Lots of tests are using an explicit target triple w/o first checking that the
target is actually available. Add a REQUIRES clause to a bunch of them. This should
hopefully unbreak bots which don't configure w/ all targets enabled.
llvm-svn: 159949
Previously it was possible to get an infinite-loop-on-invalid with a namespace
decl within @interface. Since 'namespace' is normally a safe place to retry
top-level parsing, we just didn't consume the token.
This adds a flag that tracks whether we have temporarily left Objective-C
scope to parse a C-like declaration, and uses that to better recover from
parse problems by stopping at possible method declarations and at @end. To
fix the original problem, we do /not/ stop at 'namespace' when in an
Objective-C @interface or @protocol context (but still do in @implementation).
llvm-svn: 159941
This was a regression introduced during the CallEvent changes; a call to
FunctionDecl::hasBody was also being used to replace the decl found by
lookup with the actual definition. To keep from making this mistake again
(particularly if/when we start inlining Objective-C methods), this commit
adds a "getDefinition()" method to CallEvent, which should do the right
thing under any circumstances.
llvm-svn: 159940
Chris pointed out that while the comparison is certainly problematic
and does not have well-defined behavior, it isn't any worse than some
of the other abuses that we merely warn about and doesn't need to make
the compilation fail.
Revert the release notes change (r159766) now that this is just a new warning.
llvm-svn: 159939
* When substituting a reference to a non-type template parameter pack where the
corresponding argument is a pack expansion, transform into an expression
which contains an unexpanded parameter pack rather than into an expression
which contains a pack expansion. This causes the SubstNonTypeTemplateParmExpr
to be inside the PackExpansionExpr, rather than outside, so the expression
still looks like a pack expansion and can be deduced.
* Teach MarkUsedTemplateParameters that we can deduce a reference to a template
parameter if it's wrapped in a SubstNonTypeTemplateParmExpr (such nodes are
added during alias template substitution).
llvm-svn: 159922
expression, skip over any SubstNonTypeTemplateParmExprs which alias templates
may have inserted before checking for a DeclRefExpr referring to a non-type
template parameter declaration.
llvm-svn: 159909
-ftemplate-depth limit. There are various ways to get an infinite (or merely
huge) stack of substitutions with no intervening instantiations. This is also
consistent with gcc's behavior.
llvm-svn: 159907
multidimensional array of class type. Also, preserve zero-initialization when
evaluating an initializer list for an array, in case the initializers refer to
later elements (which have preceding zero-initialization).
llvm-svn: 159904
which will appear in the vtable as used, not just those ones which were
declared within the class itself. Fixes an issue reported as comment#3 in
PR12763 -- we sometimes assert in codegen if we try to emit a reference to a
function declaration which we've not marked as referenced. This also matches
gcc's observed behavior.
llvm-svn: 159895
- Split pedantic driver flag test into separate test file, and XFAIL on cygwin,mingw32
- Fix bug in tablegen logic where a missing '{' caused errors to be included in -Wpedantic.
llvm-svn: 159892
I suspect FileCheck might match assertion failure, even if clang/test/Misc/warning-flags.c passed the test.
> 0. Program arguments: bin/./clang -### -pedantic -Wpedantic clang/test/Driver/warning-options.cpp
llvm-svn: 159886
This patch introduces some magic in tablegen to create a "Pedantic" diagnostic
group which automagically includes all warnings that are extensions. This
allows a user to suppress specific warnings traditionally under -pedantic used
an ordinary warning flag. This also allows users to use #pragma to silence
specific -pedantic warnings, or promote them to errors, within blocks of text
(just like any other warning).
-Wpedantic is NOT an alias for -pedantic. Instead, it provides another way
to (a) activate -pedantic warnings and (b) disable them. Where they differ
is that -pedantic changes the behavior of the preprocessor slightly, whereas
-Wpedantic does not (it just turns on the warnings).
The magic in the tablegen diagnostic emitter has to do with computing the minimal
set of diagnostic groups and diagnostics that should go into -Wpedantic, as those
diagnostics that already members of groups that themselves are (transitively) members
of -Wpedantic do not need to be included in the Pedantic group directly. I went
back and forth on whether or not to magically generate this group, and the invariant
was that we always wanted extension warnings to be included in -Wpedantic "some how",
but the bookkeeping would be very onerous to manage by hand.
-no-pedantic (and --no-pedantic) is included for completeness, and matches many of the
same kind of flags the compiler already supports. It does what it says: cancels out
-pedantic. One discrepancy is that if one specifies --no-pedantic and -Weverything or
-Wpedantic the pedantic warnings are still enabled (essentially the -W flags win). We
can debate the correct behavior here.
Along the way, this patch nukes some code in TextDiagnosticPrinter.cpp and CXStoredDiagnostic.cpp
that determine whether to include the "-pedantic" flag in the warning output. This is
no longer needed, as all extensions now have a -W flag.
This patch also significantly reduces the number of warnings not under flags from 229
to 158 (all extension warnings). That's a 31% reduction.
llvm-svn: 159875
We use LazyCompoundVals to avoid copying the contents of structs and arrays
around in the store, and when we need to pass a struct around that already
has a LazyCompoundVal we just use the original one. However, it's possible
that the first field of a struct may have a LazyCompoundVal of its own, and
we currently can't distinguish a LazyCompoundVal for the first element of a
struct from a LazyCompoundVal for the entire struct. In this case we should
just drop the optimization and make a new LazyCompoundVal that encompasses
the old one.
PR13264 / <rdar://problem/11802440>
llvm-svn: 159866
This flag sets the 'fp-contract' mode, which controls the formation of fused
floating point operations. Available modes are:
- Fast: Form fused operations anywhere.
- On: Form fused operations where allowed by FP_CONTRACT. This is the default
mode.
- Off: Don't form fused operations (in future this may be relaxed to forming
fused operations where it can be proved that the result won't be
affected).
Currently clang doesn't support the FP_CONTRACT pragma, so the 'On' and 'Off'
modes are equivalent.
llvm-svn: 159794
that the migrator handles) but return their instances as 'id', resulting
in the compiler resolving 'objectForKey:' as the method from NSDictionary.
When checking if we can convert to subscripting syntax, check whether
the receiver is a result of a class method from a hardcoded list of
such classes. In such a case return the specific class as the interface
of the receiver.
llvm-svn: 159788
(apart from NSDictionary/NSArray) that implement objectForKey:/objectAtIndex/etc.
and the subscripting methods as well.
Part of rdar://11734969
llvm-svn: 159783
duplicates attributes on the declaration. Also eliminates a false negative in
ReleasableMutexLock. Fixing this bug required some refactoring.
llvm-svn: 159780
of out-of-line c++ method definition which happens
to be inside an objc class implementation
until I can figure out how to do it. This is to fix
a broken project.
llvm-svn: 159772
actually perform value initialization rather than trying to fake it with a call
to the default constructor. Fixes various bugs related to the previously-missing
zero-initialization in this case.
I've also moved this and the other list initialization 'special case' from
TryConstructorInitialization into TryListInitialization where they belong.
llvm-svn: 159733
Our current inlining support (specifically RegionStore::enterStackFrame)
doesn't know that calls to overloaded operators may be calls to non-static
member functions, and that in these cases the first argument should be
treated as 'this'. This caused incorrect results and sometimes crashes.
The long-term fix will be to rewrite RegionStore::enterStackFrame to use
CallEvent and its subclasses, but for now we can just disable these
problematic calls by classifying them under a new CallEvent,
CXXMemberOperatorCall.
llvm-svn: 159692
By default on OS X 10.8, we don't link with a crt1.o file and the linker
knows to use _main as the entry point. But, when compiling with -pg, we
need to link with the gcrt1.o file, and the linker needs to be told to use
the "start" symbol as the entry point. The -no_new_main linker option does
that last part. <rdar://problem/11491405>
llvm-svn: 159683
values:
- Return integer vectors in integer registers.
- Pass vector arguments in integer registers.
- Set an upper bound for argument alignment. The largest alignment is 8-byte
for O32 and 16-byte for N32/64.
llvm-svn: 159676
if we want to ignore a result, the Dest will be null. Otherwise,
we must copy into it. This means we need to ensure a slot when
loading from a volatile l-value.
With all that in place, fix a bug with chained assignments into
__block variables of aggregate type where we were losing insight into
the actual source of the value during the second assignment.
llvm-svn: 159630
given a cursor pointing to a C++ method call or an ObjC message,
returns non-zero if the method/message is "dynamic", meaning:
For a C++ method: the call is virtual.
For an ObjC message: the receiver is an object instance, not 'super' or a
specific class.
rdar://11779185
llvm-svn: 159627
c-functions declared in implementation should have their
parsing delayed until the end so, they can access forward
declared private methods. // rdar://10387088
llvm-svn: 159626
This adds validation that the
* repro source is only rewrite-includes processed, not fully preprocessed.
* repro script contains macro definitions (-DFOO=BAR).
Based on suggestions/help by Matt Beaumont-Gay.
llvm-svn: 159605
Clang unit tests. It's not clear why we support this mode in builds
where LLVM is available (LLVM itself does not), but at least this makes
us support it correctly.
This also fixes a long-standing bug where we would pass the unit test
param flag to lit in the standalone build even though the standalone
build *never* has the unittests built and ready for testing.
llvm-svn: 159594
In C, enum constants have the type of the enum's underlying integer type,
rather than the type of the enum. (This is not true in C++.) Thus, when a
block's return type is inferred from an enum constant, it is incompatible
with expressions that return the enum type.
In r158899, I told block returns to pretend that enum constants have enum
type, like in C++. Doug Gregor pointed out that this can break existing code.
Now, we don't check the types of return statements until the end of the block.
This lets us go back and add implicit casts in blocks with mixed enum
constants and enum-typed expressions.
<rdar://problem/11662489> (again)
llvm-svn: 159591
template instantiation. I wasn't able to reproduce this down to
anything small enough to put in our test suite, but it's "obviously"
okay to set the invalid bit earlier and precludes a
known-broken-but-not-marked-broken class from being used elsewhere.
llvm-svn: 159584
in the call graph had been inlined but for whatever reason we did not inline some
of its callees.
Also, fix a related traversal bug where we meant to do a BFS of the callgraph but
instead were doing a DFS.
llvm-svn: 159577
the standalone mode. We've changed scoping and sequencing of variables
being defined and that cause this to start to be unset breaking some
cmake users. Thanks to Jordan Rose for the report.
The fix also makes the condition on the preceding line much more
sensible. =D
llvm-svn: 159576
This involved refactoring some common pointer-escapes code onto CallEvent,
then having MallocChecker use those callbacks for whether or not to consider
a pointer's /ownership/ as escaping. This still needs to be pinned down, and
probably we want to make the new argumentsMayEscape() function a little more
discerning (content invalidation vs. ownership/metadata invalidation), but
this is a good improvement.
As a bonus, also remove CallOrObjCMessage from the source completely.
llvm-svn: 159557
This ended allowing quite a bit of cleanup, and some minor changes.
- CallEvent makes it easy to use hasNonZeroCallbackArg more aggressively, which
we check in order to avoid false positives with callbacks that might release
the object.
- In order to support this for functions which consume their arguments, there
are two new ArgEffects: DecRefAndStopTracking and DecRefMsgAndStopTracking.
These act just like StopTracking, except that if the object only had a
return count of +1 it's now considered released instead (so we still get
use-after-free messages).
- On the plus side, we no longer have to special-case
+[NSObject performSelector:withObject:afterDelay:] and friends.
- The use of IdentifierInfos in the method summary cache is now hidden; only
the ObjCInterfaceDecl gets passed around most of the time.
- Since we cache all "simple" summaries and check every function call, there is
no real benefit to having NULL stand in for default summaries anymore.
- Whitespace, unused methods, etc.
Even more simplification to come when we get check::postCall and can unify all
these other post* checks.
llvm-svn: 159555
This is intended to replace CallOrObjCMessage, and is eventually intended to be
used for anything that cares more about /what/ is being called than /how/ it's
being called. For example, inlining destructors should be the same as inlining
blocks, and checking __attribute__((nonnull)) should apply to the allocator
calls generated by operator new.
llvm-svn: 159554
coming from an AST file are registered for serialization.
A static data member instantiation of in a chained PCH could be missed
when serializing decls; the result was that when emitting the visible decls
map of its DeclContext, we would use a DeclID that was not actually emitted,
leading to crashes or hangs.
Fix this by making sure such decls are always registered for serialization.
Also introduce extra sanity checks to make sure we don't register new
declarations or types after we have serialized the types/decls block.
rdar://11728990
llvm-svn: 159550
attached to a declaration in the completion string.
Since extracting comments isn't free, a new code completion option is
introduced.
A new code completion option that enables including brief comments
into CodeCompletionString should be a, err, code completion option.
But because ASTUnit caches global declarations during parsing before
even completion consumer is created, the option is duplicated as a
translation unit option (in both libclang and ASTUnit, like the option
to cache code completion results).
llvm-svn: 159539
For some targets a structure named __va_list_tag is built to help define
the __builtin_va_list type. However, __va_list_tag was not being treated as a
predefined type thus causing problems when serializing the AST. This commit
fixes that oversight by adding the necessary support to treat __va_list_tag
as a predefined type.
llvm-svn: 159508
This behaves like the existing GNU __alignof and C++11 alignof keywords;
most of the patch is simply adding the third token spelling to various places.
llvm-svn: 159494
lit testsuites. This sinks all management of the aggregate lit runs into
the LLVM CMake files, making Clang only responsible for declaring its
own testsuite. In the process we fix numerous "bugs" where the proper
method of invoking lit has changed over time, and the old system
encoded several broken artifacts of this in ABIs and compatibility
tests.
It also switches to 'check-clang' for the canonical name of the test
suite, although 'clang-test' remains as an alias.
The situation when Clang is being built in standalone mode is little
changed. It replicates just enough of the lit setup to cope with the
oddities of being run outside of an LLVM build.
llvm-svn: 159483
In future changes we should:
* use __builtin_trap rather than derefing 'random' volatile pointers.
* avoid dumping temporary files into /tmp when running tests, instead
preferring a location that is properly cleaned up by lit.
Review by Chandler Carruth.
llvm-svn: 159469
is selected. This will allow more flexibility when converting diagnostics to
use template type diffing.
Also updated the internal manual and test cases for correctly keeping the bold
attribute and for tree printing.
llvm-svn: 159463
Update the two function overloads
void TemplateSpecializationType::PrintTemplateArgumentList(raw_ostream &OS,....
to behave like
std::string TemplateSpecializationType::PrintTemplateArgumentList(const TemplateArgument *Args,...
hence making sure that clang consistently adds a space between two '>' at the end of nested template arguments.
llvm-svn: 159412
Corrected type for index of _mm256_mask_i32gather_pd
from 256-bit to 128-bit
Corrected types for src|dst|mask of _mm256_mask_i64gather_ps
from 256-bit to 128-bit
Support the following intrinsics:
_mm_mask_i32gather_epi64, _mm256_mask_i32gather_epi64,
_mm_mask_i64gather_epi64, _mm256_mask_i64gather_epi64,
_mm_mask_i32gather_epi32, _mm256_mask_i32gather_epi32,
_mm_mask_i64gather_epi32, _mm256_mask_i64gather_epi32
llvm-svn: 159403
type traits that assignment to/construction of a lifetime-qualified
object under ARC is *not* trivial. Fixes <rdar://problem/11738725>.
llvm-svn: 159401
Previously:
...the comment said DFS...
...the WorkList being instantiated said BFS...
...and the implementation was actually DFS...
...due to an unintentional change in 2010...
...and everything kept working anyway.
This fixes our std::deque implementation of BFS, but switches back to a
SmallVector-based implementation of DFS.
We should probably still investigate the ramifications of DFS vs. BFS,
especially for large functions (and especially when we hit our block path
limit), since this might completely change our memory use. It can also mask
some bugs and reveal others depending on when we halt analysis. But at least
we will not have this kind of little mistake creep in again.
llvm-svn: 159397
handy. It can be done, but we would have to build a derived-to-base cast
during codegen to compute the correct this pointer.
I will handle covariant returns next.
llvm-svn: 159350
undefined behaviour, and move the diagnostic for '' from an Error into
an ExtWarn in this group. This is important for some users of the preprocessor,
and is necessary for gcc compatibility.
llvm-svn: 159335
the correct this pointer. There is some potential for sharing a bit more
code with canDevirtualizeMemberFunctionCalls, but that can be done in an
independent patch.
llvm-svn: 159326
I've added an extra FileCheck pass for that with an extra "CURRENT" prefix.
I've carefully chosed the CURRENT/CORRECT prefixes so they
a) are self-descriptive
b) have the same length so the mangling between the current and the correct version is obvious
Feel free to ask me to change the prefixes if you know a better alternative.
llvm-svn: 159250
Support the following intrinsics:
_mm_mask_i32gather_pd, _mm256_mask_i32gather_pd, _mm_mask_i64gather_pd
_mm256_mask_i64gather_pd, _mm_mask_i32gather_ps, _mm256_mask_i32gather_ps
_mm_mask_i64gather_ps, _mm256_mask_i64gather_ps
llvm-svn: 159222
comparison between two templated types when they both appear in a diagnostic.
Type elision will remove indentical template arguments, which can be disabled
with -fno-elide-type. Cyan highlighting is applied to the differing types.
For more formatting, -fdiagnostic-show-template-tree will output the template
type as an indented text tree, with differences appearing inline. Template
tree works with or without type elision.
llvm-svn: 159216
to see if we had an underlying final class or method, but we would then
use the cast type to do the call, resulting in a direct call to the wrong
method.
llvm-svn: 159212
literal helper functions. All helper functions (global
and locals) use block_invoke as their prefix. Local literal
helper names are prefixed by their enclosing mangled function
names. Blocks in non-local initializers (e.g. a global variable
or a C++11 field) are prefixed by their mangled variable name.
The descriminator number added to end of the name starts off
with blank (for first block) and _<N> (for the N+2-th block).
llvm-svn: 159206
constexpr function evaluation, and corresponding ASan / valgrind issue in
tests, by storing the corresponding value with the relevant stack frame. This
also prevents re-evaluation of the source of the underlying OpaqueValueExpr,
which makes a major performance difference for certain contrived code (see
testcase update).
llvm-svn: 159189
This works around a quirk in the way that explicit template specializations are
handled in Clang. We generate an implicit declaration from the original
template which the explicit specialization is considered to redeclare. This
trips up the explicit delete logic.
This change only works around that strange representation. At some point it'd
be nice to remove those extra declarations to make the AST more accurately
reflect the C++ semantics.
Review by Doug Gregor.
llvm-svn: 159167
The implicit global allocation functions do not have valid source locations,
but we still want to treat them as being "system header" functions for the
purposes of how they affect program state.
llvm-svn: 159160
resulted in it being reverted. A test for that bug was added in r158950.
Original comment:
If an object (such as a std::string) with an appropriate c_str() member function
is passed to a variadic function in a position where a format string indicates
that c_str()'s return type is desired, provide a note suggesting that the user
may have intended to call the c_str() member.
Factor the non-POD-vararg checking out of DefaultVariadicArgumentPromotion and
move it to SemaChecking in order to facilitate this. Factor the call checking
out of function call checking and block call checking, and extend it to cover
constructor calls too.
Patch by Sam Panzer!
llvm-svn: 159159
This adds support for the tls_model attribute. This allows the user to
choose a TLS model that is better than what LLVM would select by
default. For example, a variable might be declared as:
__thread int x __attribute__((tls_model("initial-exec")));
if it will not be used in a shared library that is dlopen'ed.
This depends on LLVM r159077.
llvm-svn: 159078
attributes in more places where we didn't and catching a lot more issues.
This implements nearly every aspect of C++11 attribute parsing, except for:
- Attributes are permitted on explicit instantiations inside the declarator
(but not preceding the decl-spec)
- Attributes are permitted on friend declarations of functions.
- Multiple instances of the same attribute in an attribute-list (e.g.
[[noreturn, noreturn]], not [[noreturn]] [[noreturn]] which is conforming)
are allowed.
The first two are marked as expected-FIXME in the test file and the latter
is probably a defect and is currently untested.
Thanks to Richard Smith for providing the lion's share of the testcases.
llvm-svn: 159072
We don't handle exceptions yet, so we treat them as sinks. ExprEngine
hardcodes messages that are known to raise Objective-C exceptions like -raise,
but it was only checking for +raise:format: and +raise:format:arguments: on
NSException itself, not subclasses.
<rdar://problem/11724201>
llvm-svn: 159010
properly if there is a join point in the control flow graph that involves
a trylock. Also changes the source locations of some warnings to be
more consistent.
llvm-svn: 159008
Revert "If an object (such as a std::string) with an appropriate c_str() member function"
This reverts commit 7d96f6106bfbd85b1af06f34fdbf2834aad0e47e.
llvm-svn: 158949
This now correctly covers, I believe, all the pointer types:
* 'any' pointers (both function and data normal pointers and ObjC object pointers)
* member pointers (both function and data)
* block pointers
llvm-svn: 158931
In C, enum constants have the type of the enum's underlying integer type,
rather than the type of the enum. (This is not true in C++.) This leads to
odd warnings when returning enum constants directly in blocks with inferred
return types. The easiest way out of this is to pretend that, like C++, enum
constants have enum type when being returned from a block.
<rdar://problem/11662489>
llvm-svn: 158899
Also, don't warn if the used function is __attribute__((const)), in which case
it's not supposed to use global variables anyway.
The inline-in-inline thing is a heuristic, and one that's possibly incorrect
fairly often because the function being inlined could definitely use global
variables. However, even some C standard library functions are written using
other (trivial) static-inline functions in the headers, and we definitely don't
want to be warning on that (or on anything that /uses/ these trivial inline
functions). So we're using "inlined" as a marker for "fairly trivial".
(Note that __attribute__((pure)) does /not/ guarantee safety like ((const),
because ((const)) does not guarantee that global variables are not being used,
and the warning is about globals not being shared across TUs.)
llvm-svn: 158898
is passed to a variadic function in a position where a format string indicates
that c_str()'s return type is desired, provide a note suggesting that the user
may have intended to call the c_str() member.
Factor the non-POD-vararg checking out of DefaultVariadicArgumentPromotion and
move it to SemaChecking in order to facilitate this. Factor the call checking
out of function call checking and block call checking, and extend it to cover
constructor calls too.
Patch by Sam Panzer!
llvm-svn: 158887
Tanya Lattner