property references to use a new PseudoObjectExpr
expression which pairs a syntactic form of the expression
with a set of semantic expressions implementing it.
This should significantly reduce the complexity required
elsewhere in the compiler to deal with these kinds of
expressions (e.g. IR generation's special l-value kind,
the static analyzer's Message abstraction), at the lower
cost of specifically dealing with the odd AST structure
of these expressions. It should also greatly simplify
efforts to implement similar language features in the
future, most notably Managed C++'s properties and indexed
properties.
Most of the effort here is in dealing with the various
clients of the AST. I've gone ahead and simplified the
ObjC rewriter's use of properties; other clients, like
IR-gen and the static analyzer, have all the old
complexity *and* all the new complexity, at least
temporarily. Many thanks to Ted for writing and advising
on the necessary changes to the static analyzer.
I've xfailed a small diagnostics regression in the static
analyzer at Ted's request.
llvm-svn: 143867
path. That assumption should never have been true, but it was until
I fixed it. Now that its fixed, add a triple here to get correct
behavior even on Windows.
llvm-svn: 143863
edge cases and have better behavior. Specifically, we should actually
prefer the general '4.6' version string over the '4.6.1' string, as
'4.6.2' should be able to replace it without breaking rpaths or any
other place that these paths have been embedded. Debian-based
distributions are already using a path structure with symlinks to
achieve in-place upgrades for patch versions. Now our parsing reflects
this and we select the shorter paths instead of the longer paths.
A separate issue was that we would not parse a leading patch version
number even in the presence of a suffix. The above change makes this
more problematic as it would cause a suffix being added to make us treat
the entire thing as patch-version-agnostic, which it isn't. This changes
the logic to distinguish between '4.4.x' and 4.4.1-x', and retain that
the latter has *some* patch number information. Currently, we always
bias toward the shorter and more canonical version strings. If it
becomes important we can add more Debian like rules to produce sequences
such as '4.4.1b' > '4.4.1' > '4.4.1-rc3' > '4.4.1-rc2' > '4.4.1-pre5',
but I'm very doubtful this will ever matter or be desirable.
I've made the tests for this logic a bit more interesting, and added
some specific tests for logic that is now different.
llvm-svn: 143841
rather than presuming that it is 3.0. This is extra important as the
version should be 3.1, but CMake hasn't caught up with the times.
That'll be fixed in a separate commit.
llvm-svn: 143823
builtin headers are no longer going to receive the old 'implicit extern
"C" block' semantics. This hint is actually ignored by both Clang and
GCC at this point, and Clang's own builtin headers can simply be changed
if there is any issue with this. Clang should be free to include these
however it wants, and so shorter and simpler is better.
Note: *nothing* is changing about the *system* stddef.h include. That
should always have the exact same include semantics, whether with Clang
or GCC or any other compiler. Only the compiler-builtin header search
path is changing.
If anyone knows of some risk that this introduces that I've not thought
of, please chime in. So far, only Windows has switched to the Brave New
World, but others should be switching soon.
llvm-svn: 143806
to do "realistic" includes, and so need the header search logic now in
the driver. This in turn requires switching the CC1 options to the
actual driver options, and passing -Xclang where there is no analogy.
llvm-svn: 143805
is a pretty gross hack, but I don't have any significantly cleaner ideas
for this. There are several things obviously gross about it:
1) Lit shouldn't know that Clang needs this. This really that bad, as
Lit already knows about CC1 and other internal details.
2) This hard codes the '3.0' version number, which is pretty lame.
3) This hard codes every other aspect of the resource dir structure
which is less lame than the version number, but still not great.
However, it should bring the MSVC tests back to life, and it should
unblock the rest of the move from Frontend to Driver, so I think it's
worth a bit of grossness that is isolated in our testing infrastructure
while we figure out the best long term approach. I have the following
ideas, some of which only solve part of the problem (and thus might need
to be combined with other ideas):
a) Create a symlink or other convenience path instead of a version
number.
b) Run 'clang' directly in the lit.cfg, look at its resource dir, and use
that.
c) Switch all the tests to use the driver instead of CC1.
d) Hack the frontend to synthesize builtin include directories when none
are provided by the driver.
I don't like (d) because it feels very hackish and likely to break. We
can only solve a small part of the problem with (a). I wanted to vote
for (c), but lots of the tests in this bucket are really heavily using
internal-only flags like -verify and -triple. I'm loath to complicate
them with the full driver layer. Also, switching them to the driver adds
more than just builtin headers, but all of the rest of the system
headers!
This leaves me with (b). If others like (b), I'll switch to it, but it
felt a bit icky. Nothing concrete, and the other options look
significantly worse, but I felt icky enough that I wanted to start with
a more brain-dead patch to stop the bleeding, and gauge others' feelings
here.
llvm-svn: 143804
to allow us to implement the C++11 rule that a non-active union member can't be
read, and use it to implement subobject access for string literals.
llvm-svn: 143677
A PCH file keeps track of #pragma diagnostics state; when loading the preamble, they conflicted
with the #pragma diagnostic state already present in the DiagnosticsEngine object due to
parsing the preamble.
Fix this by clearing the state of the DiagnosticsEngine object.
Fixes rdar://10363572 && http://llvm.org/PR11254.
llvm-svn: 143644
definition, we may not have a scope corresponding to the namespace
where that friend function template actually lives. Work around this
issue by faking up a scope with the appropriate DeclContext.
This is a bit of a hack, but it fixes <rdar://problem/10204947>.
llvm-svn: 143614
to types. Enable this flag for code completion, where knowing whether
something is in an anonymous or inline namespace is actually not
useful, since you don't have to type it anyway. Fixes
<rdar://problem/10208818>.
llvm-svn: 143599
the injected-class-name of a class (or class template) to the
declaration that results from substituting the given template
arguments. Previously, we would actually perform a substitution into
the injected-class-name type and then retrieve the resulting
declaration. However, in certain, rare circumstances involving
deeply-nested member templates, we would get the wrong substitution
arguments.
This new approach just matches up the declaration with a declaration
that's part of the current context (or one of its parents), which will
either be an instantiation (during template instantiation) or the
declaration itself (during the definition of the template). This is
both more efficient (we're avoiding a substitution) and more correct
(we can't get the template arguments wrong in the member-template
case).
Fixes <rdar://problem/9676205>.
llvm-svn: 143551
wrong class, make sure to drop it immediately; we don't want that
constructor to be available within the DeclContext. Fixes
<rdar://problem/9677163>.
llvm-svn: 143506
does not match any declaration in the class (or class template), be
sure to mark it as invalid. Fixes PR10924 / <rdar://problem/10119422>.
llvm-svn: 143504
it contains give it a USR based on its semantic context, which is the interface.
This follows what we already did for objc methods. rdar://10371669
llvm-svn: 143464
that it retains source location information for the type. Aside from
general goodness (being able to walk the types described in that
information), we now have a proper representation for dependent
delegating constructors. Fixes PR10457 (for real).
llvm-svn: 143410
just integers and floating point types. Since we don't support evaluating class
types or performing lvalue-to-rvalue conversions on array elements yet, this
just means pointer types right now.
llvm-svn: 143298
Track the function invocation where an lvalue referring to a constexpr function
parameter originated from, and use it to substitute the correct argument and to
determine whether such an argument's lifetime has ended.
llvm-svn: 143296
constexpr function arguments outside of their function (passing or returning
them by reference) does not work correctly yet.
Calling constexpr function templates does not work yet, since the bodies are not
instantiated until the end of the translation unit.
llvm-svn: 143234
are present in all the necessary places:
In constant expression evaluation, evaluate lvalues as lvalues and rvalues as
rvalues. Remove special case for caching reference initialization and fix a
cyclic initialization crash in the process.
llvm-svn: 143204
string is part of the function call, then there is no difference. If the
format string is not, the warning will point to the call site and a note
will point to where the format string is.
Fix-it hints for strings are moved to the note if a note is emitted. This will
prevent changes to format strings that may be used in multiple places.
llvm-svn: 143168
rvalue. An assertion to catch this is in ImpCastExprToType will follow, but
vector operations currently trip over this (due to omitting the usual arithmetic
conversions). Also add an assert to catch missing lvalue-to-rvalue conversions
on the LHS of ->.
llvm-svn: 143155
The OpenCL single precision division operation is only required to
be accurate to 2.5ulp. Annotate the fdiv instruction with metadata
which signals to the backend that an imprecise divide instruction
may be used.
llvm-svn: 143136
AST file more lazy, so that we don't eagerly load that information for
all known identifiers each time a new AST file is loaded. The eager
reloading made some sense in the context of precompiled headers, since
very few identifiers were defined before PCH load time. With modules,
however, a huge amount of code can get parsed before we see an
@import, so laziness becomes important here.
The approach taken to make this information lazy is fairly simple:
when we load a new AST file, we mark all of the existing identifiers
as being out-of-date. Whenever we want to access information that may
come from an AST (e.g., whether the identifier has a macro definition,
or what top-level declarations have that name), we check the
out-of-date bit and, if it's set, ask the AST reader to update the
IdentifierInfo from the AST files. The update is a merge, and we now
take care to merge declarations before/after imports with declarations
from multiple imports.
The results of this optimization are fairly dramatic. On a small
application that brings in 14 non-trivial modules, this takes modules
from being > 3x slower than a "perfect" PCH file down to 30% slower
for a full rebuild. A partial rebuild (where the PCH file or modules
can be re-used) is down to 7% slower. Making the PCH file just a
little imperfect (e.g., adding two smallish modules used by a bunch of
.m files that aren't in the PCH file) tips the scales in favor of the
modules approach, with 24% faster partial rebuilds.
This is just a first step; the lazy scheme could possibly be improved
by adding versioning, so we don't search into modules we already
searched. Moreover, we'll need similar lazy schemes for all of the
other lookup data structures, such as DeclContexts.
llvm-svn: 143100
The code had it backwards, thinking size_t was signed, and using that for "%zd".
Also let the analysis get the types for (u)intmax_t while we are at it.
llvm-svn: 143099
itself via an asm label.
available_externally functions are supposed to correspond to an external
function, and that is not the case in the examples in pr9614.
llvm-svn: 143049
essence, the redeclaration chain for a class could end up in an
inconsistent state while deserializing multiple declarations in that
chain, where the circular linked list was not, in fact,
circular. Since only two redeclarations of the same entity will get
loaded when we're in this state, restore circularity when both have
been loaded. Fixes <rdar://problem/10324940> / PR11195.
llvm-svn: 143037
expressions: expressions which refer to a logical rather
than a physical l-value, where the logical object is
actually accessed via custom getter/setter code.
A subsequent patch will generalize the AST for these
so that arbitrary "implementing" sub-expressions can
be provided.
Right now the only client is ObjC properties, but
this should be generalizable to similar language
features, e.g. Managed C++'s __property methods.
llvm-svn: 142914
Microsoft __if_exists/__if_not_exists statement. Also note that we
weren't traversing DeclarationNameInfo *at all* within the
RecursiveASTVisitor, which would be rather fatal for variadic
templates.
llvm-svn: 142906
statements. As noted in the documentation for the AST node, the
semantics of __if_exists/__if_not_exists are somewhat different from
the way Visual C++ implements them, because our parsed-template
representation can't accommodate VC++ semantics without serious
contortions. Hopefully this implementation is "good enough".
llvm-svn: 142901
analysis to separate dependent names from non-dependent names. For
dependent names, we'll behave differently from Visual C++:
- For __if_exists/__if_not_exists at class scope, we'll just warn
and then ignore them.
- For __if_exists/__if_not_exists in statements, we'll treat the
inner statement as a compound statement, which we only instantiate
in templates where the dependent name (after instantiation)
exists. This behavior is different from VC++, but it's as close as
we can get without encroaching ridiculousness.
The latter part (dependent statements) is not yet implemented.
llvm-svn: 142864
rvalues, as C++11 constant evaluation semantics require. DeclRefs referring to
references can now use the normal initialization-caching codepath, which
incidentally fixes a crash in cyclic initialization of references.
llvm-svn: 142844
Former RUN line does not make sense on Win32 hosts.
Win32 hosts would not be expected to set appropriate $PWD.
Latter RUN line might be made valid if PathV2::is_absolute("/foo") would be true. Unfortunately, "/foo" should not be treated as absolute path on Win32.
FYI, on mingw32 with MSYS bash (it has 'shell' feature);
Former) $PWD is set as "X:/hogehoge/test/Driver"
Latter) PWD=/foo sets "X:/root/to/msys/foo" to $PWD.
llvm-svn: 142721
in such a case just write out a reference of a previously serialized Stmt, instead
of serializing it all over again.
This saves memory + space + [de]serializing time, and avoids blowing up memory
with pathological cases. rdar://10293911
llvm-svn: 142696
class declaration which forces any such class and any
class that inherits from such a class to have their
typeinfo symbols be marked as weak.
// rdar://10246395
A test/CodeGenCXX/weak-extern-typeinfo.cpp
M lib/Sema/SemaDeclCXX.cpp
M lib/Sema/SemaDeclAttr.cpp
M lib/CodeGen/CGRTTI.cpp
llvm-svn: 142693
be sure to consider all of the possible lookup results. We were
assert()'ing (but behaving correctly) for unresolved values. Fixes
PR11134 / <rdar://problem/10290422>.
llvm-svn: 142652
but trivially constructible and destructible variables in C++11 mode. Also
incidentally improve the precision of the wording for jump diagnostics in C++98
mode.
llvm-svn: 142619
actually just has an extraneous 'template<>' header, strip off the
'template<>' header and treat it as a normal friend tag. Fixes PR10660
/ <rdar://problem/9958322>.
llvm-svn: 142587
instead of a semicolon (as sometimes happens during refactorings). When such a
comma is seen at the end of a line, and is followed by something which can't
possibly be a declarator (or even something which might be a plausible typo for
a declarator), suggest that a semicolon was intended.
llvm-svn: 142544
Add test that a variadic base list which expands to 0 bases doesn't make the
class a non-aggregate. This test passed before the change, too.
llvm-svn: 142411
Use "%f" as format string to make sure it doesn't match size_t, etc.
whatever they might be typedeffed to, so that the fixit always applies.
llvm-svn: 142348
Turns out this part of the test from r142342 wasn't portable.
The errors on the bots look like this:
E:\bb-win7\cmake-clang-i686-msys\build\tools\clang\test\Sema\Output\format-strings-fixit.c.tmp:58:13: error: conversion specifies type 'unsigned int' but the argument has type 'ssize_t' (aka 'long')
printf("%zd", (ssize_t) 42);
~~^ ~~~~~~~~~~~~
%zd
Obviously we can't typedef ssize_t to someting that doesn't have the same size as size_t and expect it to work.
But it's also weird that the format string "%zd" gets interpreted as "unsigned int" when it should clearly be signed.
llvm-svn: 142345
For PR11152. Make PrintSpecifier::fixType() suggest "%zu" for size_t, etc.
rather than looking at the underlying type and suggesting "%llu" or other
platform-specific length modifiers. Applies to C99 and C++11.
llvm-svn: 142342
arguments as block literal arguments; the block literal argument code
completion should only go one level deep. Fixes <rdar://problem/10291294>.
llvm-svn: 142335
the right namespace in C++11 mode. Teach the code to prefer the 'must be in
precisely this namespace' diagnostic whenever that's true, and fix a defect
which resulted in the -Wc++11-compat warning in C++98 mode sometimes being
omitted.
llvm-svn: 142329
changed the return type of a compare of two unsigned vectors to be unsigned. This patch removes the check for hasIntegerRepresentation since its not needed and returns the appropriate signed type.
I added a new test case and updated exisiting test cases that assumed an unsigned result.
llvm-svn: 142250
as part of the hash rather than ignoring them. This means we'll end up
building more module variants (overall), but it allows configuration
macros such as NDEBUG to work so long as they're specified via command
line. More to come in this space.
llvm-svn: 142187
This also applies to C99-style aggregate literals, should they be used in C++11, since they are effectively identical to constructor call list-initialization syntax.
llvm-svn: 142147
formatting as any other diagnostic, they will be properly line wrapped and
otherwise pretty printed. Let's take advantage of that and the new factoring to
add some helpful information to them (much like template backtrace notes and
other notes): the name of the macro whose expansion is being noted. This makes
a world of difference if caret diagnostics are disabled, making the expansion
notes actually useful in this case. It also helps ensure that in edge cases the
information the user needs is present. Consider:
% nl -ba t5.cc
1 #define M(x, y, z) \
2 y
3
4 M(
5 1,
6 2,
7 3);
We now produce:
% ./bin/clang -fsyntax-only t5.cc
t5.cc:6:3: error: expected unqualified-id
2,
^
t5.cc:2:3: note: expanded from macro: M
y
^
1 error generated.
Without the added information in the note, the name of the macro being expanded
would never be shown.
This also deletes a FIXME to use the diagnostic formatting. It's not yet clear
to me that we *can* do this reasonably, and the production of this message was
my primary goal here anyways.
I'd love any comments or suggestions on improving these notes, their wording,
etc. Currently, I need to make them provide more helpful information in the
presence of a token-pasting buffer, and I'm pondering adding something along
the lines of "expanded from argument N of macro: ...".
llvm-svn: 142127
standing deficiency: we were providing no macro backtrace information
whenever caret diagnostics were turned off. This sinks the logic for
suppressing the code snippet and caret to the code that actually prints
tho code snippet and caret. Along the way, clean up the naming of
functions, remove some now fixed FIXMEs, and generally improve the
wording and logic of this process.
Add a test case exerecising this functionality. It is notable that the
resulting messages are extremely low quality. I'm working on a follow-up
patch that should address this and have left a FIXME in the test case.
llvm-svn: 142120
the important code in this test to make the test more stable. Now adding
further tests won't shift the line numbers occuring in the diagnostic
output.
llvm-svn: 142118
This also adds a -Wc++98-compat-pedantic for warning on constructs which would
be diagnosed by -std=c++98 -pedantic (that is, it warns even on C++11 features
which we enable by default, with no warning, in C++98 mode).
llvm-svn: 142034
through varargs. This only happens when we're in an unevaluated
context, where we don't want to trigger an error anyway. Fixes PR11131
/ <rdar://problem/10288375>.
llvm-svn: 141986
There are now separate Triple::MacOSX and Triple::IOS values for the OS
so comparing against Triple::Darwin will fail to match those. Note that
I changed the expected output for the Driver/rewrite-objc.m test, which had
previously not been passing Darwin-specific options with the macosx triple.
llvm-svn: 141944
If the code file is not run through the preproccessor to remove comments,
then FileCheck will match the strings within the CHECK commands rendering
the test useless.
llvm-svn: 141911
C++11 mode but keep their sources compatible with C++98. This patch implements
the -Wc++98-compat-variadic-templates sub-flag and -Wc++98-compat to include
it.
llvm-svn: 141898
Start handling debug line and scope information better:
Migrate most of the location setting within the larger API in CGDebugInfo and
update a lot of callers.
Remove the existing file/scope change machinery in UpdateLineDirectiveRegion
and replace it with DILexicalBlockFile usage.
Finishes off the rest of rdar://10246360
after fixing a few bugs that were exposed in gdb testsuite testing.
llvm-svn: 141893
part of template argument deduction is ill-formed, we mark it as
invalid and treat it as a deduction failure. If we happen to find that
specialization again, treat it as a deduction failure rather than
silently building a call to the declaration.
Fixes PR11117, a marvelous bug where deduction failed after creating
an invalid specialization, causing overload resolution to pick a
different candidate. Then we performed a similar overload resolution
later, and happily picked the invalid specialization to
call... resulting in a silent link failure.
llvm-svn: 141809
Argyrios Kyrtzidis