Dan Gohman [Fri, 16 Apr 2010 04:45:02 +0000 (04:45 +0000)]
Convert libCIndex to use the new native EXPORTED_SYMBOL_FILE mechanism.
libCIndex also has a CMakeLists.txt file which has its own code for using
the exports file. To preserve existing functionality, create a separate
darwin-specific exports file for use by this CMakeLists.txt code.
Douglas Gregor [Fri, 16 Apr 2010 00:23:51 +0000 (00:23 +0000)]
Fix a bug in caret-line-pruning logic that only happens when we have a
source line wider than the terminal where the associated fix-it line
is longer than the caret line. Previously, we would crash in this
case, which was rather unfortunate. Fixes <rdar://problem/7856226>.
Douglas Gregor [Thu, 15 Apr 2010 23:40:53 +0000 (23:40 +0000)]
Audit uses of Sema::LookupSingleName for those lookups that are
intended for redeclarations, fixing those that need it. Fixes PR6831.
This uncovered an issue where the C++ type-specifier-seq parsing logic
would try to perform name lookup on an identifier after it already had
a type-specifier, which could also lead to spurious ambiguity errors
(as in PR6831, but with a different test case).
Douglas Gregor [Thu, 15 Apr 2010 22:33:43 +0000 (22:33 +0000)]
Feed proper source-location information into Sema::LookupSingleResult,
in case it ends up doing something that might trigger diagnostics
(template instantiation, ambiguity reporting, access
reporting). Noticed while working on PR6831.
Do not generate USRs for declarations with 'no linkage' except for enums, structs, typedefs.
Those still need work to disambiguate them across translation units.
Fix PR 6844, a regression caused by the introduction of llvm_unreachable for the default
case in GRExprEngine::Visit (in r101129). Instead, enumerate all Stmt cases and have
no 'default' case in the switch statement. When we encounter a Stmt we don't handle,
we should explicitly add it to the switch statement.
Douglas Gregor [Thu, 15 Apr 2010 15:53:31 +0000 (15:53 +0000)]
Fix a few cases where enum constant handling was using
ASTContext::getTypeSize() rather than ASTContext::getIntWidth() for
the width of an integral type. The former includes padding for bools
(to the target's size) while the latter does not, so we woud end up
zero-extending bools to the target width when we shouldn't. Fixes a
crash-on-valid in the included test.
Daniel Dunbar [Thu, 15 Apr 2010 06:18:39 +0000 (06:18 +0000)]
Add TargetInfo::useBitfieldTypeAlignment().
- Used to determine whether the alignment of the type in a bit-field is
respected when laying out structures. The default is true, targets can
override this as needed.
- This is designed to correspond to the PCC_BITFIELD_TYPE_MATTERS macro in
gcc. The AST/Sema implementation only affects one line, unless I have
forgotten something. I'd appreciate further review.
- IRgen still needs to be updated to fully support this (which is effectively
PR5591).
Daniel Dunbar [Thu, 15 Apr 2010 06:09:03 +0000 (06:09 +0000)]
Driver/Frontend: Add support for -mllvm, which forwards options to the LLVM option parser.
- Note that this is a behavior change, previously -mllvm at the driver level forwarded to clang -cc1. The driver does a little magic to make sure that '-mllvm -disable-llvm-optzns' works correctly, but other users will need to be updated to use -Xclang.
Daniel Dunbar [Thu, 15 Apr 2010 03:47:33 +0000 (03:47 +0000)]
IRgen: (Reapply 101222, with fixes) Move EmitStoreThroughBitfieldLValue to use new CGBitfieldInfo::AccessInfo decomposition, instead of computing the access policy itself.
- Sadly, this doesn't seem to give any .ll size win so far. It is possible to make this routine significantly smarter & avoid various shifting, masking, and zext/sext, but I'm not really convinced it is worth it. It is tricky, and this is really instcombine's job.
- No intended functionality change; the test case is just to increase coverage & serves as a demo file, it worked before this commit.
The new fixes from r101222 are:
1. The shift to the target position needs to occur after the value is extended to the correct size. This broke Clang bootstrap, among other things no doubt.
2. Swap the order of arguments to OR, to get a tad more constant folding.
Eric Christopher [Thu, 15 Apr 2010 01:43:08 +0000 (01:43 +0000)]
Rewrite handling of 64-bit palignr intrinsics to be vector shuffles.
Stop multiplying constant by 8 accordingly in the header and change
intrinsic definition for what types we expect.
Add to existing palignr test to check that we're emitting the correct things.
Add optional timing logging for code completion results. This causes a UDP packet
containing the time taken for the code completion to be sent to a designated server
(which is specified using a compile-time -D flag).
Douglas Gregor [Thu, 15 Apr 2010 00:00:53 +0000 (00:00 +0000)]
Warn about non-aggregate classes with no user-declared constructors
that have reference or const scalar members, since those members can
never be initializer or modified. Fixes <rdar://problem/7804350>.
Douglas Gregor [Wed, 14 Apr 2010 23:11:21 +0000 (23:11 +0000)]
Always diagnose and complain about problems in
ResolveAddressOfOverloadedFunction when asked to complain. Previously,
we had some weird handshake where ResolveAddressOfOverloadedFunction
expected its caller to handle some of the diagnostics but not others,
and yet there was no way for the caller to know which case we were
in. Eliminate this madness, fixing <rdar://problem/7765884>.
Douglas Gregor [Wed, 14 Apr 2010 22:19:45 +0000 (22:19 +0000)]
Once we've emitted a fatal diagnostic, keep counting errors but with a
separate count of "suppressed" errors. This way, semantic analysis
bits that depend on the error count to determine whether problems
occured (e.g., some template argument deduction failures, jump-scope
checking) will not get confused.
The actual problem here is that a missing #include (which is a fatal
error) could cause the jump-scope checker to run on invalid code,
which it is not prepared to do. Trivial fix for both
<rdar://problem/7775941> and <rdar://problem/7775709>.
Douglas Gregor [Wed, 14 Apr 2010 20:04:41 +0000 (20:04 +0000)]
Teach typo correction about various language keywords. We can't
generally recover from typos in keywords (since we would effectively
have to mangle the token stream). However, there are still benefits to
typo-correcting with keywords:
- We don't make stupid suggestions when the user typed something
that is similar to a keyword.
- We can suggest the keyword in a diagnostic (did you mean
"static_cast"?), even if we can't recover and therefore don't have
a fix-it.
Douglas Gregor [Wed, 14 Apr 2010 17:09:22 +0000 (17:09 +0000)]
Return the corrected DeclarationName from Sema::CorrectTypo rather
than just a bool indicating that correction occurred. No actual
functionality change (it's still always used like a bool), but this
refactoring will be used to support typo correction to keywords.
Douglas Gregor [Wed, 14 Apr 2010 16:09:52 +0000 (16:09 +0000)]
When diagnosing suspicious precedence or assignments, move the fix-it
that adds parentheses from the main diagnostic down to a new
note. This way, when the fix-it represents a choice between two
options, each of the options is associted with a note. There is no
default option in such cases. For example:
/Users/dgregor/t.c:2:9: warning: & has lower precedence than ==; ==
will be
evaluated first [-Wparentheses]
if (x & y == 0) {
^~~~~~~~
/Users/dgregor/t.c:2:9: note: place parentheses around the &
expression to
evaluate it first
if (x & y == 0) {
^
( )
/Users/dgregor/t.c:2:9: note: place parentheses around the ==
expression to
silence this warning
if (x & y == 0) {
^
( )
Daniel Dunbar [Wed, 14 Apr 2010 05:48:35 +0000 (05:48 +0000)]
Speculatively revert "IRgen: Move EmitStoreThroughBitfieldLValue to use new CGBitfieldInfo::AccessInfo decomposition, instead of computing the access policy itself.", I think it might be breaking bootstrap.
Daniel Dunbar [Wed, 14 Apr 2010 04:08:03 +0000 (04:08 +0000)]
IRgen: Move EmitStoreThroughBitfieldLValue to use new CGBitfieldInfo::AccessInfo decomposition, instead of computing the access policy itself.
- Sadly, this doesn't seem to give any .ll size win so far. It is possible to make this routine significantly smarter & avoid various shifting, masking, and zext/sext, but I'm not really convinced it is worth it. It is tricky, and this is really instcombine's job.
- No intended functionality change; the test case is just to increase coverage & serves as a demo file, it worked before this commit.
Douglas Gregor [Wed, 14 Apr 2010 02:46:37 +0000 (02:46 +0000)]
Implement typo correction for Objective-C message sends when the
receiver is a mis-typed class name. Previously, we would give a non-specific
typo-correction diagnostic from the expression-parsing code, but there
was no fix-it because it was too late to recover. Now, we give a nice
diagnostic
honk.m:6:4: error: unknown receiver 'Hnk'; did you mean 'Honk'?
[Hnk method];
^~~
Honk
honk.m:1:1: note: 'Honk' declared here
@interface Honk
^
which includes a fix-it.
We still need to recover better from mis-typing "super".
Douglas Gregor [Wed, 14 Apr 2010 02:22:16 +0000 (02:22 +0000)]
Introduce a parsing action to distinguish between class, instance, and
super message sends in Objective-C. No actual functionality change
here, but it provides a hook so that Sema can typo-correct the
receiver in some cases.
John McCall [Wed, 14 Apr 2010 01:27:20 +0000 (01:27 +0000)]
Mark a function declaration invalid if any of its parameter declarations
are invalid. Prevents a crash-on-invalid during template instantiation.
I... really don't understand how this wasn't already present.
John McCall [Wed, 14 Apr 2010 00:24:33 +0000 (00:24 +0000)]
Parse friend template ids as types instead of ending up in
ActOnClassTemplateSpecialization and being very confused.
Fixes PR6514 (for non-templated-scope friends).
Use ASTVector instead of std::vector for the Exprs in InitListExpr. Performance
measurements of '-fsyntax-only' on combine.c (403.gcc) shows no real performance
change, but now the vector isn't leaked.
Introduce ASTVector, which is a std::vector-like class that allocates all memory
using the allocator associated with an ASTContext. This is largely copy-and-paste
from SmallVector, and should be refactored one day.
Daniel Dunbar [Tue, 13 Apr 2010 23:34:15 +0000 (23:34 +0000)]
IRgen: Move EmitLoadOfBitfieldLValue to use new CGBitfieldInfo::AccessInfo decomposition, instead of computing the access policy itself.
- This lets the method focus slightly more on emitting clean IR to honor the policy which has been selected. On 403.gcc's combine.c, x86_64, -O0, this reduces the number of lines in the .ll file (~= # of instructions) by 2.5%.
- No intended functionality change -- at -O3 this should produce equivalent if not identical output. On 403.gcc's combine.c, x86_64, -O3, this isn't quite true and some of the changes are regressions, but I'm not going to worry about that until we move to a new access policy.
- There is still some room for improvement in the generated IR, in particular we can usually fold the sign-extension of the bit-field into one of the component access. See the FIXME.
Daniel Dunbar [Tue, 13 Apr 2010 20:58:55 +0000 (20:58 +0000)]
IRgen: Enhance CGBitFieldInfo with enough information to fully describe the "policy" with which a bit-field should be accessed.
- For now, these policies are computed to match the current IRgen strategy, although the new information isn't being used yet (except in -fdump-record-layouts).
Chris Lattner [Tue, 13 Apr 2010 18:16:19 +0000 (18:16 +0000)]
Rework the ConstStructBuilder code to emit missing initializer
elements with explicit zero values instead of with tail padding.
On an example like this:
struct foo { int a; int b; };
struct foo fooarray[] = {
{1, 2},
{4},
};
We now lay this out as:
@fooarray = global [2 x %struct.foo] [%struct.foo { i32 1, i32 2 }, %struct.foo { i32 4, i32 0 }]
Chris Lattner [Tue, 13 Apr 2010 17:33:56 +0000 (17:33 +0000)]
make the rewriter add a #ifndef around the #define of __attribute__.
Without it, there is no reason for a compiler that supports it to
emit the dead static globals that the rewriter labels attribute(used).
Douglas Gregor [Tue, 13 Apr 2010 16:31:36 +0000 (16:31 +0000)]
Refactor and simplify the computation of implicit conversion sequences
for reference binding. The code attempted to handle both the
computation of the ICS and the actual conversion, but the latter is an
anachronism: we now use InitializationSequence for that.
Sema::CheckReferenceInit is now a static function TryReferenceInit
that's only use within overload resolution, and has been simplified
slightly. It still needs to be updated per C++ [over.ics.ref], by
eliminating more of the lvalue/rvalue checks.
Douglas Gregor [Tue, 13 Apr 2010 15:50:39 +0000 (15:50 +0000)]
When returning the result of a call to an object of class type, do not
return a NULL expression; return either an error or a proper
expression. Fixes PR6078.
Douglas Gregor [Tue, 13 Apr 2010 15:07:45 +0000 (15:07 +0000)]
During referencing binding, only consider conversion functions for
direct reference binding when the source and target types are not
reference-related. Fixes PR6066.
John McCall [Tue, 13 Apr 2010 01:44:10 +0000 (01:44 +0000)]
Don't try to find a scope corresponding to the search DC for an unfound
friend declaration; this used to be important but is now just a waste of time
plus an unreasonable assertion. Fixes PR6174.
Douglas Gregor [Mon, 12 Apr 2010 23:42:09 +0000 (23:42 +0000)]
Implement C++ [over.ics.user]p3, which restricts the final conversion
from a conversion function template specialization to one of exact
match rank. We only know how to test this in C++0x with default
function template arguments, but it's also in the C++03 spec. Fixes
PR6285.
Douglas Gregor [Mon, 12 Apr 2010 23:19:01 +0000 (23:19 +0000)]
Improve source-location information for C++ conversion functions, by
copying the type location information from the conversion-type-id into
the type location information for the function type. Do something
similar for constructors and destructors, by giving their "void"
return type source-location information.
In all of these cases, we previously left this type-source information
uninitialized, which led to various unfortunate crashes.
We still aren't tracking good source-location information for the
actual names. That's PR6357.