Douglas Gregor [Mon, 28 Nov 2011 16:30:08 +0000 (16:30 +0000)]
When checking a call to a builtin atomic operation, be sure to
consider the _<width> variants as well, which we'll see if we're
performing the type checking in a template instantiation where the
call expression itself was originally not type-dependent. Fixes
PR11411.
Chandler Carruth [Mon, 28 Nov 2011 13:02:29 +0000 (13:02 +0000)]
And that completes my sweep through commits for things that might be
worth noting in the release notes. These remain raw notes. I'll be
re-writing them into nice prose first thing tomorrow, with help from
others. A couple of notes for any reading the commits:
If you don't see something that should be mentioned, feel free to add
a note (or even a nicely written section) about it! I haven't really
done the static analyzer justice here as I don't really know what the
significant changes are other than mile-high stuff like watching it grow
C++ support and a more robust CFG. I also worry I've missed important
stuff in the Objective-C world.
If you see something that isn't worth mentioning, just delete it. I know
there are several things like this. I plan to prune the list down as
I flesh things out.
If you're name or email is on a bullet, I'll likely be sending you an
email asking for any input on that subject. For many of these I can fill
in something generic, and I'll just want you to give it a once-over.
However, if you have time, feel free to just write the blurb yourself
and drop it in, or drop it in an email to me.
Finally, *WOW* has a lot happened in Clang... I shouldn't have dreaded
(and put off) this so much, it was kind of awesome to go back and watch
the evolution. Anyways, these should be in a reasonable draft state
early tomorrow.
Bob Wilson [Mon, 28 Nov 2011 08:03:54 +0000 (08:03 +0000)]
Install c-index-test and clang-c/Index.h as internal files. rdar://10217046
Specify that these files should be installed to the optional internal
install location as specified by configure's --with-internal-prefix.
If that option is not used, they'll be installed to the default prefix
as before.
Chandler Carruth [Mon, 28 Nov 2011 07:16:19 +0000 (07:16 +0000)]
Begin sketching out the Clang release notes. The plan is to link to
these more detailed notes from the primary LLVM release notes for Clang.
This gives us a nice place to flesh out in plenty of detail the major
changes that have happened in Clang land since 2.9.
I've outlined a very rough structure based on the LLVM release notes
structure and what seems like useful divisions in the Clang landscape
(e.g., language-specific stuff is relevant to a narrower audience).
I'll be first converting my brain-dump-ish notes from the commit logs,
and then cleaning here. Suggestions on structure welcome. Typo
corrections, spelling fixes (oh how I'll need them), all welcome; just
commit away.
When destroying temporaries, instead of a custom cleanup use the
generic pushDestroy function.
This would reduce the number of useful declarations in
CGTemporaries.cpp to one. Since CodeGenFunction::EmitCXXTemporary
does not deserve its own file, move it to CGCleanup.cpp and delete
CGTemporaries.cpp.
Rafael Espindola [Sun, 27 Nov 2011 20:00:43 +0000 (20:00 +0000)]
Make our handling of MMX x SSE closer to what gcc does:
* Enabling sse enables mmx.
* Disabling (-mno-mmx) mmx, doesn't disable sse (we got this right already).
* The order in not important. -msse -mno-mmx is the same as -mno-mmx -msse.
Sebastian Redl [Sun, 27 Nov 2011 16:50:07 +0000 (16:50 +0000)]
Reference initialization with initializer lists.
This supports single-element initializer lists for references according to DR1288, as well as creating temporaries and binding to them for other initializer lists.
Rafael Espindola [Sat, 26 Nov 2011 20:53:19 +0000 (20:53 +0000)]
Add the minimum implementation of cpuid.h. This works on "modern" intel cpus
and on clang, which seams to handled "=b" correctly even when ebx is the
PIC register.
Francois Pichet [Fri, 25 Nov 2011 01:10:54 +0000 (01:10 +0000)]
In Microsoft mode, make "Unqualified lookup into dependent bases of class templates" works inside a friend function definition at class scope.
Basically we have to look into the parent *lexical* DeclContext for friend functions at class scope. That's because calling GetParent() return the namespace or file DeclContext.
This fixes all remaining cases of "Unqualified lookup into dependent bases of class templates" when parsing MFC code with clang.
Richard Trieu [Wed, 23 Nov 2011 22:32:32 +0000 (22:32 +0000)]
Add feature to diagnostics that will provide more information on function
pointer mismatch. Cases covered are: initialization, assignment, and function
arguments. Additional text will give the extra information about the nature
of the mismatch: different classes for member functions, wrong number of
parameters, different parameter type, different return type, and function
qualifier mismatch.
[libclang] Fix operations (token annotation, getting cursor, etc.) with a file region
inside an objc container that "contains" other file-level declarations.
When getting the array of file-level declarations that overlap with a file region,
we failed to report that the region overlaps with an objc container, if
the container had other file-level declarations declared lexically inside it.
Fix this by marking such declarations as "isTopLevelDeclInObjCContainer" in the AST
and handling them appropriately.
Richard Smith [Mon, 21 Nov 2011 19:36:32 +0000 (19:36 +0000)]
Add driver arguments -ftemplate-depth=N and -fconstexpr-depth=N, with the same
semantics and defaults as the corresponding g++ arguments. The historical g++
argument -ftemplate-depth-N is kept for compatibility, but modern g++ versions
no longer document that option.
Add -cc1 argument -fconstexpr-depth N to implement the corresponding
functionality.
Teach the driver about failure result files, which are compilation
output files that are valid regardless of whether the compilation
succeeded or failed (but not if we crash). Add depfiles to the
failure result file list.
Benjamin Kramer [Sat, 19 Nov 2011 10:24:49 +0000 (10:24 +0000)]
Driver: Remove the signal number from the "command failed" diagnostic.
- With the current implementation of sys::Program this always printed "2".
- The command execution code will output the right number anyway (including the signal name).
Ted Kremenek [Fri, 18 Nov 2011 04:32:13 +0000 (04:32 +0000)]
Refine placement of LangOptions object in CompilerInvocation by adding a new baseclass CompilerInvocationBase with a custom copy constructor. This ensures that whenever the CompilerInvocation object's copy constructor is used we always clone the LangOptions object.
Eli Friedman [Fri, 18 Nov 2011 03:47:20 +0000 (03:47 +0000)]
Fix the meaning of an "empty" record for the case of a zero-length array. Use isEmptyRecord for arguments on x86-32; there are structs of size 0 which don't count as empty.
Eli Friedman [Fri, 18 Nov 2011 01:25:50 +0000 (01:25 +0000)]
Simplify code for returning a struct for Darwin x86-32 ABI. Use a better type for a function returning a struct containing only a pointer. Handle the edge case of a struct containing only a float or double plus some dead padding instead of asserting.
-For indexDeclaration, also pass the declaration attributes as an array of cursors.
-Rename CXIndexOpt_OneRefPerFile -> CXIndexOpt_SuppressRedundantRefs, and only pass
a reference if a declaration/definition does not exist in the file.
-Other fixes.
Anna Zaks [Thu, 17 Nov 2011 23:07:28 +0000 (23:07 +0000)]
[analyzer] Do not conjure a symbol when we need to propagate taint.
When the solver and SValBuilder cannot reason about symbolic expressions (ex: (x+1)*y ), the analyzer conjures a new symbol with no ties to the past. This helps it to recover some path-sensitivity. However, this breaks the taint propagation.
With this commit, we are going to construct the expression even if we cannot reason about it later on if an operand is tainted.
Ted Kremenek [Thu, 17 Nov 2011 23:01:24 +0000 (23:01 +0000)]
Make 'LangOptions' in CompilerInvocation a heap-allocated, reference counted object. I discovered that llvm::RefCountedBase<T> has
a bug where the reference count is copied in the copy constructor, which means that there were cases when the CompilerInvocation
objects created by ASTUnit were actually leaked. When I fixed that bug locally, it showed that a whole bunch of code assumed
that the LangOptions object that was part of CompilerInvocation was still alive. By making it heap-allocated and reference counted,
we can keep it around after the CompilerInvocation object goes away.
As part of this change, change CompilerInvocation:getLangOptions() to return a pointer, acting as another clue that this
object may outlive the CompilerInvocation object.
This commit doesn't fix the CompilerInvocation leak itself. That will come when I commit the fix to llvm::RefCountedBase<T> to
mainline LLVM.
Ted Kremenek [Thu, 17 Nov 2011 23:01:17 +0000 (23:01 +0000)]
Simplify crash cleanup logic in ASTUnit::LoadFromCommandLine() by zeroing out two IntrusiveRefCnt pointers after we have assigned their respective values into fields of ASTUnit.
Douglas Gregor [Thu, 17 Nov 2011 22:44:56 +0000 (22:44 +0000)]
When making a suggestion regarding which module to load rather than
preprocess/parse a header, report back with an actual module (which
may be a submodule) rather than just the name of the module.
Douglas Gregor [Thu, 17 Nov 2011 22:09:43 +0000 (22:09 +0000)]
Add the notion of "framework" modules to module maps. Framework
modules (obviously) describe frameworks, and understand the header
layout of frameworks.
projects / clang / log