use an "IgnoreSysRoot" argument. HeaderSearchOptions had been using the
opposite form with "IsSysRootRelative", which made for much confusion when
looking at true/false values in calls in AddPath. No functional change.
Bob Wilson [Tue, 21 Jun 2011 21:12:29 +0000 (21:12 +0000)]
Fix the default libc++ header search path to be sysrooted. Radar 9639692.
The -cxx-isystem path is not prefixed with the sysroot directory, so it's
not a good way for the driver to set the system default C++ search path.
Instead, add -stdlib as a cc1 option and teach the frontend how to find the
headers. The driver can then just pass -stdlib through to "cc1".
Daniel Dunbar [Tue, 21 Jun 2011 20:55:11 +0000 (20:55 +0000)]
Driver/Darwin: When invoking the linker, automatically pass -object_path_lto so
that the linker has a place to put the temporary object file and can leave it
around (for the driver to clean up). This is important so that the object file
references in the debug info are preserved for possible use by dsymutil.
- <rdar://problem/8294279> executable has no debug symbols when compiled with LTO
objc-arc: Add support for unbridged cast of
__builtin___CFStringMakeConstantString and CF typed function calls
with explicit cf_returns_retained/cf_returns_not_retained attributes.
// rdar://9544832
Daniel Dunbar [Tue, 21 Jun 2011 18:54:46 +0000 (18:54 +0000)]
IRgen: Add a -fuse-register-sized-bitfield-access option, for testing.
- Changes bit-field access policy to try to use (aligned) register sized accesses.
The idea here is that by using larger accesses we expose more coalescing
potential to the backend when we have situations like adjacent bit-fields in the
same structure (which is common), and that the backend should be smart enough to
narrow the accesses down when no coalescing is done or when it is shown not to
be profitable.
Chandler Carruth [Tue, 21 Jun 2011 18:07:33 +0000 (18:07 +0000)]
Switch the order of the notes for the parentheses suggested in the case
of: a + b ? x : y. In our testing of this flag we've yet to hit a single
case where the existing precedence was correct, so we should suggest
grouping the ?: first.
Chandler Carruth [Tue, 21 Jun 2011 17:22:09 +0000 (17:22 +0000)]
Fix a crash when a pointer-to-member function is called in the condition
expression of '?:'. Add a test case for this pattern, and also test the
code that led to the crash in a "working" case as well.
Douglas Gregor [Tue, 21 Jun 2011 17:03:29 +0000 (17:03 +0000)]
Introduce a new AST node describing reference binding to temporaries.
MaterializeTemporaryExpr captures a reference binding to a temporary
value, making explicit that the temporary value (a prvalue) needs to
be materialized into memory so that its address can be used. The
intended AST invariant here is that a reference will always bind to a
glvalue, and MaterializeTemporaryExpr will be used to convert prvalues
into glvalues for that binding to happen. For example, given
const int& r = 1.0;
The initializer of "r" will be a MaterializeTemporaryExpr whose
subexpression is an implicit conversion from the double literal "1.0"
to an integer value.
IR generation benefits most from this new node, since it was
previously guessing (badly) when to materialize temporaries for the
purposes of reference binding. There are likely more refactoring and
cleanups we could perform there, but the introduction of
MaterializeTemporaryExpr fixes PR9565, a case where IR generation
would effectively bind a const reference directly to a bitfield in a
struct. Addresses <rdar://problem/9552231>.
Nick Lewycky [Tue, 21 Jun 2011 00:14:18 +0000 (00:14 +0000)]
Add support for -Wa,--noexecstack when building from a non-assembly file. For
an assembly file it worked correctly, while for a .c file it would given an
error about how --noexecstack is not a supported argument to -Wa.
llvm-gcc treats a tentative definition with a previous
(or follow up) extern declaration with weak_import as
an actual definition. make clang follows this behavior.
// rdar://9538608
llvm-gcc treats an extern declaration with weak_import
Jay Foad [Mon, 20 Jun 2011 14:38:01 +0000 (14:38 +0000)]
Change how PHINodes store their operands.
Change PHINodes to store simple pointers to their incoming basic blocks,
instead of full-blown Uses.
Note that this loses an optimization in SplitCriticalEdge(), because we
can no longer walk the use list of a BasicBlock to find phi nodes. See
the comment I removed starting "However, the foreach loop is slow for
blocks with lots of predecessors".
Extend replaceAllUsesWith() on a BasicBlock to also update any phi
nodes in the block's successors. This mimics what would have happened
when PHINodes were proper Users of their incoming blocks. (Note that
this only works if OldBB->replaceAllUsesWith(NewBB) is called when
OldBB still has a terminator instruction, so it still has some
successors.)
Chandler Carruth [Mon, 20 Jun 2011 08:59:25 +0000 (08:59 +0000)]
Use an explicitly 64-bit triple flag to ensure we can easily verify the
types printed in various diagnostics.
We could omit checking for the types, but that can mask errors where the
wrong type is streamed into the diagnostic. There was at least one of
these caught by this test already.
Chandler Carruth [Mon, 20 Jun 2011 07:52:11 +0000 (07:52 +0000)]
Fix a problem with the diagnostics of invalid arithmetic with function
pointers I found while working on the NULL arithmetic warning. We here
always assuming the LHS was the pointer, instead of using the selected
pointer expression.
Chandler Carruth [Mon, 20 Jun 2011 07:38:51 +0000 (07:38 +0000)]
Move away from the poor "abstraction" I added to Type. John argued
effectively that this abstraction simply doesn't exist. That is
highlighted by the fact that by using it we were papering over a more
serious error in this warning: the fact that we warned for *invalid*
constructs involving member pointers and block pointers.
I've fixed the obvious issues with the warning here, but this is
confirming an original suspicion that this warning's implementation is
flawed. I'm looking into how we can implement this more reasonably. WIP
on that front.
Chris Lattner [Mon, 20 Jun 2011 04:01:35 +0000 (04:01 +0000)]
Update to match mainline ConstantStruct::get API change. Also, use
ConvertType on InitListExprs as they are being converted. This is
needed for a forthcoming patch, and improves the IR generated anyway
(see additional type names in testcases).
This patch also converts a bunch of std::vector's in CGObjCMac to use
C arrays. There are a ton more that should be converted as well.
Jordy Rose [Mon, 20 Jun 2011 03:49:16 +0000 (03:49 +0000)]
[analyzer] Re-enable checking for strncpy, along with a new validation of the size argument. strncat is not yet up-to-date, but I'm leaving it enabled for now (there shouldn't be any false positives, at least...)
Jordy Rose [Mon, 20 Jun 2011 02:06:40 +0000 (02:06 +0000)]
[analyzer] Eliminate "byte string function" from CStringChecker's diagnostics, and make it easier to provide custom messages for overflow checking, in preparation for re-enabling strncpy checking.
Chandler Carruth [Mon, 20 Jun 2011 01:23:19 +0000 (01:23 +0000)]
Restructure the API in Type based on a conversation with Richard Smith.
This makes 'isPointerLikeType' a little less confusing, and pulls the
decay check into a separate interface that is much more clear and
concrete. Also, just implement these as logical wrappers around other
predicates. Having a switch based implementation isn't likely to be
necessary. We can try to optimize them later if they show up on
a profile.
Chandler Carruth [Sun, 19 Jun 2011 09:05:19 +0000 (09:05 +0000)]
Turn -Wnull-arithmetic back on by default -- we now have tests for the
false positives, including those in the GCC test suite, and don't warn
about them.
Let me know if this causes fallout, we can turn it back off if needed.
Chandler Carruth [Sun, 19 Jun 2011 09:05:14 +0000 (09:05 +0000)]
Add test cases for false positives on -Wnull-arithmetic from Richard
Trieu, and fix them by checking for array and function types as well as
pointer types.
I've added a predicate method on Type to bundle together the logic we're
using here: isPointerLikeType(). I'd welcome better names for this
predicate, this is the best I came up with. It's implemented as a switch
to be a touch lighter weight than all the chained isa<...> casts that
would result otherwise.
John McCall [Sat, 18 Jun 2011 07:31:30 +0000 (07:31 +0000)]
A couple of minor changes to the ARC spec, plus a new section
specifying that retain/release/autorelease/retainCount are forbidden,
plus a section talking about the behavior of dealloc.
Chandler Carruth [Sat, 18 Jun 2011 01:19:03 +0000 (01:19 +0000)]
Accept no-return stripping conversions for pointer type arguments after
deducing template parameter types. Recently Clang began enforcing the
more strict checking that the argument type and the deduced function
parameter type (after substitution) match, but that only consideres
qualification conversions.
One problem with this patch is that we check noreturn conversions and
qualification conversions independently. If a valid conversion would
require *both*, perhaps interleaved with each other, it will be
rejected. If this actually occurs (I'm not yet sure it does) and is in
fact a problem (I'm not yet sure it is), there is a FIXME to implement
more intelligent conversion checking.
However, this step at least allows Clang to resume accepting valid code
we're seeing in the wild.
Douglas Gregor [Fri, 17 Jun 2011 23:16:24 +0000 (23:16 +0000)]
Objective-C++ ARC: eliminate the utterly unjustified loophole that
silently dropped ownership qualifiers that were being applied to
ownership-qualified, substituted type that was *not* a substituted
template type parameter. We now provide a diagnostic in such cases,
and recover by dropping the added qualifiers.
Douglas Gregor [Fri, 17 Jun 2011 22:26:49 +0000 (22:26 +0000)]
Objective-C++ ARC: do not mangle __unsafe_unretained lifetime
qualifiers, so that an __unsafe_unretained-qualified type T in ARC code
will have the same mangling as T in non-ARC code, improving ABI
interoperability. This works now because we infer or require a
lifetime qualifier everywhere one can appear in an external
interface. Another part of <rdar://problem/9595486>.
Douglas Gregor [Fri, 17 Jun 2011 22:11:49 +0000 (22:11 +0000)]
Objective-ARC++: infer template type arguments of
ownership-unqualified retainable object type as __strong. This allows
us to write, e.g.,
std::vector<id>
and we'll infer that the vector's element types have __strong
ownership semantics, which is far nicer than requiring:
std::vector<__strong id>
Note that we allow one to override the ownership qualifier of a
substituted template type parameter, e.g., given
template<typename T>
struct X {
typedef __weak T type;
};
X<id> is treated the same as X<__strong id>. At instantiation type,
the __weak in "__weak T" overrides the (inferred or specified)
__strong on the template argument type, so that we can still provide
metaprogramming transformations.
Richard Trieu [Fri, 17 Jun 2011 20:35:48 +0000 (20:35 +0000)]
Put the new warning from revision 133196 on NULL arithmetic behind the flag -Wnull-arthimetic and set to DefaultIgnore. A few edge cases need to be worked out before this can be set to default.
Douglas Gregor [Fri, 17 Jun 2011 16:37:20 +0000 (16:37 +0000)]
When emitting a compound literal of POD type, continue to emit a
separate aggregate temporary and then memcpy it over to the
destination. This fixes a regression I introduced with r133235, where
the compound literal on the RHS of an assignment makes use of the
structure on the LHS of the assignment.
I'm deeply suspicious of AggExprEmitter::VisitBinAssign()'s
optimization where it emits the RHS of an aggregate assignment
directly into the LHS lvalue without checking whether there is any
aliasing between the LHS/RHS. However, I'm not in a position to
revisit this now.
John McCall [Fri, 17 Jun 2011 06:42:21 +0000 (06:42 +0000)]
Objective-C fast enumeration loop variables are not retained in ARC, but
they should still be officially __strong for the purposes of errors,
block capture, etc. Make a new bit on variables, isARCPseudoStrong(),
and set this for 'self' and these enumeration-loop variables. Change
the code that was looking for the old patterns to look for this bit,
and change IR generation to find this bit and treat the resulting
variable as __unsafe_unretained for the purposes of init/destroy in
the two places it can come up.
projects / clang / log