David Blaikie [Sat, 10 May 2014 02:44:57 +0000 (02:44 +0000)]
Add FIXME describing the limitation of using column info to disambiguate inlining.
Also tidy up, simplify, and extend the test coverage to demonstrate the
limitations. This test should now fail if the bugs are fixed (&
hopefully whoever ends up in this situation sees the FIXMEs and realizes
that the test needs to be updated to positively test their change that
has fixed some or all of these issues).
I do wonder whether I could demonstrate breakage without a macro here,
but any way I slice it I can't think of a way to get two calls to the
same function on the same line/column in non-macro C++ - implicit
conversions happen at the same location as an explicit function, but
you'd never get an implicit conversion on the result of an explicit call
to the same implicit conversion operator (since the value is already
converted to the desired result)...
Richard Smith [Sat, 10 May 2014 01:17:36 +0000 (01:17 +0000)]
Stop all the Decl classes poking at Redeclarable's data member directly, and make getNextRedeclaration follow the pattern of its friends getPreviousDecl and getMostRecentDecl.
Reid Kleckner [Fri, 9 May 2014 22:46:15 +0000 (22:46 +0000)]
MS ABI: Pass 'sret' as the second parameter of instance methods
Summary:
MSVC always passes 'sret' after 'this', unlike GCC. This required
changing a number of places in Clang that assumed the sret parameter was
always first in LLVM IR.
This fixes win64 MSVC ABI compatibility for methods returning structs.
Nico Weber [Fri, 9 May 2014 22:33:11 +0000 (22:33 +0000)]
Don't leak the CXStoredDiagnostics returned by clang_codeCompleteGetDiagnostic()
r144269 changed clang_disposeDiagnostic() to be a no-op, but didn't update
code completion diagnostics. Let CXCodeCompleteResults store all diagnostics
returned by clang_codeCompleteGetDiagnostic() and then free them up in
clang_disposeCodeCompleteResults().
Code completion diagnostics referred to data stored in CXCodeCompleteResults
before already, so it wasn't possible to refer to the results of
clang_codeCompleteGetDiagnostic() after clang_disposeCodeCompleteResults()
before this change already -- hence this should be a safe, backwards-compatible
change.
David Blaikie [Fri, 9 May 2014 22:02:28 +0000 (22:02 +0000)]
Don't emit -Wnon-virtual-dtor on final classes, since it's not a problem there.
The base class is the culprit/risk here - a sealed/final derived class
with virtual functions and a non-virtual dtor can't accidentally be
polymorphically destroyed (if the base class's dtor is protected - which
also suppresses this warning).
Richard Smith [Fri, 9 May 2014 21:08:59 +0000 (21:08 +0000)]
PR19698, PR19674: enable __has_feature checks for cxx_generic_lambdas and
cxx_decltype_auto, and fix documentation of cxx_generic_lambdas and
cxx_init_captures to specify the right feature-check name.
Objective-C. Reduce false positive warnings with -Wselector by issuing warning
only when named selector is declared in TU and it is not declared in a system
header. rdar://16600230
Aaron Ballman [Fri, 9 May 2014 18:26:23 +0000 (18:26 +0000)]
Add the ability to use logical expressions for capability attributes. This is to allow requirements to be expressed not just in terms of lists, but in terms of logical expressions. Eg)
Nico Weber [Fri, 9 May 2014 18:09:42 +0000 (18:09 +0000)]
Don't leak MacroArgs when using code completion, PR19688.
MacroArgs are owned by TokenLexer, and when a TokenLexer is destroyed, it'll
call its MacroArgs's destroy() method. destroy() only appends the MacroArg to
Preprocessor's MacroArgCache list, and Preprocessor's destructor then calls
deallocate() on all MacroArgs in that list. This method then ends up freeing
the MacroArgs's memory.
In a code completion context, Parser::cutOffParsing() gets called when a code
completion token is hit, which changes the type of the current token to
tok::eof. eof tokens aren't always ConsumeToken()ed, so
Preprocessor::HandleEndOfFile() isn't always called, and that function is
responsible for popping the macro stack.
Due to this, Preprocessor::CurTokenLexer can be non-NULL when
~Preprocessor runs. It's a unique_ptr, so it ended up being destructed after
~Preprocessor completed, and its MacroArgs thus got added to the freelist after
the code freeing things on the freelist had already completed. The fix is to
explicitly call reset() before the freelist processing happens. (See the bug
for more notes.)
Benjamin Kramer [Fri, 9 May 2014 17:08:01 +0000 (17:08 +0000)]
ThreadSafetyAnalysis: Don't crash when trying to analyze objc methods.
The thread safety analysis isn't very useful in ObjC (you can't annotate
ObjC classes or methods) but we can still analyze the actual code and
show violations in usage of C/C++ functions.
Fixes PR19541, which does not use thread safety attributes but crashes
with -Weverything.
James Molloy [Fri, 9 May 2014 16:28:56 +0000 (16:28 +0000)]
Pacify bots again - turns out my checkout was slightly polluted when I was reverting a olista01s change, and this pollution made it upstream during the revert checkin :/ Sorryemacs -nw lib/CodeGen/CodeGenModule.cpp
Ben Langmuir [Thu, 8 May 2014 22:36:02 +0000 (22:36 +0000)]
Switch Wmodule-build to a remark
On reflection, this is better despite the missing command-line handling
bits for remarks. Making this a remark makes it much clearer that
this is purely informational and avoids the negative connotations of a
'warning'.
Ben Langmuir [Thu, 8 May 2014 18:09:29 +0000 (18:09 +0000)]
Remove -Wnon-modular-include
But keep -Wnon-modular-include-in-[framework-]module
This warning is too noisy and doesn't really indicate a problem for most
people. Even though it would only really affect people using
-Weverything, that seems bad so remove it.
Large is CodeModel::Model::Large, not CodeModel::Model::Medium. Thanks to
majnemer for pointing out the typo! Its unclear how to test the mapped value in
the compiler, the tests already cover the driver side.
Diego Novillo [Thu, 8 May 2014 13:49:54 +0000 (13:49 +0000)]
Fix segmentation fault when mixing -Rpass with #line.
Summary:
When using #line directives, FileManager::getFile() will return a nil
entry. This triggers an assert in translateFileLineCol().
This patch handles nil FileEntry instances by emitting a note that the
location could not be translated back to a SourceLocation. I don't
really like this solution, but we are translating presumed locations,
so some information has already been lost.
Karthik Bhat [Thu, 8 May 2014 13:16:20 +0000 (13:16 +0000)]
Fix PR19169 [Crash on invalid attempting to specialize a template method as a template variable].
A template declaration of a template name can be null in case we have a dependent name or a set of function templates.
Hence use dyn_cast_or_null instead of dyn_cast. Also improve the diagnostic emitted in this case.
Nico Weber [Thu, 8 May 2014 04:26:47 +0000 (04:26 +0000)]
Let ASTReader optionally delete its ASTDeserializationListener.
Use this to fix the leak of DeserializedDeclsDumper and DeserializedDeclsChecker
in FrontendAction (found by LSan), PR19560.
The "delete this" bool is necessary because both PCHGenerator and ASTUnit
return the same object from both getDeserializationListener() and
getASTMutationListener(), so ASTReader can't just have a unique_ptr.
It's also not possible to just let FrontendAction (or CompilerInstance) own
these listeners due to lifetime issues (see comments on PR19560).
Finally, ASTDeserializationListener can't easily be refcounted, since several of
the current listeners are allocated on the stack.
Having this bool isn't ideal, but it's a pattern that's used in other places in
the codebase too, and it seems better than leaking.
This addresses an existing FIXME item in the driver. The code model flag was
parsed in the actual tool rather than in the driver. This was problematic since
the value may be invalid. In that case, we would silently treat it as a default
value in non-assert builds, and abort in assert builds. Add a check in the
driver to validate that the value being passed is valid, and if not provide a
proper error message.
Richard Smith [Wed, 7 May 2014 21:53:27 +0000 (21:53 +0000)]
Add an Extension warning for applying unary * to an operand of type 'void*' in
C++. This seems like a pointless (and indeed harmful) restriction to me, so
I've suggested removing it to -core and disabled this diagnostic by default.
James Molloy [Wed, 7 May 2014 17:41:15 +0000 (17:41 +0000)]
When doing int<->ptr coercion for big-endian, calculate the shift amount correctly.
Previously we calculated the shift amount based upon DataLayout::getTypeAllocSizeInBits.
This will only work for legal types - types such as i24 that are created as part of
structs for bitfields will return "32" from that function. Change to using
getTypeSizeInBits.
It turns out that AArch64 didn't run across this problem because it always returned
[1 x i64] as the type for a bitfield, whereas ARM64 returns i64 so goes down this
(better, but wrong) codepath.
David Blaikie [Wed, 7 May 2014 16:56:58 +0000 (16:56 +0000)]
PR19562: Fix another temporary node leak in Clang debug info emission
While constructing ObjC Interface types we might create the declaration
of some normal C++ types, thus adding things to the ReplaceMap. Make
sure we process the ReplaceMap after the ObjC interfaces.
In theory we know at this point, since we're at the end of the TU, that
we won't be upgrading any declarations to definitions, so we could just
construct non-temporary nodes, but that would require extra state in
CGDebugInfo to conditionalize the creation of declaration nodes which
seems annoying/more work than is appropriate.
Nico Weber [Wed, 7 May 2014 16:25:32 +0000 (16:25 +0000)]
Remove "CH_ECK" line (which is ignored) from test after r154191.
r154191 switched to atexit() instead of global destructors, so the intent
was probably to check for _GLOBAL__D_a _not_ being in the output. There already
is a line for _ZN3barD1Ev further up, so just remove the CH_ECK line referring
to that.
The only circumstance in which clang emits _GLOBAL__D_a destructor symbols is
for -fapple-kext, and that is tested by test/CodeGenCXX/cxx-apple-kext.cpp.
James Molloy [Wed, 7 May 2014 14:51:57 +0000 (14:51 +0000)]
Re-satisfy the bots. Due to a de-sync between my clang and LLVM trees, I didn't notice that the generated BE alignment register is now unnamed and is not called '%align_be'.
Oliver Stannard [Wed, 7 May 2014 10:39:12 +0000 (10:39 +0000)]
ARM: Fix assertion caused by passing bitfield struct using ABIArgInfo::getExpandWithPadding
In cases where a struct must, according to the AAPCS, not be split between
general purpose and floating point registers, we use
ABIArgInfo::getExpandWithPadding to add the padding arguments. However,
ExpandWithPadding does not work if the struct contains bitfields, so we
instead must use ABIArgInfo::getDirect.
Drop libgcc and default to libc++ for NetBSD/ARM on 6.99.40+.
Don't bother with keeping the old support for x86_64 in 6.99.23+, just
use a single range. Update test cases for the always-on --eh-frame-hdr.
Richard Smith [Wed, 7 May 2014 02:25:43 +0000 (02:25 +0000)]
If an instantiation of a template is required to be a complete type, check
whether the definition of the template is visible rather than checking whether
the instantiated definition happens to be in an imported module.