Richard Smith [Sun, 10 Aug 2014 02:20:15 +0000 (02:20 +0000)]
[modules] When considering merging a newly-declared typedef into an imported
one, perform the import if the types match even if the imported declaration is
hidden. Otherwise, NamedDecl::declarationReplaces will drop one of the name
lookup entries, making the typedef effectively inaccessible from one of the
modules that declared it.
Yaron Keren [Sat, 9 Aug 2014 21:24:04 +0000 (21:24 +0000)]
Make this test unsupported when there is no real 'env' from a real shell.
While the test was already requiring shell-preserves-root (such as the
internal shell), it wasn't aware that the MSYS 1.0 env command knows how
to expand root by itself!
From cmd.exe try:
env SDKROOT=/ cmd //c echo %SDKROOT%
And get:
C:/MINGW/MSYS/1.0
To be certain we have a good 'env' program the test now requires a shell.
Suprisingly the normalize_separators() was no-op when LLVM_ON_WIN32.
Its replacement native() does change path separators into \ as expected,
breaking these tests.
I had fixed the tests by #ifndef LLVM_ON_WIN32 on the native call,
to match the previous behaviour.
If this logic is not used on Windows host, it might be completely
deleted as there should not be windows path seperators on Linux hosts.
I can't test on Linux but if someone can run tests on Linux after
commenting out the line
llvm::sys::path::native(NormalizedPath);
and the tests pass, the whole if (LangOpts.MSVCCompat) could be deleted.
Justin Bogner [Sat, 9 Aug 2014 03:55:09 +0000 (03:55 +0000)]
Revert "Add tests for coverage mapping generation."
I reverted one of the added tests from r215261 in r215274, since it
was failing on quite a few bots. It looks like this wasn't sufficient,
as we're still getting failures on windows, like the following:
Richard Smith [Sat, 9 Aug 2014 01:24:07 +0000 (01:24 +0000)]
[modules] Weaken an out-of-date assertion: an #undef can have no location if we
imported it from a module when performing finalization before writing out
an AST file.
Ben Langmuir [Sat, 9 Aug 2014 00:57:23 +0000 (00:57 +0000)]
Refactor the module map file used for uniquing a module name out of
class Module. It's almost always going to be the same as
getContainingModule() for top-level modules, so just add a map to cover
the remaining cases. This lets us do less bookkeeping to keep the
ModuleMap fields up to date.
Alex Lorenz [Fri, 8 Aug 2014 23:49:58 +0000 (23:49 +0000)]
Add tests for coverage mapping generation.
This patch adds the tests for the coverage mapping generation.
Most of the tests check the mapping regions produced by
the generator, and one checks the llvm IR.
David Blaikie [Fri, 8 Aug 2014 23:36:37 +0000 (23:36 +0000)]
Simplify ownership of ExplodedGraph in the CoreEngine by removing unique_ptr indirection.
Summary: I was going to fix the use of raw pointer ownership in "takeGraph" when I realized that function was unused and the whole ExplodedGraph could just be owned by value without the std::unique_ptr indirection at all.
Alexey Samsonov [Fri, 8 Aug 2014 22:47:17 +0000 (22:47 +0000)]
Add -link-cxx-sanitizer driver flag.
Summary:
This flag can be used to force linking of CXX-specific parts
of sanitizer runtimes into the final executable. It gives more precise
control than --driver-mode=g++ and comes handy when user links several
object files with sanitized C++ code into an executable, but wants
to provide libstdc++ himself, instead of relying on Clang dirver's
behavior.
Richard Trieu [Fri, 8 Aug 2014 22:41:43 +0000 (22:41 +0000)]
Extend tautological pointer compare and pointer to bool conversion warnings to
macro arguments.
Previously, these warnings skipped any code in a macro expansion. Preform an
additional check and warn when the expression and context locations are both
in the macro argument.
The most obvious case not caught is passing a pointer directly to a macro,
i.e 'assert(&array)' but 'assert(&array && "valid array")' is caught. This is
because macro arguments are not typed and the conversion happens inside the
macro.
Objective-C [qoi]. Issue warning and fixit if property-dot syntax
use mis-cased property name (which is currently accepted silently
due to the way property setters are named). rdar://17911746
Bob Wilson [Fri, 8 Aug 2014 21:45:53 +0000 (21:45 +0000)]
Change __ENVIRONMENT_MAC_OS_X_VERSION_MIN_REQUIRED__ for versions > 10.9.
The previous encoding only allowed a single digit for the minor version
number. This changes it to use 2 digits for both the minor version and the
revision number.
Daniel Sanders [Fri, 8 Aug 2014 18:39:01 +0000 (18:39 +0000)]
Partially revert r215204 - [mips] Add -mabicalls/-mno-abicalls to the driver
It wasn't actually a bug that -mabicalls/-mno-abicalls wasn't being passed to
GAS. The only reason we pass it to the integrated assembler is because it shares
the same framework with CodeGen.
David Blaikie [Fri, 8 Aug 2014 17:10:14 +0000 (17:10 +0000)]
DebugInfo: Blocks: Do not depend on LLVM argument numbering when choosing the debug info argument numbering.
Due to the possible presence of return-by-out parameters, using the LLVM
argument number count when numbering debug info arguments can end up
off-by-one. This could produce two arguments with the same number, which
would in turn cause LLVM to emit only one of those arguments (whichever
it found last) or assert (r215157).
David Blaikie [Fri, 8 Aug 2014 16:06:15 +0000 (16:06 +0000)]
CompilationDatabase: Sure-up ownership of compilation databases using std::unique_ptr
Diving into the memory leaks fixed by r213851 there was one case of a
memory leak of a CompilationDatabase due to not properly taking
ownership of the result of "CompilationDatabase::autoDetectFromSource".
Given that both implementations and callers have been using unique_ptr
to own CompilationDatabase objects - make this explicit in the API to
reduce the risk of further leaks.
Daniel Sanders [Fri, 8 Aug 2014 15:47:17 +0000 (15:47 +0000)]
[mips] Invert the abicalls feature bit to be noabicalls so that it's possible for -mno-abicalls to take effect.
Also added the testcase that should have been in r215194.
This behaviour has surprised me a few times now. The problem is that the
generated MipsSubtarget::ParseSubtargetFeatures() contains code like this:
if ((Bits & Mips::FeatureABICalls) != 0) IsABICalls = true;
so '-abicalls' means 'leave it at the default' and '+abicalls' means 'set it to
true'. In this case, (and the similar -modd-spreg case) I'd like the code to be
Manuel Klimek [Fri, 8 Aug 2014 07:37:13 +0000 (07:37 +0000)]
Fix branch reachabiliy annotation for temp dtor branches.
As we only create temp dtor decision branches when a temp dtor needs to
be run (as opposed to for each logical branch in the original
expression), we must include the information about all previous logical
branches when we annotate the temp dtor decision branch.
David Majnemer [Fri, 8 Aug 2014 07:21:18 +0000 (07:21 +0000)]
Parser: Array decls with static but without array size are illformed
Array declarators involving the static keyword take on two forms:
D[ static type-qualifier-listopt assignment-expression ]
D[ type-qualifier-list static assignment-expression ]
Raise a diagnostic if the assignment-expression is missing.
David Majnemer [Fri, 8 Aug 2014 00:10:39 +0000 (00:10 +0000)]
MS ABI: Don't force bases to have an inheritance model
Previously, assigning an inheritance model to a derived class would
trigger further assiginments to the various bases of the class. This
was done to fix a bug where we couldn't handle an implicit
base-to-derived conversion for pointers-to-members when the conversion
was ambiguous at an earlier point.
However, this is not how the MS scheme works. Instead, assign
inheritance models to *just* the class which owns to declaration we
ended up referencing.
N.B. This result is surprising in many ways. It means that it is
possible for a base to have a "larger" inheritance model than it's
derived classes. It also means that bases in the conversion path do not
get assigned a model.
struct A { void f(); void f(int); };
struct B : A {};
struct C : B {};
void f() { void (C::*x)() = &A::f; }
We can only begin to assign an inheritance model *after* we've seen the
address-of but *before* we've done the implicit conversion the more
derived pointer-to-member type. After that point, both 'A' and 'C' will
have an inheritance model but 'B' will not. Surprising, right?
Reid Kleckner [Thu, 7 Aug 2014 21:29:25 +0000 (21:29 +0000)]
Debug info: Use the vbtable offset for virtual bases in the MS ABI
There are no vtable offset offsets in the MS ABI, but vbtable offsets
are analogous. There are no consumers of this information yet, but at
least we don't crash now.
Objective-C arc. Switch the Objective-C dictionary literal in ARC mode
to use non-retain/autorelease API variants of ObjC objects. wip.
rdar://17554063
Benjamin Kramer [Thu, 7 Aug 2014 20:51:16 +0000 (20:51 +0000)]
Flip the order the preprocessor and frontendaction are informed of the end of a file.
This allows using EndOfMainFile from a PPCallback to access data from the
action. The pattern of PPCallback referencing an action is common in clang-tidy.
Richard Smith [Thu, 7 Aug 2014 18:53:08 +0000 (18:53 +0000)]
[modules] When emitting an update record containing the body of a destructor,
also emit the updated 'operator delete' looked up for that destructor. Switch
from UpdateDecl to an actual update record when this happens due to implicitly
defining a special member function and unify this code path and the one for
instantiating a function definition.
Justin Bogner [Thu, 7 Aug 2014 18:45:21 +0000 (18:45 +0000)]
test/Modules: Use FileCheck's -allow-empty instead of "count 0"
Piping stderr into "count 0" in tests doesn't work - things like guard
malloc write to stderr and mess up the count. This comes up all the
time, so I've added a feature to FileCheck to fix it this time.
Fixes test failures caused by r215046 under guard malloc.
Manuel Klimek [Thu, 7 Aug 2014 16:05:51 +0000 (16:05 +0000)]
Model temporary destructors from logical operators with known values.
If the truth value of a LHS is known, we can build the knowledge whether
a temporary destructor is executed or not into the CFG. This is needed
by the return type analysis.
Manuel Klimek [Thu, 7 Aug 2014 10:42:17 +0000 (10:42 +0000)]
Re-applying r214962.
Changes to the original patch:
- model the CFG for temporary destructors in conditional operators so that
the destructors of the true and false branch are always exclusive. This
is necessary because we must not have impossible paths for the path
based analysis to work.
- add multiple regression tests with ternary operators
Original description:
Fix modelling of non-lifetime-extended temporary destructors in the
analyzer.
Changes to the CFG:
When creating the CFG for temporary destructors, we create a structure
that mirrors the branch structure of the conditionally executed
temporary constructors in a full expression.
The branches we create use a CXXBindTemporaryExpr as terminator which
corresponds to the temporary constructor which must have been executed
to enter the destruction branch.
2. Changes to the Analyzer:
When we visit a CXXBindTemporaryExpr we mark the CXXBindTemporaryExpr as
executed in the state; when we reach a branch that contains the
corresponding CXXBindTemporaryExpr as terminator, we branch out
depending on whether the corresponding CXXBindTemporaryExpr was marked
as executed.
Richard Trieu [Thu, 7 Aug 2014 02:09:05 +0000 (02:09 +0000)]
Update the context location of the condition of a conditional operator to the
question mark instead of the context of the conditional operator. The
condition does not need the context of the conditional operator at all.
Richard Smith [Thu, 7 Aug 2014 00:24:21 +0000 (00:24 +0000)]
Use -Rblah, not -Wblah, to control remark diagnostics. This was always the
intent when we added remark support, but was never implemented in the general
case, because the first -R flags didn't need it. (-Rpass= had special handling
to accomodate its argument.)
-Rno-foo, -Reverything, and -Rno-everything can be used to turn off a remark,
or to turn on or off all remarks. Per discussion on cfe-commits, -Weverything
does not affect remarks, and -Reverything does not affect warnings or errors.
The only "real" -R flag we have right now is -Rmodule-build; that flag is
effectively renamed from -Wmodule-build to -Rmodule-build by this change.
-Wpass and -Wno-pass (and their friends) are also renamed to -Rpass and
-Rno-pass by this change; it's not completely clear whether we intended to have
a -Rpass (with no =pattern), but that is unchanged by this commit, other than
the flag name. The default pattern is effectively one which matches no passes.
In future, we may want to make the default pattern be .*, so that -Reverything
works for -Rpass properly.
Daniel Jasper [Wed, 6 Aug 2014 14:15:41 +0000 (14:15 +0000)]
clang-format: Correct SBPO_Always-behavior after function-like keywords
Before:
auto f (int x) -> decltype(x) { return sizeof(x); }
int g () noexcept(someCall ());
static_assert(sizeof(char) == 1, "Your compiler is broken");
After:
auto f (int x) -> decltype (x) { return sizeof (x); }
int g () noexcept (someCall ());
static_assert (sizeof (char) == 1, "Your compiler is broken");
This fixes llvm.org/PR20559.
Patch by Roman Kashitsyn, thank you!
Manuel Klimek [Wed, 6 Aug 2014 12:45:51 +0000 (12:45 +0000)]
Fix modelling of non-lifetime-extended temporary destructors in the analyzer.
1. Changes to the CFG:
When creating the CFG for temporary destructors, we create a structure
that mirrors the branch structure of the conditionally executed
temporary constructors in a full expression.
The branches we create use a CXXBindTemporaryExpr as terminator which
corresponds to the temporary constructor which must have been executed
to enter the destruction branch.
2. Changes to the Analyzer:
When we visit a CXXBindTemporaryExpr we mark the CXXBindTemporaryExpr as
executed in the state; when we reach a branch that contains the
corresponding CXXBindTemporaryExpr as terminator, we branch out
depending on whether the corresponding CXXBindTemporaryExpr was marked
as executed.
Simon Atanasyan [Wed, 6 Aug 2014 05:44:47 +0000 (05:44 +0000)]
[Driver] Move information about toolchain specific include directories
from the common driver code to the corresponding `MultilibSet` declarations.
Now the `MultilibSet` can hold an optional callback function which is
responsible to return a set of include directories specific for the toolchain.
That allows to remove MIPS toolchain specific directories from
`Linux::AddClangSystemIncludeArgs` method and simplify adding new directories
in the future.
David Majnemer [Wed, 6 Aug 2014 03:12:47 +0000 (03:12 +0000)]
MS ABI: Mangle lambdas which are given the same mangling number
It is possible for lambdas to get the same mangling number because they
may exist in different mangling contexts. To handle this correctly,
mangle the context into the name as well.
David Majnemer [Tue, 5 Aug 2014 22:43:45 +0000 (22:43 +0000)]
MS ABI: Mangle empty type parameter packs compatibly
The MS mangling scheme apparently has separate manglings for type and
non-type parameter packs when they are empty. Match template arguments
with parameters during mangling; check the parameter to see if it was
destined to hold type-ish things or nontype-ish things.
Introduce f[no-]max-unknown-pointer-align=[number] option
to instruct the code generator to not enforce a higher alignment
than the given number (of bytes) when accessing memory via an opaque
pointer or reference. Patch reviewed by John McCall (with post-commit
review pending). rdar://16254558
Arthur Marble [Tue, 5 Aug 2014 18:21:20 +0000 (18:21 +0000)]
Added f and m flags to be ignored. These will not display a warning. The revision
for this patch is here: http://reviews.llvm.org/D4570. This will help with the
rebuild of Debian with clang. Here is a link to the errors that Debian is
experiencing: http://clang.debian.net/status.php?version=3.4.2&key=UNKNOWN_ARG
Arthur Marble [Tue, 5 Aug 2014 18:13:48 +0000 (18:13 +0000)]
Added flags that should be ignored for compatibility. These flags will display
a warning. Revision for this patch is here: http://reviews.llvm.org/D4565. This
patch will help with the rebuild of Debian with clang and many other projects
that wish to use clang. Here is a link to the errors that Debian is experiencing:
http://clang.debian.net/status.php?version=3.4.2&key=UNKNOWN_ARG
Chad Rosier [Tue, 5 Aug 2014 17:58:54 +0000 (17:58 +0000)]
[PR19983] SBPO_Always not covering all the cases.
Patch by "Roman Kashitsyn" <romankashicin@gmail.com>.
Phabricator revision: http://reviews.llvm.org/D4788
Adam Nemet [Tue, 5 Aug 2014 17:28:23 +0000 (17:28 +0000)]
[AVX512] Add intrinsic for valignd/q
Note that similar to palingr, we could further optimize these to emit
shufflevector when the shift count is <=64. This however does not
change the overall design that unlike palignr we would still need the LLVM
intrinsic corresponding to this intruction to handle the >64 cases. (palignr
uses the psrldq intrinsic in this case.)