this was useful, and on review Doug and I decided it was probably on the level
of a bug in the standard and therefore not worth a warning even in -pedantic.
If someone disagrees and urgently wants clang++ to warn about this in strict
c++98 mode, we can talk about it.
llvm-svn: 91868
the redeclaration problems in the [temp.explicit]p3 testcase worse, but I can
live with that; they'll need to be fixed more holistically anyhow.
llvm-svn: 91771
Magically fixes all the terrible lookup problems associated with not pushing
a new scope. Resolves an ancient xfail and an LLVM misparse.
llvm-svn: 91769
new InitializationSequence. This fixes some bugs (e.g., PR5808),
changed some diagnostics, and caused more churn than expected. What's
new:
- InitializationSequence now has a "C conversion sequence" category
and step kind, which falls back to
- Changed the diagnostics for returns to always have the result type
of the function first and the type of the expression second.
CheckSingleAssignmentConstraints to peform checking in C.
- Improved ASTs for initialization of return values. The ASTs now
capture all of the temporaries we need to create, but
intentionally do not bind the tempoary that is actually returned,
so that it won't get destroyed twice.
- Make sure to perform an (elidable!) copy of the class object that
is returned from a class.
- Fix copy elision in CodeGen to properly see through the
subexpressions that occur with elidable copies.
- Give "new" its own entity kind; as with return values and thrown
objects, we don't bind the expression so we don't call a
destructor for it.
Note that, with this patch, I've broken returning move-only types in
C++0x. We'll fix it later, when we tackle NRVO.
llvm-svn: 91669
than using its own partial implementation of initialization.
Switched CheckInitializerTypes over to
InitializedEntity/InitializationKind, to help move us closer to
InitializationSequence.
Added InitializedEntity::getName() to retrieve the name of the entity,
for diagnostics that care about such things.
Implemented support for default initialization in
InitializationSequence.
Clean up the determination of the "source expressions" for an
initialization sequence in InitializationSequence::Perform.
Taught CXXConstructExpr to store more location information.
llvm-svn: 91492
is difficult because they're so terribly, terribly ambiguous.
We implement access declarations in terms of using declarations, which is
quite reasonable. However, we should really persist the access/using
distinction in the AST and use the appropriate name in diagnostics. This
isn't a priority, so I'll just file a PR and hope someone else does it. :)
llvm-svn: 91095
declaration. Rename note_using_decl to note_using, which is possibly less confusing.
Add a test for non-class-scope using decl collisions and be sure to note the case
we can't diagnose yet.
llvm-svn: 91057
are a couple of O(n^2) operations in this, some analogous to the usual O(n^2)
redeclaration problem and some not. In particular, retroactively removing
shadow declarations when they're hidden by later decls is pretty unfortunate.
I'm not yet convinced it's worse than the alternative, though.
llvm-svn: 91045
new notion of an "initialization sequence", which encapsulates the
computation of the initialization sequence along with diagnostic
information and the capability to turn the computed sequence into an
expression. At present, I've only switched one CheckReferenceInit
callers over to this new mechanism; more will follow.
Aside from (hopefully) being much more true to the standard, the
diagnostics provided by this reference-initialization code are a bit
better than before. Some examples:
p5-var.cpp:54:12: error: non-const lvalue reference to type 'struct
Derived'
cannot bind to a value of unrelated type 'struct Base'
Derived &dr2 = b; // expected-error{{non-const lvalue reference to
...
^ ~
p5-var.cpp:55:9: error: binding of reference to type 'struct Base' to
a value of
type 'struct Base const' drops qualifiers
Base &br3 = bc; // expected-error{{drops qualifiers}}
^ ~~
p5-var.cpp:57:15: error: ambiguous conversion from derived class
'struct Diamond' to base class 'struct Base':
struct Diamond -> struct Derived -> struct Base
struct Diamond -> struct Derived2 -> struct Base
Base &br5 = diamond; // expected-error{{ambiguous conversion from
...
^~~~~~~
p5-var.cpp:59:9: error: non-const lvalue reference to type 'long'
cannot bind to
a value of unrelated type 'int'
long &lr = i; // expected-error{{non-const lvalue reference to type
...
^ ~
p5-var.cpp:74:9: error: non-const lvalue reference to type 'struct
Base' cannot
bind to a temporary of type 'struct Base'
Base &br1 = Base(); // expected-error{{non-const lvalue reference to
...
^ ~~~~~~
p5-var.cpp:102:9: error: non-const reference cannot bind to bit-field
'i'
int & ir1 = (ib.i); // expected-error{{non-const reference cannot
...
^ ~~~~~~
p5-var.cpp:98:7: note: bit-field is declared here
int i : 17; // expected-note{{bit-field is declared here}}
^
llvm-svn: 90992
instantiation, to ensure that we mark class template specilizations as
abstract when we need to and perform checking of abstract classes.
Also, move the checking that determines whether we are creating a
variable of abstract class type *after* we check whether the type is
complete. Otherwise, we won't see when we have an abstract class
template specialization that is implicitly instantiated by this
declaration. This is the "something else" that Sebastian had noted
earlier.
llvm-svn: 90467
overloaded-operator resolution is wildly untested, but the parallel code for
methods seems to satisfy some trivial tests.
Also change some overload-resolution APIs to take a type instead of an expression,
which lets us avoid creating a spurious CXXThisExpr when resolving implicit
member accesses.
llvm-svn: 90410
common to both parsing and template instantiation, so that we'll find
overridden virtuals for member functions of class templates when they
are instantiated.
Additionally, factor out the checking for pure virtual functions, so
that it will be executed both at parsing time and at template
instantiation time.
These changes fix PR5656 (for real), although one more tweak
w.r.t. member function templates will be coming along shortly.
llvm-svn: 90241
ValueDecl, because that isn't always the case in ill-formed
code. Diagnose a common mistake (forgetting to provide a template
argument list for a class template, PR5655) and dyn_cast so that we
handle the general problem of referring to a non-value declaration
gracefully.
llvm-svn: 90239
function templates (in C++98), friend function templates, and
out-of-line definitions of members of class templates.
Also handles merging of default template arguments from previous
declarations of function templates, for C++0x. However, we don't yet
make use of those default template arguments.
llvm-svn: 89872
All statements that involve conditions can now hold on to a separate
condition declaration (a VarDecl), and will use a DeclRefExpr
referring to that VarDecl for the condition expression. ForStmts now
have such a VarDecl (I'd missed those in previous commits).
Also, since this change reworks the Action interface for
if/while/switch/for, use FullExprArg for the full expressions in those
expressions, to ensure that we're emitting
Note that we are (still) not generating the right cleanups for
condition variables in for statements. That will be a follow-on
commit.
llvm-svn: 89817
into pretty much everything about overload resolution in order to wean
BuildDeclarationNameExpr off LookupResult::getAsSingleDecl(). Replace
UnresolvedFunctionNameExpr with UnresolvedLookupExpr, which generalizes the
idea of a non-member lookup that we haven't totally resolved yet, whether by
overloading, argument-dependent lookup, or (eventually) the presence of
a function template in the lookup results.
Incidentally fixes a problem with argument-dependent lookup where we were
still performing ADL even when the lookup results contained something from
a block scope.
Incidentally improves a diagnostic when using an ObjC ivar from a class method.
This just fell out from rewriting BuildDeclarationNameExpr's interaction with
lookup, and I'm too apathetic to break it out.
The only remaining uses of OverloadedFunctionDecl that I know of are in
TemplateName and MemberExpr.
llvm-svn: 89544
The following attributes are currently supported in C++0x attribute
lists (and in GNU ones as well):
- align() - semantics believed to be conformant to n3000, except for
redeclarations and what entities it may apply to
- final - semantics believed to be conformant to CWG issue 817's proposed
wording, except for redeclarations
- noreturn - semantics believed to be conformant to n3000, except for
redeclarations
- carries_dependency - currently ignored (this is an optimization hint)
llvm-svn: 89543
two classes, one for typenames and one for values; this seems to have some
support from Doug if not necessarily from the extremely-vague-on-this-point
standard. Track the location of the 'typename' keyword in a using-typename
decl. Make a new lookup result for unresolved values and deal with it in
most places.
llvm-svn: 89184
LookupResult RAII powers to diagnose ambiguity in the results. Other diagnostics
(e.g. access control and deprecation) will be moved to automatically trigger
during lookup as part of this same mechanism.
This abstraction makes it much easier to encapsulate aliasing declarations
(e.g. using declarations) inside the lookup system: eventually, lookup will
just produce the aliases in the LookupResult, and the standard access methods
will naturally strip the aliases off.
llvm-svn: 89027
sugared types. The basic problem is that our qualifier accessors
(getQualifiers, getCVRQualifiers, isConstQualified, etc.) only look at
the current QualType and not at any qualifiers that come from sugared
types, meaning that we won't see these qualifiers through, e.g.,
typedefs:
typedef const int CInt;
typedef CInt Self;
Self.isConstQualified() currently returns false!
Various bugs (e.g., PR5383) have cropped up all over the front end due
to such problems. I'm addressing this problem by splitting each
qualifier accessor into two versions:
- the "local" version only returns qualifiers on this particular
QualType instance
- the "normal" version that will eventually combine qualifiers from this
QualType instance with the qualifiers on the canonical type to
produce the full set of qualifiers.
This commit adds the local versions and switches a few callers from
the "normal" version (e.g., isConstQualified) over to the "local"
version (e.g., isLocalConstQualified) when that is the right thing to
do, e.g., because we're printing or serializing the qualifiers. Also,
switch a bunch of
Context.getCanonicalType(T1).getUnqualifiedType() == Context.getCanonicalType(T2).getQualifiedType()
expressions over to
Context.hasSameUnqualifiedType(T1, T2)
llvm-svn: 88969
- Also, perform calculated implicit cast sequences if they're determined to work. This finally diagnoses static_cast to ambiguous or implicit bases and fixes two long-standing fixmes in the test case. For the C-style cast, this requires propagating the access check suppression pretty deep into other functions.
- Pass the expressions for TryStaticCast and TryStaticImplicitCast by reference. This should lead to a better AST being emitted for such casts, and also fixes a memory leak, because CheckReferenceInit and PerformImplicitConversion wrap the node passed to them. These wrappers were previously lost.
llvm-svn: 88809
cast) that is converting to a class type, enumerate its constructors
as in any other direct initialization. This ensures that we get the
proper conversion sequence.
llvm-svn: 88751
like a copy constructor to the overload set, just ignore it. This
ensures that we don't try to use such a constructor as a copy
constructor *without* triggering diagnostics at the point of
declaration.
Note that we *do* diagnose such copy constructors when explicitly
written by the user (e.g., as an explicit specialization).
llvm-svn: 88733
instantiation), be sure to finish the expression statement by
providing a FullExprArg, making sure that temporaries get
destroyed. Fixes an obscure failure when parsing
llvm/LinkAllPasses.h.
llvm-svn: 88668
in "if" statements like:
if (CanQual<ReferenceType> RefType = T.getAs<ReferenceType>())
Thanks to Clang for pointing out this mistake :)
llvm-svn: 86995
handling template template parameters properly. This refactoring:
- Parses template template arguments as id-expressions, representing
the result of the parse as a template name (Action::TemplateTy)
rather than as an expression (lame!).
- Represents all parsed template arguments via a new parser-specific
type, ParsedTemplateArgument, which stores the kind of template
argument (type, non-type, template) along with all of the source
information about the template argument. This replaces an ad hoc
set of 3 vectors (one for a void*, which was either a type or an
expression; one for a bit telling whether the first was a type or
an expression; and one for a single source location pointing at
the template argument).
- Moves TemplateIdAnnotation into the new Parse/Template.h. It never
belonged in the Basic library anyway.
llvm-svn: 86708
and implicitly defined constructors. This has a number of benefits:
1. Less code.
2. Explicit and implicit constructors get the same diagnostics.
3. The AST explicitly contains constructor calls from implicit default
constructors. This allows handing some cases that previously weren't handled
correctly in IRGen without any additional code. Specifically, implicit default
constructors containing calls to constructors with default arguments are now
handled correctly.
llvm-svn: 86500
instead of all assignment operators. The mistake messes up IRGen because
it ends up assuming that the assignment operator is actually the implicit
copy assignment operator, and therefore tries to emit the RHS as an lvalue.
llvm-svn: 86307
(without complaining if it fails) to get proper semantics: reference
binding with a derived-to-base conversion and the enumeration of
constructors for user-defined conversions. There are probably more
cases to fix, but my prior attempt at statically ensuring that
complete-type checking always happens failed. Perhaps I'll try again.
With this change, Clang can parse include/llvm/*.h!
llvm-svn: 86129
representation of a C++ unqualified-id, along with a single parsing
function (Parser::ParseUnqualifiedId) that will parse all of the
various forms of unqualified-id in C++.
Replace the representation of the declarator name in Declarator with
the new UnqualifiedId class, simplifying declarator-id parsing
considerably and providing more source-location information to
Sema. In the future, I hope to migrate all of the other
unqualified-id-parsing code over to this single representation, then
begin to merge actions that are currently only different because we
didn't have a unqualified notion of the name in the parser.
llvm-svn: 85851
of ConvertDeclSpecToType, which I'd like to keep private to SemaType.cpp.
We do this by cons'ing up a trivial Declarator for the type.
John, please review.
llvm-svn: 85060
the DeclaratorInfo, one for semantic analysis), just build a single type whose
canonical type will reflect the semantic analysis (assuming the type is
well-formed, of course).
To make that work, make a few changes to the type system:
* allow the nominal pointee type of a reference type to be a (possibly sugared)
reference type. Also, preserve the original spelling of the reference type.
Both of these can be ignored on canonical reference types.
* Remove ObjCProtocolListType and preserve the associated source information on
the various ObjC TypeLocs. Preserve the spelling of protocol lists except in
the canonical form.
* Preserve some level of source type structure on parameter types, but
canonicalize on the canonical function type. This is still a WIP.
Drops code size, makes strides towards accurate source location representation,
slight (~1.7%) progression on Cocoa.h because of complexity drop.
llvm-svn: 84907
what we found when we looked into <blah>", where <blah> is a
DeclContext*. We can now format DeclContext*'s in nice ways, e.g.,
"namespace N", "the global namespace", "'class Foo'".
This is part of PR3990, but we're not quite there yet.
llvm-svn: 84028
function templates.
This commit ensures that friend function templates are constructed as
FunctionTemplateDecls rather than partial FunctionDecls (as they
previously were). It then implements template instantiation for friend
function templates, injecting the friend function template only when
no previous declaration exists at the time of instantiation.
Oh, and make sure that explicit specialization declarations are not
friends.
llvm-svn: 83970
The exception specification of the assignee must be the same or a subset of the target. In addition, exception specifications on arguments and return types must be equivalent, but this is not implemented yet.
This currently produces two diagnostics for every invalid assignment/initialization, due to the diagnostic produced outside PerformImplicitConversion, e.g. in CheckSingleInitializer. I don't know how to suppress this; in any case I think it is the wrong place for a diagnostic, since there are other diagnostics produced inside the function. So I'm leaving it as it is for the moment.
llvm-svn: 83710
for bases, members, overridden virtual methods, etc. The operations
isDerivedFrom and lookupInBases are now provided by CXXRecordDecl,
rather than by Sema, so that CodeGen and other clients can use them
directly.
llvm-svn: 83396
concrete types. Use unqualified desugaring for getAs<> and sundry.
Fix a few users to either not desugar or use qualified desugar, as seemed
appropriate. Removed Type's qualified desugar method, as it was easy
to accidentally use instead of QualType's.
llvm-svn: 83116
class templates. We now treat friend class templates much more like
normal class templates, except that they still get special name lookup
rules. Fixes PR5057 and eliminates a bunch of spurious diagnostics in
<iostream>.
llvm-svn: 82848
Type hierarchy. Demote 'volatile' to extended-qualifier status. Audit our
use of qualifiers and fix a few places that weren't dealing with qualifiers
quite right; many more remain.
llvm-svn: 82705
It uses a recent API to find inherited conversion functions to do
the initializer to reference lvalue conversion (and removes a FIXME).
It issues the ambiguity diagnostics when multiple conversions are found.
WIP.
llvm-svn: 82649
Several of the existing methods were identical to their respective
specializations, and so have been removed entirely. Several more 'leaf'
optimizations were introduced.
The getAsFoo() methods which imposed extra conditions, like
getAsObjCInterfacePointerType(), have been left in place.
llvm-svn: 82501
give them the appropriate exception specifications. This,
unfortunately, requires us to maintain and/or implicitly generate
handles to namespace "std" and the class "std::bad_alloc". However,
every other approach I've come up with was more hackish, and this
standard requirement itself is quite the hack.
Fixes PR4829.
llvm-svn: 81939
- Diagnose attempts to add default arguments to templates (or member
functions of templates) after the initial declaration (DR217).
- Improve diagnostics when a default argument is redefined. Now, the
note will always point at the place where the default argument was
previously defined, rather than pointing to the most recent
declaration of the function.
llvm-svn: 81548
such initializations properly convert constructor arguments and fill
in default arguments where necessary. This also makes the ownership
model more clear.
llvm-svn: 81394
Now that parsing, semantic analysis, and (I think) code generation of
pseudo-destructor expressions and explicit destructor calls works,
update the example-dynarray.cpp test to destroy the objects it
allocates and update the test to actually compile + link.
The code seems correct, but the Clang-compiled version dies with a
malloc error. Time to debug!
llvm-svn: 81025
declarations of same, introduce a single AST class and add appropriate bits
(encoded in the namespace) for whether a decl is "real" or not. Much hackery
about previously-declared / not-previously-declared, but it's essentially
mandated by the standard that friends alter lookup, and this is at least
fairly non-intrusive.
Refactor the Sema methods specific to friends for cleaner flow and less nesting.
Incidentally solve a few bugs, but I remain confident that we can put them back.
llvm-svn: 80353
that type. Note that we do not produce a diagnostic if the type is
incomplete; rather, we just don't look for conversion functions. Fixes PR4660.
llvm-svn: 79919
and will participate in overload resolution. Unify the instantiation
of CXXMethodDecls and CXXConstructorDecls, which had already gotten
out-of-sync.
llvm-svn: 79658
DeclaratorDecl contains a DeclaratorInfo* to keep type source info.
Subclasses of DeclaratorDecl are FieldDecl, FunctionDecl, and VarDecl.
EnumConstantDecl still inherits from ValueDecl since it has no need for DeclaratorInfo.
Decl/Sema interfaces accept a DeclaratorInfo as parameter but no DeclaratorInfo is created yet.
llvm-svn: 79392
FriendFunctionDecl, and create instances as appropriate.
The design of FriendFunctionDecl is still somewhat up in the air; you can
befriend arbitrary types of functions --- methods, constructors, etc. ---
and it's not clear that this representation captures that very well.
We'll have a better picture when we start consuming this data in access
control.
llvm-svn: 78653
Permit a class to friend its class members in C++ 98, as long as extensions
are enabled (and even when they aren't, only give an extwarn about it).
llvm-svn: 78332
Type::getAsReferenceType() -> Type::getAs<ReferenceType>()
Type::getAsRecordType() -> Type::getAs<RecordType>()
Type::getAsPointerType() -> Type::getAs<PointerType>()
Type::getAsBlockPointerType() -> Type::getAs<BlockPointerType>()
Type::getAsLValueReferenceType() -> Type::getAs<LValueReferenceType>()
Type::getAsRValueReferenceType() -> Type::getAs<RValueReferenceType>()
Type::getAsMemberPointerType() -> Type::getAs<MemberPointerType>()
Type::getAsReferenceType() -> Type::getAs<ReferenceType>()
Type::getAsTagType() -> Type::getAs<TagType>()
And remove Type::getAsReferenceType(), etc.
This change is similar to one I made a couple weeks ago, but that was partly
reverted pending some additional design discussion. With Doug's pending smart
pointer changes for Types, it seemed natural to take this approach.
llvm-svn: 77510
1. Make it work correctly with anonymous unions.
2. Don't compute it if the warning isn't enabled.
3. Optimize the algorithm slightly to make it linear time in the
case where we don't produce any warnings.
llvm-svn: 76630
until Doug Gregor's Type smart pointer code lands (or more discussion occurs).
These methods just call the new Type::getAs<XXX> methods, so we still have
reduced implementation redundancy. Having explicit getAsXXXType() methods makes
it easier to set breakpoints in the debugger.
llvm-svn: 76193
This method is intended to eventually replace the individual
Type::getAsXXXType<> methods.
The motivation behind this change is twofold:
1) Reduce redundant implementations of Type::getAsXXXType() methods. Most of
them are basically copy-and-paste.
2) By centralizing the implementation of the getAs<Type> logic we can more
smoothly move over to Doug Gregor's proposed canonical type smart pointer
scheme.
Along with this patch:
a) Removed 'Type::getAsPointerType()'; now clients use getAs<PointerType>.
b) Removed 'Type::getAsBlockPointerTypE()'; now clients use getAs<BlockPointerType>.
llvm-svn: 76098
in their order of construction for each class and use it
to to check on propery order of base class construction
under -Wreorder option.
llvm-svn: 75270
Remove ASTContext parameter from DeclContext's methods. This change cascaded down to other Decl's methods and changes to call sites started "escalating".
Timings using pre-tokenized "cocoa.h" showed only a ~1% increase in time run between and after this commit.
llvm-svn: 74506
templates.
For example, this now type-checks (but does not instantiate the body
of deref<int>):
template<typename T> T& deref(T* t) { return *t; }
void test(int *ip) {
int &ir = deref(ip);
}
Specific changes/additions:
* Template argument deduction from a call to a function template.
* Instantiation of a function template specializations (just the
declarations) from the template arguments deduced from a call.
* FunctionTemplateDecls are stored directly in declaration contexts
and found via name lookup (all forms), rather than finding the
FunctionDecl and then realizing it is a template. This is
responsible for most of the churn, since some of the core
declaration matching and lookup code assumes that all functions are
FunctionDecls.
llvm-svn: 74213
C++. This logic is required to trigger implicit instantiation of
function templates and member functions of class templates, which will
be implemented separately.
This commit includes support for -Wunused-parameter, printing warnings
for named parameters that are not used within a function/Objective-C
method/block. Fixes <rdar://problem/6505209>.
llvm-svn: 73797
parser. Rather than placing all of the delayed member function
declarations and inline definitions into a single bucket corresponding
to the top-level class, we instead mirror the nesting structure of the
nested classes and place the delayed member functions into their
appropriate place. Then, when we actually parse the delayed member
function declarations, set up the scope stack the same way as it was
when we originally saw the declaration, so that we can find, e.g.,
template parameters that are in scope.
llvm-svn: 72502
an integral constant expression, maintain a cache of the value and the
is-an-ICE flag within the VarDecl itself. This eliminates
exponential-time behavior of the Fibonacci template metaprogram.
llvm-svn: 72428
statement was using an rvalue reference during the template
definition. However, template instantiations based on an lvalue
reference type are well-formed, so we delay checking of these property
until template instantiation time.
llvm-svn: 72041
specialization" within a C++ template, and permit name lookup into the
current instantiation. For example, given:
template<typename T, typename U>
struct X {
typedef T type;
X* x1; // current instantiation
X<T, U> *x2; // current instantiation
X<U, T> *x3; // not current instantiation
::X<type, U> *x4; // current instantiation
X<typename X<type, U>::type, U>: *x5; // current instantiation
};
llvm-svn: 71471
into the left-hand side of an assignment expression. This completes
most of PR3500; the only remaining part is to deal with the
GCC-specific implementation-defined behavior for "unsigned long" (and
other) bit-fields.
llvm-svn: 70623
This gets rid of a bunch of random InvalidDecl bools in sema, changing
us to use the following approach:
1. When analyzing a declspec or declarator, if an error is found, we
set a bit in Declarator saying that it is invalid.
2. Once the Decl is created by sema, we immediately set the isInvalid
bit on it from what is in the declarator. From this point on, sema
consistently looks at and sets the bit on the decl.
This gives a very clear separation of concerns and simplifies a bunch
of code. In addition to this, this patch makes these changes:
1. it renames DeclSpec::getInvalidType() -> isInvalidType().
2. various "merge" functions no longer return bools: they just set the
invalid bit on the dest decl if invalid.
3. The ActOnTypedefDeclarator/ActOnFunctionDeclarator/ActOnVariableDeclarator
methods now set invalid on the decl returned instead of returning an
invalid bit byref.
4. In SemaType, refering to a typedef that was invalid now propagates the
bit into the resultant type. Stuff declared with the invalid typedef
will now be marked invalid.
5. Various methods like CheckVariableDeclaration now return void and set the
invalid bit on the decl they check.
There are a few minor changes to tests with this, but the only major bad
result is test/SemaCXX/constructor-recovery.cpp. I'll take a look at this
next.
llvm-svn: 70020
Remove an atrocious amount of trailing whitespace in the overloaded operator mangler. Sorry, couldn't help myself.
Change the DeclType parameter of Sema::CheckReferenceInit to be passed by value instead of reference. It wasn't changed anywhere.
Let the parser handle C++'s irregular grammar around assignment-expression and conditional-expression.
And finally, the reason for all this stuff: implement C++ semantics for the conditional operator. The implementation is complete except for determining lvalueness.
llvm-svn: 69299
Implement the rvalue reference overload dance for returning local objects. Returning a local object first tries to find a move constructor now.
The error message when no move constructor is defined (or is not applicable) and the copy constructor is deleted is quite ugly, though.
llvm-svn: 68902
productions (except the already broken ObjC cases like @class X,Y;) in
the parser that can produce more than one Decl return a DeclGroup instead
of a Decl, etc.
This allows elimination of the Decl::NextDeclarator field, and exposes
various clients that should look at all decls in a group, but which were
only looking at one (such as the dumper, printer, etc). These have been
fixed.
Still TODO:
1) there are some FIXME's in the code about potentially using
DeclGroup for better location info.
2) ParseObjCAtDirectives should return a DeclGroup due to @class etc.
3) I'm not sure what is going on with StmtIterator.cpp, or if it can
be radically simplified now.
4) I put a truly horrible hack in ParseTemplate.cpp.
I plan to bring up #3/4 on the mailing list, but don't plan to tackle
#1/2 in the short term.
llvm-svn: 68002
pointer. Its purpose in life is to be a glorified void*, but which does not
implicitly convert to void* or other OpaquePtr's with a different UID.
Introduce Action::DeclPtrTy which is a typedef for OpaquePtr<0>. Change the
entire parser/sema interface to use DeclPtrTy instead of DeclTy*. This
makes the C++ compiler enforce that these aren't convertible to other opaque
types.
We should also convert ExprTy, StmtTy, TypeTy, AttrTy, BaseTy, etc,
but I don't plan to do that in the short term.
The one outstanding known problem with this patch is that we lose the
bitmangling optimization where ActionResult<DeclPtrTy> doesn't know how to
bitmangle the success bit into the low bit of DeclPtrTy. I will rectify
this with a subsequent patch.
llvm-svn: 67952
uniqued representation that should both save some memory and make it
far easier to properly build canonical types for types involving
dependent nested-name-specifiers, e.g., "typename T::Nested::type".
This approach will greatly simplify the representation of
CXXScopeSpec. That'll be next.
llvm-svn: 67799
const DiagnosticBuilder &operator<<(const DiagnosticBuilder &DB,
AccessSpecifier AS);
so we can easily add access specifiers to diagnostics.
llvm-svn: 67795
class C {
C() { }
int a;
};
C::C() : a(10) { }
We also diagnose when initializers are used on declarations that aren't constructors:
t.cpp:1:10: error: only constructors take base initializers
void f() : a(10) { }
^
Doug and/or Sebastian: I'd appreciate a review, especially the nested-name-spec test results (from the looks of it we now match gcc in that test.)
llvm-svn: 67672
class C {
void g(C c);
virtual void f() = 0;
};
In this case, C is not known to be abstract when doing semantic analysis on g. This is done by recursively traversing the abstract class and checking the types of member functions.
llvm-svn: 67594
qualified name, e.g.,
foo::x
so that we retain the nested-name-specifier as written in the source
code and can reproduce that qualified name when printing the types
back (e.g., in diagnostics). This is PR3493, which won't be complete
until finished the other tasks mentioned near the end of this commit.
The parser's representation of nested-name-specifiers, CXXScopeSpec,
is now a bit fatter, because it needs to contain the scopes that
precede each '::' and keep track of whether the global scoping
operator '::' was at the beginning. For example, we need to keep track
of the leading '::', 'foo', and 'bar' in
::foo::bar::x
The Action's CXXScopeTy * is no longer a DeclContext *. It's now the
opaque version of the new NestedNameSpecifier, which contains a single
component of a nested-name-specifier (either a DeclContext * or a Type
*, bitmangled).
The new sugar type QualifiedNameType composes a sequence of
NestedNameSpecifiers with a representation of the type we're actually
referring to. At present, we only build QualifiedNameType nodes within
Sema::getTypeName. This will be extended to other type-constructing
actions (e.g., ActOnClassTemplateId).
Also on the way: QualifiedDeclRefExprs will also store a sequence of
NestedNameSpecifiers, so that we can print out the property
nested-name-specifier. I expect to also use this for handling
dependent names like Fibonacci<I - 1>::value.
llvm-svn: 67265
Type pointer. This allows our nested-name-specifiers to retain more
information about the actual spelling (e.g., which typedef did the
user name, or what exact template arguments were used in the
template-id?). It will also allow us to have dependent
nested-name-specifiers that don't map to any DeclContext.
llvm-svn: 67140
instantiation for binary operators. This change moves most of the
operator-overloading code from the parser action ActOnBinOp to a new,
parser-independent semantic checking routine CreateOverloadedBinOp.
Of particular importance is the fact that CreateOverloadedBinOp does
*not* perform any name lookup based on the current parsing context (it
doesn't take a Scope*), since it has to be usable during template
instantiation, when there is no scope information. Rather, it takes a
pre-computed set of functions that are visible from the context or via
argument-dependent lookup, and adds to that set any member operators
and built-in operator candidates. The set of functions is computed in
the parser action ActOnBinOp based on the current context (both
operator name lookup and argument-dependent lookup). Within a
template, the set computed by ActOnBinOp is saved within the
type-dependent AST node and is augmented with the results of
argument-dependent name lookup at instantiation time (see
TemplateExprInstantiator::VisitCXXOperatorCallExpr).
Sadly, we can't fully test this yet. I'll follow up with template
instantiation for sizeof so that the real fun can begin.
llvm-svn: 66923
(default: 99). Beyond this limit, produce an error and consider the
current template instantiation a failure.
The stack we're building to track the instantiations will, eventually,
be used to produce instantiation backtraces from diagnostics within
template instantiation. However, we're not quite there yet.
This adds a new Clang driver option -ftemplate-depth=NNN, which should
eventually be generated from the GCC command-line operation
-ftemplate-depth-NNN (note the '-' rather than the '='!). I did not
make the driver changes to do this mapping.
llvm-svn: 66513
such as replacing 'T' in vector<T>. There are a few aspects to this:
- Extend TemplateArgument to allow arbitrary expressions (an
Expr*), and switch ClassTemplateSpecializationType to store
TemplateArguments rather than it's own type-or-expression
representation.
- ClassTemplateSpecializationType can now store dependent types. In
that case, the canonical type is another
ClassTemplateSpecializationType (with default template arguments
expanded) rather than a declaration (we don't build Decls for
dependent types).
- Split ActOnClassTemplateId into ActOnClassTemplateId (called from
the parser) and CheckClassTemplateId (called from
ActOnClassTemplateId and InstantiateType). They're smart enough to
handle dependent types, now.
llvm-svn: 66509
response to attempts to diagnose an "incomplete" type. This will force
us to use DiagnoseIncompleteType more regularly (rather than looking at
isIncompleteType), but that's also a good thing.
Implicit instantiation is still very simplistic, and will create a new
definition for the class template specialization (as it should) but it
only actually instantiates the base classes and attaches
those. Actually instantiating class members will follow.
Also, instantiate the types of non-type template parameters before
checking them, allowing, e.g.,
template<typename T, T Value> struct Constant;
to work properly.
llvm-svn: 65924
giving them rough classifications (normal types, never-canonical
types, always-dependent types, abstract type representations) and
making it far easier to make sure that we've hit all of the cases when
decoding types.
Switched some switch() statements on the type class over to using this
mechanism, and filtering out those things we don't care about. For
example, CodeGen should never see always-dependent or non-canonical
types, while debug info generation should never see always-dependent
types. More switch() statements on the type class need to be moved
over to using this approach, so that we'll get warnings when we add a
new type then fail to account for it somewhere in the compiler.
As part of this, some types have been renamed:
TypeOfExpr -> TypeOfExprType
FunctionTypeProto -> FunctionProtoType
FunctionTypeNoProto -> FunctionNoProtoType
There shouldn't be any functionality change...
llvm-svn: 65591
(as GCC does), except when we've performed overload resolution and
found an unavailable function: in this case, we actually error.
Merge the checking of unavailable functions with the checking for
deprecated functions. This unifies a bit of code, and makes sure that
we're checking for unavailable functions in the right places. Also,
this check can cause an error. We may, eventually, want an option to
make "unavailable" warnings into errors.
Implement much of the logic needed for C++0x deleted functions, which
are effectively the same as "unavailable" functions (but always cause
an error when referenced). However, we don't have the syntax to
specify deleted functions yet :)
llvm-svn: 64955
- If a declaration is an invalid redeclaration of an existing name,
complain about the invalid redeclaration then avoid adding it to
the AST (we can still parse the definition or initializer, if any).
- If the declaration is invalid but there is no prior declaration
with that name, introduce the invalid declaration into the AST
(for later error recovery).
- If the declaration is an invalid redeclaration of a builtin that
starts with __builtin_, we produce an error and drop the
redeclaration. If it is an invalid redeclaration of a library
builtin (e.g., malloc, printf), warn (don't error!) and drop the
redeclaration.
If a user attempts to define a builtin, produce an error and (if it's
a library builtin like malloc) suggest -ffreestanding.
This addresses <rdar://problem/6097585> and PR2892. However, PR3588 is
still going to cause some problems when builtins are redeclared
without a prototype.
llvm-svn: 64639
than a Decl, which gives us some more flexibility to express the
results with the type system. There are no clients using this
flexibility yet, but it's meant to be able to describe qualified names
as written in the source (e.g., "foo::type") or template-ids that name
a class template specialization (e.g., "std::vector<INT>").
DeclSpec's TST_typedef has become TST_typename, to reflect its use to
describe types found by name (that may or may not be typedefs). The
type representation of a DeclSpec with TST_typename is an opaque
QualType pointer. All users of TST_typedef, both direct and indirect,
have been updated for these changes.
llvm-svn: 64141
- Made allocation of Stmt objects using vanilla new/delete a *compiler
error* by making this new/delete "protected" within class Stmt.
- Now the only way to allocate Stmt objects is by using the new
operator that takes ASTContext& as an argument. This ensures that
all Stmt nodes are allocated from the same (pool) allocator.
- Naturally, these two changes required that *all* creation sites for
AST nodes use new (ASTContext&). This is a large patch, but the
majority of the changes are just this mechanical adjustment.
- The above changes also mean that AST nodes can no longer be
deallocated using 'delete'. Instead, one most do
StmtObject->Destroy(ASTContext&) or do
ASTContextObject.Deallocate(StmtObject) (the latter not running the
'Destroy' method).
Along the way I also...
- Made CompoundStmt allocate its array of Stmt* using the allocator in
ASTContext (previously it used std::vector). There are a whole
bunch of other Stmt classes that need to be similarly changed to
ensure that all memory allocated for ASTs comes from the allocator
in ASTContext.
- Added a new smart pointer ExprOwningPtr to Sema.h. This replaces
the uses of llvm::OwningPtr within Sema, as llvm::OwningPtr used
'delete' to free memory instead of a Stmt's 'Destroy' method.
Big thanks to Doug Gregor for helping with the acrobatics of making
'new/delete' private and the new smart pointer ExprOwningPtr!
llvm-svn: 63997
- Changes Lookup*Name functions to return NamedDecls, instead of
Decls. Unfortunately my recent statement that it will simplify lot of
code, was not quite right, but it simplifies some...
- Makes MergeLookupResult SmallPtrSet instead of vector, following
Douglas suggestions.
- Adds %qN format for printing qualified names to Diagnostic.
- Avoids searching for using-directives in Scopes, which are not
DeclScope, during unqualified name lookup.
llvm-svn: 63739
sequence. Previously, we weren't permitting the second step to call
copy constructors, which left user-defined conversion sequences
surprisingly broken.
Now, we perform overload resolution among all of the constructors, but
only accept the result if it makes the conversion a standard
conversion. Note that this behavior is different from both GCC and EDG
(which don't agree with each other, either); I've submitted a core
issue on the matter.
llvm-svn: 63450
LookupName et al. Instead, use an enum and a bool to describe its
contents.
Optimized the C/Objective-C path through LookupName, eliminating any
unnecessarily C++isms. Simplify IdentifierResolver::iterator, removing
some code and arguments that are no longer used.
Eliminated LookupDeclInScope/LookupDeclInContext, moving all callers
over to LookupName, LookupQualifiedName, or LookupParsedName, as
appropriate.
All together, I'm seeing a 0.2% speedup on Cocoa.h with PTH and
-disable-free. Plus, we're down to three name-lookup routines.
llvm-svn: 63354
The previous interface was very confusing. This is much more explicit, which will be easier to understand/optimize/convert.
The plan is to eventually deprecate both of these functions. For now, I'm focused on performance.
llvm-svn: 63256
Even though Sema::LookupDecl() is deprecated, it's still used all over the place. Simplifying the interface will make it easier to understand/optimize/convert.
llvm-svn: 63208
.def file for each library. This means that adding a diagnostic
to sema doesn't require all the other libraries to be rebuilt.
Patch by Anders Johnsen!
llvm-svn: 63111
that every declaration lives inside a DeclContext.
Moved several things that don't have names but were ScopedDecls (and,
therefore, NamedDecls) to inherit from Decl rather than NamedDecl,
including ObjCImplementationDecl and LinkageSpecDecl. Now, we don't
store empty DeclarationNames for these things, nor do we try to insert
them into DeclContext's lookup structure.
The serialization tests are temporarily disabled. We'll re-enable them
once we've sorted out the remaining ownership/serialiazation issues
between DeclContexts and TranslationUnion, DeclGroups, etc.
llvm-svn: 62562
new DiagnoseIncompleteType. It provides additional information about
struct/class/union/enum types when possible, either by pointing to the
forward declaration of that type or by pointing to the definition (if
we're in the process of defining that type).
Fixes <rdar://problem/6500531>.
llvm-svn: 62521
This change refactors and cleans up our handling of name lookup with
LookupDecl. There are several aspects to this refactoring:
- The criteria for name lookup is now encapsulated into the class
LookupCriteria, which replaces the hideous set of boolean values
that LookupDecl currently has.
- The results of name lookup are returned in a new class
LookupResult, which can lazily build OverloadedFunctionDecls for
overloaded function sets (and, eventually, eliminate the need to
allocate member for OverloadedFunctionDecls) and contains a
placeholder for handling ambiguous name lookup (for C++).
- The primary entry points for name lookup are now LookupName (for
unqualified name lookup) and LookupQualifiedName (for qualified
name lookup). There is also a convenience function
LookupParsedName that handles qualified/unqualified name lookup
when given a scope specifier. Together, these routines are meant
to gradually replace the kludgy LookupDecl, but this won't happen
until after we have base class lookup (which forces us to cope
with ambiguities).
- Documented the heck out of name lookup. Experimenting a little
with using Doxygen's member groups to make some sense of the Sema
class. Feedback welcome!
- Fixes some lingering issues with name lookup for
nested-name-specifiers, which now goes through
LookupName/LookupQualifiedName.
llvm-svn: 62245
Small cleanup in the handling of user-defined conversions.
Also, implement an optimization when constructing a call. We avoid
recomputing implicit conversion sequences and instead use those
conversion sequences that we computed as part of overload resolution.
llvm-svn: 62231
introduce a Scope for the body of a tag. This reduces the number of
semantic differences between C and C++ structs and unions, and will
help with other features (e.g., anonymous unions) in C. Some important
points:
- Fields are now in the "member" namespace (IDNS_Member), to keep
them separate from tags and ordinary names in C. See the new test
in Sema/member-reference.c for an example of why this matters. In
C++, ordinary and member name lookup will find members in both the
ordinary and member namespace, so the difference between
IDNS_Member and IDNS_Ordinary is erased by Sema::LookupDecl (but
only in C++!).
- We always introduce a Scope and push a DeclContext when we're
defining a tag, in both C and C++. Previously, we had different
actions and different Scope/CurContext behavior for enums, C
structs/unions, and C++ structs/unions/classes. Now, it's one pair
of actions. (Yay!)
There's still some fuzziness in the handling of struct/union/enum
definitions within other struct/union/enum definitions in C. We'll
need to do some more cleanup to eliminate some reliance on CurContext
before we can solve this issue for real. What we want is for something
like this:
struct X {
struct T { int x; } t;
};
to introduce T into translation unit scope (placing it at the
appropriate point in the IdentifierResolver chain, too), but it should
still have struct X as its lexical declaration
context. PushOnScopeChains isn't smart enough to do that yet, though,
so there's a FIXME test in nested-redef.c
llvm-svn: 61940
- ObjCContainerDecl's (ObjCInterfaceDecl/ObjCCategoryDecl/ObjCProtocolDecl), ObjCCategoryImpl, & ObjCImplementation are all DeclContexts.
- ObjCMethodDecl is now a ScopedDecl (so it can play nicely with DeclContext).
- ObjCContainerDecl now does iteration/lookup using DeclContext infrastructure (no more linear search:-)
- Removed ASTContext argument to DeclContext::lookup(). It wasn't being used and complicated it's use from an ObjC AST perspective.
- Added Sema::ProcessPropertyDecl() and removed Sema::diagnosePropertySetterGetterMismatch().
- Simplified Sema::ActOnAtEnd() considerably. Still more work to do.
- Fixed an incorrect casting assumption in Sema::getCurFunctionOrMethodDecl(), now that ObjCMethodDecl is a ScopedDecl.
- Removed addPropertyMethods from ObjCInterfaceDecl/ObjCCategoryDecl/ObjCProtocolDecl.
This passes all the tests on my machine. Since many of the changes are central to the way ObjC finds it's methods, I expect some fallout (and there are still a handful of FIXME's). Nevertheless, this should be a step in the right direction.
llvm-svn: 61929
Eli Friedman