This is a follow-up to r194907, which added a new -arch setting to make it
easier to specify AVX2 targets. The "-arch x86_64h" option needs to be passed
on to the linker, but it was getting canonicalized to x86_64 by the code
in getArchTypeForDarwinArchName.
llvm-svn: 198096
Most importantly, this makes our vtable layout match MSVC's. Previously
we would emit a return adjusting thunk whenever the return types
differed, even if the adjustment would have been trivial.
MSVC does emit some trivial return adjusting thunks, but only if there
was already an overridden method that required a return adjustment.
llvm-svn: 198080
Thisadds a new warning that warns on code like this:
if (memcmp(a, b, sizeof(a) != 0))
The warning looks like:
test4.cc:5:30: warning: size argument in 'memcmp' call is a comparison [-Wmemsize-comparison]
if (memcmp(a, b, sizeof(a) != 0))
~~~~~~~~~~^~~~
test4.cc:5:7: note: did you mean to compare the result of 'memcmp' instead?
if (memcmp(a, b, sizeof(a) != 0))
^ ~
)
test4.cc:5:20: note: explicitly cast the argument to size_t to silence this warning
if (memcmp(a, b, sizeof(a) != 0))
^
(size_t)( )
1 warning generated.
This found 2 bugs in chromium and has 0 false positives on both chromium and
llvm.
The idea of triggering this warning on a binop in the size argument is due to
rnk.
llvm-svn: 198063
With pragma pack, the layout engine would produce vfptrs that were
packed width rather than pointer width. This patch addresses the issue
and adds a test case.
llvm-svn: 198059
Since this warning was generalized, it was also given a sensible warning group flag and the corresponding test was updated to reflect this.
llvm-svn: 198053
Even g++ considers this a valid C++ identifier and it should only have been
visible in C mode.
Also drop the associated low-value diagnostic.
llvm-svn: 197995
These names weren't referred to anywhere in the source so don't need a written
name.
Depends on the TableGen fix for anonymous records in LLVM r197869.
llvm-svn: 197896
This new warning detects when a function will recursively call itself on every
code path though that function. This catches simple recursive cases such as:
void foo() {
foo();
}
As well as more complex functions like:
void bar() {
if (test()) {
bar();
return;
} else {
bar();
}
return;
}
This warning uses the CFG. As with other CFG-based warnings, this is off
by default. Due to false positives, this warning is also disabled for
templated functions.
llvm-svn: 197853
Without this patch, record decls with invalid out-of-line method delcs would
sometimes be marked invalid, but not always. With this patch, they are
consistently never marked invalid.
(The code to do this was added in
http://lists.cs.uiuc.edu/pipermail/cfe-commits/Week-of-Mon-20100809/033154.html
, but the test from that revision is still passing.)
As far as I can tell, this was the only place where a class was marked invalid
after its definition was complete.
llvm-svn: 197848
These members will still be lazily added to the relevant DWARF DIEs in
LLVM but when enumerating the members they will not appear. This allows
DWARF type units to be more consistent - the type unit will never
contain these special members (so all instances of the type should have
the same DIEs without some having some special members and others having
others) and the special members will be added to the skeletal
declaration that appears in the relevant compile_unit.
llvm-svn: 197844
This matches llc's behavior.
Before this patch clang would create a TargetInfo base on -triple but a llvm
CodeGen based on the triple in the module.
llvm-svn: 197837
This was part of the cause for PR17655. We were generating thunks when
we shouldn't have. I suspect that if we tweak the test case for PR17655
to actually require thunks, we can reproduce the same crash.
llvm-svn: 197836
If a header file belonging to a certain module is not found on the
filesystem, that header gets marked as unavailable. Now, the layering
warning (-fmodules-decluse) should still warn about headers of this
module being wrongfully included. Currently, headers belonging to those
modules are just treated as not belonging to modules at all which means
they can be included freely from everywhere.
To implement this (somewhat) cleanly, I have moved most of the layering
checks into the ModuleMap. This will also help with showing FixIts
later.
llvm-svn: 197805
We have assertions for this, but a few edge cases had snuck through where
we were still unconditionally using 'int'.
<rdar://problem/15703011>
llvm-svn: 197733
Checked on VS10(multiconfig) and some singleconfig builders.
* Assumptions
- You should specify llvm-config as LLVM_CONFIG.
CMake could find one in $PATH by default.
- ENABLE_ASSERTIONS obeys LLVM's.
* Use cases
a) With LLVM build tree
Assume llvm-config is in your build tree.
Everything should work as ever.
b) With *installed* LLVM
Assume distributions. The source tree can be optional.
b1) The source tree is provided on the location `llvm-config --src-root`
- Test utils, FileCheck &c., are imported and built in the new tree.
- Gtest is built in the tree if gtest library is not found.
- Lit is used in $(SRCROOT)/utils/lit/lit.py.
b2) The source tree is not provided
- clang and utilities can be built.
- All tests, unittests and check-clang are invalidated and not built.
llvm-svn: 197697
We started by trying to deserialize decltype(func-param) in a trailing return
type, which causes the function parameter decl to be deserialized, which pulls
in the function decl, which pulls the function type, which pulls the same
decltype() in the return type, and then we crashed.
llvm-svn: 197644
The alignment impact of the virtual bases apperas to be applied in
order, rather than up front. This patch adds the new behavior and
provides a test case.
llvm-svn: 197639
Fixes <rdar://problem/15584219> and <rdar://problem/12241361>.
This change looks large, but all it does is reuse and consolidate
the delayed diagnostic logic for deprecation warnings with unavailability
warnings. By doing so, it showed various inconsistencies between the
diagnostics, which were close, but not consistent. It also revealed
some missing "note:"'s in the deprecated diagnostics that were showing
up in the unavailable diagnostics, etc.
This change also changes the wording of the core deprecation diagnostics.
Instead of saying "function has been explicitly marked deprecated"
we now saw "'X' has been been explicitly marked deprecated". It
turns out providing a bit more context is useful, and often we
got the actual term wrong or it was not very precise
(e.g., "function" instead of "destructor"). By just saying the name
of the thing that is deprecated/deleted/unavailable we define
this issue away. This diagnostic can likely be further wordsmithed
to be shorter.
llvm-svn: 197627
The problem here is more serious than the fix implies. Adding a field
to a class updates the triviality bits for the class (among other
things). Failing to require a complete type before adding the field
meant that these updates don't happen in the well-formed case where
the capture is an uninstantiated class template specialization,
leading the lambda itself to be treated as having a trivial copy
constructor when it shouldn't. Fixes <rdar://problem/15560464>.
llvm-svn: 197623
Move some of the verifier directives away from the end of the pragma line.
This ensures that the diagnostics relate to the trailing token being tested and
not the verifier comments which are themselves part of the token stream.
llvm-svn: 197616
Bob Wilson