don't repeatedly loop through identifiers, correcting the same typo'd
identifier over and over again.
We still bail out after 20 typo corrections, but this should help
improve performance in the common case where we're typo-correcting
because the user forgot to include a header.
llvm-svn: 116901
Here's example code:
---
template<class T> class MyClass {
struct S { };
S* NewS() { return new S; }
void DeleteS() { delete NewS(); }
};
---
CXXDeleteExpr::getDestroyedType() on the 'delete NewS()' expression
would crash before this change. Now it returns a dependent type
object. Solution suggested by dgregor.
llvm-svn: 116891
Now MICache is a linked list (per the FIXME), where we tradeoff between MacroInfo objects being in MICache
and MIChainHead. MacroInfo objects in the MICache chain are already "Destroy()'ed", so they can be reused. When
inserting into MICache, we need to remove them from the regular linked list so that they aren't destroyed more than
once.
llvm-svn: 116869
The problem was not the management of MacroInfo objects, but that when we recycle them
via the MICache the memory of the underlying SmallVector (within MacroInfo) was not getting
released. This is because objects stashed into MICache simply are reused with a placement
new, and never have their destructor called.
llvm-svn: 116862
computation to compute the lower bound of the edit distance, so that
we can avoid computing the edit distance for names that will clearly
be rejected later. Since edit distance is such an expensive algorithm
(M x N), this leads to a 7.5x speedup when correcting NSstring ->
NSString in the presence of a Cocoa PCH.
llvm-svn: 116849
list of allocated MacroInfos. This requires only 1 extra pointer per MacroInfo object, and allows us to blow them
away in one place. This fixes an elusive memory leak with MacroInfos (whose exact location I couldn't still figure
out despite substantial digging).
Fixes <rdar://problem/8361834>.
llvm-svn: 116842
within a default argument), recurse into default arguments. Fixes
PR8401, a regression I introduced in r113700 while refactoring our
handling of "used" declarations in default arguments.
llvm-svn: 116817
'../lib/clang/<version>'. Actually use '..' rather than removing the trailing
component to correctly handle paths containing '.' or symlinks in the presence
of -no-canonical-prefixes, etc. This shouldn't change any existing behavior.
llvm-svn: 116803
construct an unsupported friend when there's a friend with a templated
scope specifier. Fixes a consistency crash, rdar://problem/8540527
llvm-svn: 116786
type matches have a bigger impact. The impetus for this change was
that, when initializing an enumeration value, we want enumerators of
that enumeration type to have a higher priority than, e.g., unrelated
local variables.
llvm-svn: 116774
doesn't hold. This fix is to increase the loop unrolling count to 4, which experiments show doesn't typically impact
analysis time. The real fix is to modify the IdempotentOperationsChecker to suppress warnings where an analysis point
could be preceded by a point where we gave up due to loop unrolling.
llvm-svn: 116769
-Wa,-force_cpusubtype_ALL t.c'.
- Tweaks -Wa, and -Xassembler handling to only accept an explicit short list of
arguments and give an obvious unsupported error on others.
llvm-svn: 116759
C++/C99/Objective-C, so that we properly include types. This fix
affects global caching of code-completion results; without caching,
the behavior was already correct.
llvm-svn: 116757
declaring methods and when sending messages to them, by bringing all
of the selector into TypedCheck chunks in the completion result. This
way, we can improve the sorting of these results to account for the
full selector name rather than just the first chunk.
llvm-svn: 116746
The previous method used the DESTDIR environment variable at configure
time, but sometimes it is only available at install time. See PR8397.
llvm-svn: 116689
function parameters weren't converted to use the correct type (x86_mmx). Add a
check, similar to the one in llvm-gcc, to see if we need the x86_mmx type for
that function parameter. If so, it coerces the type to be that.
llvm-svn: 116684
by marking the decl invalid isn't. Make some steps towards supporting these
and then hastily shut them down at the last second by marking them as
unsupported.
llvm-svn: 116661
objc_exception_rethrow, so we don't...", since something is actually trying to
call this with the wrong signature (!). Unfortunately I don't understand the new
EH infrastructure well enough to fix it immediately.
llvm-svn: 116660
flexible array member, so long as the flexibility array member is
either not initialized or is initialized with an empty initializer
list. Fixes <rdar://problem/8540437>.
llvm-svn: 116647
find a copy constructor/assignment operator used
in getter/setter synthesis. This removes an unintended
diagnostics and makes objc++ consistant with objective-c.
// rdar: //8550657.
llvm-svn: 116631
we did was an acceptable lookup. If it is, then we can re-use that
lookup result. If it isn't, we have to perform the lookup again. This
is almost surely the cause behind the mysterious typo.m failures on
some builders; we were getting the wrong lookup results returned.
llvm-svn: 116586
identifiers to determine good typo-correction candidates. Once we've
identified those candidates, we perform name lookup on each of them
and the consider the results.
This optimization makes typo correction > 2x faster on a benchmark
example using a single typo (NSstring) in a tiny file that includes
Cocoa.h from a precompiled header, since we are deserializing far less
information now during typo correction.
There is a semantic change here, which is interesting. The presence of
a similarly-named entity that is not visible can now affect typo
correction. This is both good (you won't get weird corrections if the
thing you wanted isn't in scope) and bad (you won't get good
corrections if there is a similarly-named-but-completely-unrelated
thing). Time will tell whether it was a good choice or not.
llvm-svn: 116528
The failing was due to this:
1. preamble.c contains CR+LF new lines
2. write() is called with a buffer containing the original (CR+LF) to output the result on the console.
3. In text mode(the default), write() convert LF to CR+LF even if LF is preceded by CR, hence we have CR+CR+LF which filecheck interprets as 2 lines.
llvm-svn: 116513
solely based on the names it sees, rather than actual declarations it
gets. In essence, we determine the set of names that are "close
enough" to the typo'd name. Then, we perform name lookup for each of
those names, filtering out those that aren't actually visible, and
typo-correct from the remaining results.
Overall, there isn't much of a change in the behavior of typo
correction here. The only test-suite change comes from the fact that
we make good on our promise to require that the user type 3 characters
for each 1 character corrected.
The real intent behind this change is to set the stage for an
optimization to typo correction (so that we don't need to deserialize
all declarations in a translation unit) and future work in finding
missing qualification ("'vector' isn't in scope; did you mean
'std::vector'?). Plus, the code is cleaner this way.
llvm-svn: 116511
instead of deserializing the complete declaration context of the record.
Iterating over the fields of a record is very common (e.g to determine the layout), unfortunately we needlessly deserialize every declaration
that the declaration context of the record contains; this can be bad for large C++ classes that contain a lot of methods.
Fix this by allow deserialization of just the fields when we want to iterate over them.
Progress for rdar://7260160.
llvm-svn: 116507
and emits an error if a declaration with this name is deserialized from PCH.
This is for testing, to make sure that we don't deserialize stuff needlessly.
llvm-svn: 116505