using new/delete and OwningPtrs. After memory profiling Clang, I witnessed periodic leaks of these
objects; digging deeper into the code, it was clear that our management of these objects was a mess. The ownership rules were murky at best, and not always followed. Worse, there are plenty of error paths where we could screw up.
This patch introduces AttributeList::Factory, which is a factory class that creates AttributeList
objects and then blows them away all at once. While conceptually simple, most of the changes in
this patch just have to do with migrating over to the new interface. Most of the changes have resulted in some nice simplifications.
This new strategy currently holds on to all AttributeList objects during the lifetime of the Parser
object. This is easily tunable. If we desire to have more bound the lifetime of AttributeList
objects more precisely, we can have the AttributeList::Factory object (in Parser) push/pop its
underlying allocator as we enter/leave key methods in the Parser. This means that we get
simple memory management while still having the ability to finely control memory use if necessary.
Note that because AttributeList objects are now BumpPtrAllocated, we may reduce malloc() traffic
in many large files with attributes.
This fixes the leak reported in: <rdar://problem/8650003>
llvm-svn: 118675
construct an unsupported friend when there's a friend with a templated
scope specifier. Fixes a consistency crash, rdar://problem/8540527
llvm-svn: 116786
with comma-separated lists. We never actually used the comma
locations, nor did we store them in the AST, but we did manage to
waste time during template instantiation to produce fake locations.
llvm-svn: 113495
One who seeks the Tao unlearns something new every day.
Less and less remains until you arrive at non-action.
When you arrive at non-action,
nothing will be left undone.
llvm-svn: 112244
- move DeclSpec &c into the Sema library
- move ParseAST into the Parse library
Reflect this change in a thousand different includes.
Reflect this change in the link orders.
llvm-svn: 111667
-There are 2 instances that change the TokenID for GNU libstdc++ 4.2 compatibility.
To handler those cases introduce a RevertedTokenID bitfield, RevertTokenIDToIdentifier() and hasRevertedTokenIDToIdentifier() methods.
Store the bitfield in PCH.
llvm-svn: 110868
allows Sema some limited access to the current scope, which we only
use in one way: when Sema is performing some kind of declaration that
is not directly driven by the parser (e.g., due to template
instantiatio or lazy declaration of a member), we can find the Scope
associated with a DeclContext, if that DeclContext is still in the
process of being parsed.
Use this to make the implicit declaration of special member functions
in a C++ class more "scope-less", rather than using the NULL Scope hack.
llvm-svn: 107491
This is more targeted, as it simply provides toggle actions for the parser to
turn access checking on and off. We then use these to suppress access checking
only while we parse the template-id (included scope specifier) of an explicit
instantiation and explicit specialization of a class template. The
specialization behavior is an extension, as it seems likely a defect that the
standard did not exempt them as it does explicit instantiations.
This allows the very common practice of specializing trait classes to work for
private, internal types. This doesn't address instantiating or specializing
function templates, although those apparently already partially work.
The naming and style for the Action layer isn't my favorite, comments and
suggestions would be appreciated there.
llvm-svn: 106993
just skip over the body of the class or class template: it's a
semantic disaster that's likely to cause invariants to break. Fixes
part of <rdar://problem/8104754>.
llvm-svn: 106496
"previous token" location at the end of the class definition. This
eliminates a badly-placed error + Fix-It when the ';' following a
class definition is missing. Fixes <rdar://problem/8066414>.
llvm-svn: 106175
1) Suppress diagnostics as soon as we form the code-completion
token, so we don't get any error/warning spew from the early
end-of-file.
2) If we consume a code-completion token when we weren't expecting
one, go into a code-completion recovery path that produces the best
results it can based on the context that the parser is in.
llvm-svn: 104585
for the purposes of parsing default arguments. In effect, we would
re-introduce the parameter with a default argument N times (where N is
the number of parameters preceding the parameter with a default
argument). This showed up when a defaulted parameter of a member
function of a local class shadowed a parameter of the enclosing
function. Fixes PR6383.
llvm-svn: 97534
*not* entering the context of the nested-name-specifier. This was
causing us to look into an uninstantiated template that we shouldn't
look into. Fixes PR6376.
llvm-svn: 97524
an *almost* always incorrect case. This only does the lookahead
in the insanely unlikely case, so it shouldn't impact performance.
On this testcase:
struct foo {
}
typedef int x;
Before:
t.c:3:9: error: cannot combine with previous 'struct' declaration specifier
typedef int x;
^
After:
t.c:2:2: error: expected ';' after struct
}
^
;
llvm-svn: 97403
propagating error conditions out of the various annotate-me-a-snowflake
routines. Generally (but not universally) removes redundant diagnostics
as well as, you know, not crashing on bad code. On the other hand,
I have just signed myself up to fix fiddly parser errors for the next
week. Again.
llvm-svn: 97221
now cope with the destruction of types named as dependent templates,
e.g.,
y->template Y<T>::~Y()
Nominally, we implement C++0x [basic.lookup.qual]p6. However, we don't
follow the letter of the standard here because that would fail to
parse
template<typename T, typename U>
X0<T, U>::~X0() { }
properly. The problem is captured in core issue 339, which gives some
(but not enough!) guidance. I expect to revisit this code when the
resolution of 339 is clear, and/or we start capturing better source
information for DeclarationNames.
Fixes PR6152.
llvm-svn: 96367
we would just leak them all over the place, with no clear ownership of
these objects at all. AttributeList objects would get leaked on both
error and non-error paths.
Note: I introduced the usage of llvm::OwningPtr<AttributeList> to
manage these objects, which is particularly useful for methods with
multiple return sites. In at least one method I used them even when
they weren't strictly necessary because it clarified the ownership
semantics and made the code easier to read. Should the excessive
'take()' and 'reset()' calls become a performance issue we can always
re-evaluate.
Note+1: I believe I have not introduced any double-frees, but it would
be nice for someone to review this.
This fixes <rdar://problem/7635046>.
llvm-svn: 95847
forgetting a ';' at the end of a struct. For something like:
class c {
}
void foo() {}
we now produce:
t.cc:3:2: error: expected ';' after class
}
^
;
instead of:
t.cc:4:1: error: cannot combine with previous 'class' declaration specifier
void foo() {}
^
t.cc:2:7: error: 'class c' can not be defined in the result type of a function
class c {
^
GCC produces:
t.cc:4: error: new types may not be defined in a return type
t.cc:4: note: (perhaps a semicolon is missing after the definition of ‘c’)
t.cc:4: error: two or more data types in declaration of ‘foo’
I *think* I got the follow set right, but if I forgot anything, we'll start
getting spurious "expected ';' after class" errors, let me know if you see
any.
llvm-svn: 95042
t.cc:4:3: error: expected ';' at end of declaration list
int y;
^
t.cc:6:1: error: expected ';' at end of declaration list
};
^
After:
t.cc:3:8: error: expected ';' at end of declaration list
int x
^
;
t.cc:5:8: error: expected ';' at end of declaration list
int z
^
;
llvm-svn: 95039
(necessarily simultaneous) changes:
- CXXBaseOrMemberInitializer now contains only a single initializer
rather than a set of initialiation arguments + a constructor. The
single initializer covers all aspects of initialization, including
constructor calls as necessary but also cleanup of temporaries
created by the initializer (which we never handled
before!).
- Rework + simplify code generation for CXXBaseOrMemberInitializers,
since we can now just emit the initializer as an initializer.
- Switched base and member initialization over to the new
initialization code (InitializationSequence), so that it
- Improved diagnostics for the new initialization code when
initializing bases and members, to match the diagnostics produced
by the previous (special-purpose) code.
- Simplify the representation of type-checked constructor initializers in
templates; instead of keeping the fully-type-checked AST, which is
rather hard to undo at template instantiation time, throw away the
type-checked AST and store the raw expressions in the AST. This
simplifies instantiation, but loses a little but of information in
the AST.
- When type-checking implicit base or member initializers within a
dependent context, don't add the generated initializers into the
AST, because they'll look like they were explicit.
- Record in CXXConstructExpr when the constructor call is to
initialize a base class, so that CodeGen does not have to infer it
from context. This ensures that we call the right kind of
constructor.
There are also a few "opportunity" fixes here that were needed to not
regress, for example:
- Diagnose default-initialization of a const-qualified class that
does not have a user-declared default constructor. We had this
diagnostic specifically for bases and members, but missed it for
variables. That's fixed now.
- When defining the implicit constructors, destructor, and
copy-assignment operator, set the CurContext to that constructor
when we're defining the body.
llvm-svn: 94952
the tag kind (union, struct, class, enum) over to the name of the tag,
if there is a name, since most clients want to point at the name.
llvm-svn: 94424
distinguish between nested classes (whose member functions cannot be
parsed until the innermost non-nested class is complete) and local
classes (that are defined within a function but are not necessarily
nested). The upshot of this change, which fixes PR5764, is that the
bodies of member functions of local (non-nested) classes need to be
parsed when the local class is complete (and no later), since they may
refer to function-local static variables, typedefs, enums, etc.
llvm-svn: 93653
name a template, when they occur in a base-specifier. This is one of
the (few) places where we know for sure that an identifier followed by
a '<' must be a template name, so we can diagnose and recover well:
test/SemaTemplate/dependent-base-classes.cpp:9:16: error: missing
'template'
keyword prior to dependent template name 'T::apply'
struct X1 : T::apply<U> { }; // expected-error{{missing 'template' ...
^
template
test/SemaTemplate/dependent-base-classes.cpp:12:13: error: unknown
template name
'vector'
struct X2 : vector<T> { }; // expected-error{{unknown template name
'vector'}}
^
2 diagnostics generated.
llvm-svn: 93257
initializers. This isn't actually in the C++ grammar (in any version),
but that's clearly an oversight: both GCC and EDG support this syntax,
and it's used within Boost code. I'll file a core issue proposing
precisely the change made here. Fixes PR6008.
llvm-svn: 93243
Because of the rules of base-class lookup* and the restrictions on typedefs, it
was actually impossible for this to cause any problems more serious than the
spurious acceptance of
template <class T> class A : B<A> { ... };
instead of
template <class T> class A : B<A<T> > { ... };
but I'm sure we can all agree that that is a very important restriction which
is well worth making another Parser->Sema call for.
(*) n.b. clang++ does not implement these rules correctly; we are not ignoring
non-type names
llvm-svn: 91792
Anders Carlsson