Lang Hames [Fri, 6 Jul 2012 00:59:19 +0000 (00:59 +0000)]
Add -ffp-contract = { fast | on | off } command line option support.
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.
Dmitri Gribenko [Fri, 6 Jul 2012 00:28:32 +0000 (00:28 +0000)]
Implement AST classes for comments, a real parser for Doxygen comments and a
very simple semantic analysis that just builds the AST; minor changes for lexer
to pick up source locations I didn't think about before.
Comments AST is modelled along the ideas of HTML AST: block and inline content.
* Block content is a paragraph or a command that has a paragraph as an argument
or verbatim command.
* Inline content is placed within some block. Inline content includes plain
text, inline commands and HTML as tag soup.
[objcmt] Check for classes that accept 'objectForKey:' (or the other selectors
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.
objective-c: don't involve properties when checking
under -Wsuper-class-method-mismatch for method
mismatches in current and suprt class.
// rdar://11793793
[objcmt] Allow migrating to subscripting syntax for other classes
(apart from NSDictionary/NSArray) that implement objectForKey:/objectAtIndex/etc.
and the subscripting methods as well.
Thread-safety analysis: eliminate false positives in case where the definition
duplicates attributes on the declaration. Also eliminates a false negative in
ReleasableMutexLock. Fixing this bug required some refactoring.
objective-c++ parsing. Turn off delayed parsing
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.
Jordan Rose [Thu, 5 Jul 2012 18:48:30 +0000 (18:48 +0000)]
Release notes for r158230 "Disallow using ObjC literals in direct comparisons"
This may turn out to be a controversial change, due to string literals being
uniqued at link time, but Apple's docs only say "The compiler makes such
object constants unique on a per-module basis..."[1] without actually saying
what a "module" is. (It's not a clang module.) Furthermore, this uniqueness
guarantee often can't be guaranteed once the string has been passed through
framework code.
If this does turn out very controversial, we could downgrade this to a
DefaultError warning for strings, and leave it as a true Error for other
kinds of literals.
Manuel Klimek [Thu, 5 Jul 2012 18:13:01 +0000 (18:13 +0000)]
Adapts the FrontendAction convenience functions so that it can be
used with classes that generate ASTConsumers; this allows decoupling
the ASTConsumer generation from the Frontend library (like, for example,
the MatchFinder in the upcoming ASTMatcher patch).
Richard Smith [Thu, 5 Jul 2012 08:39:21 +0000 (08:39 +0000)]
PR13273: When performing list-initialization with an empty initializer list,
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.
Benjamin Kramer [Wed, 4 Jul 2012 17:04:04 +0000 (17:04 +0000)]
Drop the ASTContext.h include from Stmt.h and fix up transitive users.
This required moving the ctors for IntegerLiteral and FloatingLiteral out of
line which shouldn't change anything as they are usually called through Create
methods that are already out of line.
ASTContext::Deallocate has been a nop for a long time, drop it from ASTVector
and make it independent from ASTContext.h
Pass the StorageAllocator directly to AccessedEntity so it doesn't need to
have a definition of ASTContext around.
David Chisnall [Wed, 4 Jul 2012 11:52:24 +0000 (11:52 +0000)]
Hoist the logic for selecting the Objective-C dispatch method into the runtime
class, from the target. No functionality change, just less duplicated logic.
Jordan Rose [Tue, 3 Jul 2012 22:55:57 +0000 (22:55 +0000)]
[analyzer] For now, don't inline non-static member overloaded operators.
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.
Bob Wilson [Tue, 3 Jul 2012 20:42:10 +0000 (20:42 +0000)]
When using -pg targeting OS X 10.8, pass -no_new_main to the linker.
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>
Make the following changes in the way Mips handles vector arguments and return
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.
Remove a redundant assignment to the FDecl variable from *inside* it's
initializer.
I really feel like Clang should warn about this, but I can't describe
a good reason. GCC will warn on this in some cases under
-Wsequence-point, but it actually seems like a false positive for that
warning....
John McCall [Mon, 2 Jul 2012 23:58:38 +0000 (23:58 +0000)]
Significantly simplify CGExprAgg's logic about ignored results:
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.
[libclang] Introduce clang_Cursor_isDynamicCall which,
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.
objective-c: just as we have done for method definitions,
c-functions declared in implementation should have their
parsing delayed until the end so, they can access forward
declared private methods. // rdar://10387088
Richard Smith [Mon, 2 Jul 2012 23:23:04 +0000 (23:23 +0000)]
-Wuninitialized: assume that an __attribute__((returns_twice)) function might
initialize any variable. This is extremely conservative, but is sufficient for
now.
James Dennett [Mon, 2 Jul 2012 23:17:34 +0000 (23:17 +0000)]
Documentation cleanup:
* Primarily, added \brief to most of include/clang/Basic, instead of prefixing
the comments with "DeclaredName - ";
* Made some brief summaries significantly briefer;
* Fixed up some erroneous uses of \see and \arg;
* Fixed up some extraneous backslashes in \code...\endcode blocks;
* Fixed up some typos/spelling errors.
David Blaikie [Mon, 2 Jul 2012 22:07:20 +0000 (22:07 +0000)]
Improve crash reporting test coverage.
This adds validation that the
* repro source is only rewrite-includes processed, not fully preprocessed.
* repro script contains macro definitions (-DFOO=BAR).
Jordan Rose [Mon, 2 Jul 2012 21:41:53 +0000 (21:41 +0000)]
Revert "Remove unused member (& consequently unused parameter) in SA's Call code."
...and instead add an accessor. We're not using this today, but it's something
that should probably stay in the source for potential clients, and it doesn't
cost a lot. (ObjCPropertyAccess is only created on the stack, and right now
there's only ever one alive at a time.)
We support a mode where the Clang regression tests are run but not the
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.
David Blaikie [Mon, 2 Jul 2012 21:28:00 +0000 (21:28 +0000)]
Include -D files in crash report repro scripts. (PR13255)
Now that we're only using -frewrite-includes rather than full preprocessing
when producing repro source files, we should also include command line macro
definitions in the repro script.
I don't have a test case for this because I'm not sure if/how I can open the
crash report file when the name is only known by scraping the crash report
output. Suggestions welcome if anyone thinks it'd be helpful.
Jordan Rose [Mon, 2 Jul 2012 21:19:23 +0000 (21:19 +0000)]
In blocks, only pretend that enum constants have enum type if necessary.
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.
Douglas Gregor [Mon, 2 Jul 2012 21:00:41 +0000 (21:00 +0000)]
Be more eager about setting the 'Invalid' bit on an invalid class
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.
Bail out the LiveVariables analysis when the CFG is very large, as
we are encountering some scalability issues with memory usage. The
appropriate long term fix is to make the analysis more scalable, but
this will at least prevent the analyzer swapping when
analyzing very large functions.
Fix subtle bug in AnalysisConsumer where we would not analyze functions whose parent
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.
Switch from using a lit.cfg-generator variable to the real variable in
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
Jordan Rose [Mon, 2 Jul 2012 19:28:04 +0000 (19:28 +0000)]
[analyzer] Finish replacing ObjCMessage with ObjCMethodDecl and friends.
The preObjCMessage and postObjCMessage callbacks now take an ObjCMethodCall
argument, which can represent an explicit message send (ObjCMessageSend) or an
implicit message generated by a property access (ObjCPropertyAccess).
Jordan Rose [Mon, 2 Jul 2012 19:27:56 +0000 (19:27 +0000)]
[analyzer] Begin replacing ObjCMessage with ObjCMethodCall and friends.
Previously, the CallEvent subclass ObjCMessageInvocation was just a wrapper
around the existing ObjCMessage abstraction (over message sends and property
accesses). Now, we have abstract CallEvent ObjCMethodCall with subclasses
ObjCMessageSend and ObjCPropertyAccess.
In addition to removing yet another wrapper object, this should make it easy
to add a ObjCSubscriptAccess call event soon.
Jordan Rose [Mon, 2 Jul 2012 19:27:51 +0000 (19:27 +0000)]
[analyzer] Move the last bits of CallOrObjCMessage over to CallEvent.
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.
Jordan Rose [Mon, 2 Jul 2012 19:27:46 +0000 (19:27 +0000)]
[analyzer] Convert CallAndMessageChecker and ObjCSelfInitChecker to CallEvent.
Both of these got uglier rather than cleaner because we don't have preCall and
postCall yet; properly wrapping a CallExpr in a CallEvent requires doing a bit
of deconstruction on the callee. Even when we have preCall and postCall we may
want to expose the current CallEvent to pre/postStmt<CallExpr>.
Jordan Rose [Mon, 2 Jul 2012 19:27:43 +0000 (19:27 +0000)]
[analyzer] Convert RetainCountChecker to use CallEvent as much as possible.
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.
Jordan Rose [Mon, 2 Jul 2012 19:27:35 +0000 (19:27 +0000)]
[analyzer] Add a new abstraction over all types of calls: CallEvent
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.
[PCH] Make sure that all newly introduced visible decls in a DeclContext
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.
Dmitri Gribenko [Mon, 2 Jul 2012 17:35:10 +0000 (17:35 +0000)]
Add a new libclang completion API to get brief documentation comment that is
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).
James Dennett [Mon, 2 Jul 2012 07:01:42 +0000 (07:01 +0000)]
Documentation cleanup: reformatting/fixing up file comments so that they have
\file and \brief markup and appear in Doxygen's summaries (and eventually at
http://clang.llvm.org/doxygen/files.html). Fixed up another couple of minor
glitches in the docs at the same time.