Various pieces of code, like base initialization in Sema and RTTI IRGen,
don't properly ignore qualifiers on base classes. Instead of auditing the
whole codebase, just strip them off in the getter. (The type as written is
still available in the TypeSourceInfo for code that cares.)
Fixes PR16596.
llvm-svn: 186125
- code cleanup, improved reporting when failures take place
- ensure known thread is interrupted by using pthread_kill() instead of kill() in the signal worker thread
- above should avoid llvm.org/pr16567 on Mac OS X (though kill() could still cause threads to pop out of existance temporarily)
- added an additional check that all threads have exited after worker threads are all join()ed
- logged llvm.org/pr16603 for the new Linux bug discovered
llvm-svn: 186124
Before:
int i; // indented 2 space more than clang-format would use.
SomeReturnType // clang-format invoked on this line.
SomeFunctionMakingLBraceEndInColumn80() {
} // This is the indent clang-format would prefer.
After:
int i; // indented 2 space more than clang-format would use.
SomeReturnType // clang-format invoked on this line.
SomeFunctionMakingLBraceEndInColumn80() {
}
llvm-svn: 186120
(if they are not function-like).
Before:
SomeFunction(aaaaaaaaaaaaaaaaaaaaaa, aaaaaaaaaaaaaaaaaaaaaaaa)
OVERRIDE;
After:
SomeFunction(aaaaaaaaaaaaaaaaaaaaaa,
aaaaaaaaaaaaaaaaaaaaaaaa) OVERRIDE;
llvm-svn: 186117
This puts a slight penalty on the linebreak before the first "<<", so
that clang-format generally tries to keep things on the first line.
User feedback has shown that this is generally desirable.
Before:
llvm::outs()
<< "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaa =" << aaaaaaaaaaaaaaaaaaaaaaaaaaa;
After:
llvm::outs() << "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaa ="
<< aaaaaaaaaaaaaaaaaaaaaaaaaaa;
llvm-svn: 186115
First, the reason I came here: I forgot to look at readdir64_r which had
the exact same bug as readdir_r. However, upon applying the same
quick-fix and testing it I discovered that it still didn't work at all.
As a consequence, I spent some time studying the code and thinking about
it and fixed several other problems.
Second, the code was checking for a null entry and result pointer, but
there is no indication that null pointers are viable here. Certainly,
the spec makes it extremely clear that there is no non-error case where
the implementation of readdir_r fails to dereference the 'result'
pointer and store NULL to it. Thus, our checking for a non-null 'result'
pointer before reflecting that write in the instrumentation was
trivially dead. Remove it.
Third, the interceptor was marking the write to the actual dirent struct
by looking at the entry pointer, but nothing in the spec requires that
the dirent struct written is actually written into the entry structure
provided. A threadlocal buffer would be just as conforming, and the spec
goes out of its way to say the pointer to the *actual* result dirent
struct is stored into *result, so *that* is where the interceptor should
reflect a write occuring. This also obviates the need to even consider
whether the 'entry' parameter is null.
Fourth, I got to the bottom of why nothing at all worked in readdir64_r
-- the interceptor structure for dirent64 was completely wrong in that
it was the same as dirent. I fixed this struct to be correct (64-bit
inode and 64-bit offset! just a 64-bit offset isn't enough!) and added
several missing tests for the size and layout of this struct.
llvm-svn: 186109
We had patterns to match v4i32 immAllZerosV -> V_SET0, but not patterns for
v8i16 (which occurs in the test case) or v16i8. The same was true for
V_SETALLONES (so I added the associated patterns for those as well).
Another bug found by llvm-stress.
llvm-svn: 186108
Patch by Michele Scandale!
Adds a special handling of the case where, during the loop exit
condition rewriting, the exit value is a constant of bitwidth lower
than the type of the induction variable: instead of introducing a
trunc operation in order to match correctly the operand types, it
allows to convert the constant value to an equivalent constant,
depending on the initial value of the induction variable and the trip
count, in order have an equivalent comparison between the induction
variable and the new constant.
llvm-svn: 186107
This fixes a bug (found by csmith) at -O0 where we attempt to create a RLWIMI
with an out-of-range operand. Most uses of the isRunOfOnes function are guarded
by a condition that the value is not zero. This was not true in two places, and
in both places a zero input would result in an out-of-rage MB value (= 32).
To fix this, isRunOfOnes returns false on a zero input (and I've remove one
now-redundant guard).
llvm-svn: 186101
We can vectorize them because in the case where we wrap in the address space the
unvectorized code would have had to access a pointer value of zero which is
undefined behavior in address space zero according to the LLVM IR semantics.
(Thank you Duncan, for pointing this out to me).
Fixes PR16592.
llvm-svn: 186088
Trailing return types can only occur in declaration contexts.
Before:
void f() { auto a = b -> c(); }
After:
void f() { auto a = b->c(); }
llvm-svn: 186087
* All systems we support have some form of long name support.
* The options has different names and semantics in different implementations
('f' on gnu, 'T' on OS X), which makes it unlikely it is normally used on
build systems.
* It was completely untested.
llvm-svn: 186078
Daniel Malea