Rafael Espindola [Tue, 26 Jun 2012 17:45:31 +0000 (17:45 +0000)]
During codegen of a virtual call we would extract any casts in the expression
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.
block literal irgen: several improvements on naming block
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).
Richard Smith [Tue, 26 Jun 2012 08:12:11 +0000 (08:12 +0000)]
Fix lifetime issue for backing APValue of OpaqueValueExpr in recursive
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).
David Blaikie [Mon, 25 Jun 2012 21:55:30 +0000 (21:55 +0000)]
PR12937: Explicitly deleting an explicit template specialization.
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.
Jordan Rose [Mon, 25 Jun 2012 20:48:28 +0000 (20:48 +0000)]
[analyzer] Be careful about implicitly-declared operator new/delete. (PR13090)
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.
Richard Smith [Mon, 25 Jun 2012 20:30:08 +0000 (20:30 +0000)]
Unrevert r158887, reverted in r158949, along with a fix for the bug which
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.
Richard Smith [Sun, 24 Jun 2012 23:56:26 +0000 (23:56 +0000)]
Add testing for CommentHandler, and fix a bug where trailing comments in #else
and #endif in non-skipped blocks were not passed to the CommentHandler. Patch
by Andy Gibbs!
Hans Wennborg [Sat, 23 Jun 2012 11:51:46 +0000 (11:51 +0000)]
Support the tls_model attribute (PR9788)
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.
Sean Hunt [Sat, 23 Jun 2012 05:07:58 +0000 (05:07 +0000)]
Clean up a large number of C++11 attribute parse issues, including parsing
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.
Jordan Rose [Fri, 22 Jun 2012 17:15:58 +0000 (17:15 +0000)]
[analyzer] Check for +raise:format: on subclasses of NSException as well.
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.
Thread safety analysis: fixes a bug in which locksets are not handled
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.
James Dennett [Fri, 22 Jun 2012 08:10:18 +0000 (08:10 +0000)]
Documentation cleanup:
* Primarily fixed \param commands with names not matching any actual
parameters of the documented functions. In many cases this consists
just of fixing up the parameter name in the \param to match the code,
in some it means deleting obsolete documentation and occasionally it
means documenting the parameter that has replaced the older one that
was documented, which sometimes means some simple reverse-engineering
of the docs from the implementation;
* Fixed \param ParamName [out] to the correct format with [out] before
the parameter name;
* Fixed some \brief summaries.
James Dennett [Fri, 22 Jun 2012 05:59:27 +0000 (05:59 +0000)]
Documentation cleanup:
* Added \file documentation for PPCallbacks.h;
* Added/formated \brief summaries;
* Deleted documentation for parameters that no longer exist;
* Used \param more systematically for documentation of parameters;
* Escaped # characters in Doxygen comments.
James Dennett [Fri, 22 Jun 2012 05:54:32 +0000 (05:54 +0000)]
Documentation cleanup:
* Add \file documentation;
* Add \verbatim...\endverbatim markup as needed;
* Add \brief summaries;
* Escaped "::" in Doxygen comments when preceded by space, to avoid a
Doxygen warning where Doxygen takes this as an explicit link request;
* Add \code...\endcode markup to code examples;
* Fix a grammatical glitch in "is this declarator is a".
James Dennett [Fri, 22 Jun 2012 05:37:13 +0000 (05:37 +0000)]
Documentation cleanup:
* Use \p param for a parameter reference, not the (erroneous) form \arg param;
* Escape # characters in Doxygen comments as needed.
James Dennett [Fri, 22 Jun 2012 05:32:40 +0000 (05:32 +0000)]
Documentation cleanup: Fixed the format of the file header (that I broke in
a recent commit), and eliminated a Doxygen error by changing a comment inside
a function to not be a Doxygen comment.
James Dennett [Fri, 22 Jun 2012 05:20:59 +0000 (05:20 +0000)]
Documentation cleanup:
* Escaped # and < characters in Doxygen comments as needed;
* Fixed up some \brief summaries;
* Marked up some parameter references with \p;
* Added \code...\endcode around code examples;
* Used \returns a little more.
James Dennett [Fri, 22 Jun 2012 05:06:35 +0000 (05:06 +0000)]
Documentation cleanup:
* Made \brief documentation be brief, mostly by adding a blank line to make
the rest of the text be part of the detailed description only;
* Removed "FunctionOrClassName - " and other redundant text from the start
of Doxygen comments.
Dmitri Gribenko [Thu, 21 Jun 2012 22:04:37 +0000 (22:04 +0000)]
Handle include directive with comments. It turns out that in this case comments are not coming in source order. Instead of trying to std::sort() comments (which can be costly), just remove comments that are not in order.
objective-c: If an ivar is (1) the first ivar in a root class and (2) named `isa`,
then it should get the same warnings that id->isa gets. // rdar://11702488
Anna Zaks [Thu, 21 Jun 2012 20:36:11 +0000 (20:36 +0000)]
[analyzer] Do not step into statements while collecting function decls.
CallGraph's recursive visitor only needs to collect declarations; their
bodies will be processed later on. RecursiveASTVisitor will recurse on
the bodies if the definition is provided along with declaration.
Optimize, by not recursing on any of the statements.
David Blaikie [Thu, 21 Jun 2012 18:51:10 +0000 (18:51 +0000)]
PR13165: False positive when initializing member data pointers with NULL.
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
John McCall [Thu, 21 Jun 2012 17:46:38 +0000 (17:46 +0000)]
Adjust this code so that it strictly honors
TargetSimulatroVersionFromDefines if present; this also makes
it easier to chain things correctly. Noted by an internal
review.
Chandler Carruth [Thu, 21 Jun 2012 09:51:42 +0000 (09:51 +0000)]
Clang side of a refactoring of the CMake unit test build strategy.
The fundamental change is to put a CMakeLists.txt file in the unittest
directory, with a single test binary produced from it. This has several
advantages.
Among other fundamental advantages, we start to get the checking logic
for when a file is missing from the CMake build, and this caught one
missing file already! More fun details in the LLVM commit corresponding
to this one.
Note that the LLVM commit and this one most both be applied, or neither.
Sorry for any skew issues.
Alexey Samsonov [Thu, 21 Jun 2012 08:22:39 +0000 (08:22 +0000)]
Improve support for -g options accepted by Clang:
1. Accept flags -g[0-3], -ggdb[0-3], -gdwarf-[2-4] and collapse them to simple -g (except -g0/-ggdb0).
2. Produce driver error on unsupported formats (-gcoff, -gstabs, -gvms) and options (-gtoggle).
3. Recognize and ignore flags -g[no-]strict-dwarf, -g[no-]record-gcc-switches.
Jordan Rose [Thu, 21 Jun 2012 05:54:55 +0000 (05:54 +0000)]
Pretend that enum constants have enum type when inferring a block return type.
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.
Jordan Rose [Thu, 21 Jun 2012 05:54:50 +0000 (05:54 +0000)]
Don't warn for -Wstatic-in-inline if the used function is also inline.
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.)
Chandler Carruth [Thu, 21 Jun 2012 02:04:39 +0000 (02:04 +0000)]
Simplify the Clang unittest function in the CMake build, and make it
match the LLVM implemenation. This also simplifies the name management
and splits the custom library management out from the unittest specific
management. It finally drops the dependency on parsing cmake arguments.
Richard Smith [Thu, 21 Jun 2012 01:08:35 +0000 (01:08 +0000)]
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.
Dmitri Gribenko [Thu, 21 Jun 2012 00:28:14 +0000 (00:28 +0000)]
RawCommentList::addComment: fix the assertion so it actually checks that new comment is after the last one (change Comments[0] to Comments.back()), and handle the case of two consecutive comments, e.g. /** *//* */. There is already a testcase for that (but it didn't trigger the assert because the assert itself was wrong).
objective-c: Normally, a property cannot be both 'readonly' and having a
"write" attribute (copy/retain/etc.). But, property declaration in
primary class and protcols are tentative as they may be overridden
into a 'readwrite' property in class extensions. Postpone diagnosing
such warnings until the class implementation is seen.
// rdar://11656982
John McCall [Wed, 20 Jun 2012 21:58:02 +0000 (21:58 +0000)]
Introduce ObjCRuntime::hasWeakClassImport() and use it in the appropriate
places. I've turned this off for the GNU runtimes --- I don't know if
they support weak class import, but it's easy enough for them to opt in.
projects / clang / log