#1: It is cleaner. I never "liked" storing keyword selectors (i.e. foo:bar:baz) in the IdentifierTable.
#2: It is more space efficient. Since Cocoa keyword selectors can be quite long, this technique is space saving. For Cocoa.h, pulling the keyword selectors out saves ~180k. The cost of the SelectorInfo data is ~100k. Saves ~80k, or 43%.
#3: It results in many API simplifications. Here are some highlights:
- Removed 3 actions (ActOnKeywordMessage, ActOnUnaryMessage, & one flavor of ObjcBuildMethodDeclaration that was specific to unary messages).
- Removed 3 funky structs from DeclSpec.h (ObjcKeywordMessage, ObjcKeywordDecl, and ObjcKeywordInfo).
- Removed 2 ivars and 2 constructors from ObjCMessageExpr (fyi, this space savings has not been measured).
I am happy with the way it turned out (though it took a bit more hacking than I expected). Given the central role of selectors in ObjC, making sure this is "right" will pay dividends later.
Thanks to Chris for talking this through with me and suggesting this approach.
llvm-svn: 42395
1. Handles saving and checking on protocols used in an @interface declaration
2. Checks and saves class's super class.
3. Adds semantic check to category declarations.
llvm-svn: 42218
Rationale:
We currently have a separate table to unique ObjC selectors. Since I don't need all the instance data in IdentifierInfo, I thought this would save space (and make more sense conceptually).
It turns out the cost of having duplicate entries for unary selectors (i.e. names without colons) outweighs the cost difference between the IdentifierInfo & SelectorInfo structures. Here is the data:
Two tables:
*** Selector/Identifier Stats:
# Selectors/Identifiers: 51635
Bytes allocated: 1999824
One table:
*** Identifier Table Stats:
# Identifiers: 49500
Bytes allocated: 1990316
llvm-svn: 42139
- Add ObjcMessageExpr AST node and associated constructors.
- Add SourceLocation's to ActOnKeywordMessage/ActOnUnaryMessage API.
- Instantiate message expressions...
- Replace alloca usage with SmallString.
Next step, installing a correct type, among other tweaks...
llvm-svn: 42116
- Removed helper ObjcGetSelectorInfo(), moving the code directly into ObjcBuildMethodDeclaration().
- Many refinements to ParseObjCMessageExpression().
- Add ActOnMessageExpression().
Next step, finish the message actions and (finally) create/instantiate an ObjcMessageExpr AST.
llvm-svn: 42050
- Add SelectorInfo/SelectorTable classes, modeled after IdentifierInfo/IdentifierTable.
- Add SelectorTable instance to ASTContext, created lazily through ASTContext::getSelectorInfo().
- Add SelectorInfo slot to ObjcMethodDecl.
- Add helper function to derive a SelectorInfo from ObjcKeywordInfo.
Misc: Got the Decl stats stuff up and running again...it was missing support for ObjC AST's.
llvm-svn: 42023
- Allow classnames as the receiver (removing a FIXME from ParseObjCMessageExpression).
- Added a FIXME to ParseObjCMessageExpression()...we need to return a message expr AST node!
llvm-svn: 42001
The previous naming scheme was confusing, since it resulted in both the Parser and Action modules having methods with the same name. In addition, the Action module never does any parsing...
llvm-svn: 41986
Remove Action::ObjcAddVisibilityToIvars(). No need for an extra API when it is trivial to add this info to the previous hook.
In general, I want to start migrating away from having Actions prefixed with "Parse" (which is confusing, since the Action API doesn't do any parsing, per se).
llvm-svn: 41973
- Adding a safer prologue. The previous prologue would accept a null and therefore assume we had an interface (which was incorrect).
- Fixed FieldDecl's classof method. This allowed me to simplify some unnecessary casting.
- When diagnosing errors, make sure the FieldDecl/EnclosingDecl are marked as invalid. In addition, don't delete the field...rather, add all fields to the enclosing decl. Memory management can/should be done elsewhere. This code was never "upgraded" to the recently added invalid decl strategy.
llvm-svn: 41964
Move Identifier/Loc instance variables (and associated getters/setters) down from Decl to ScopedDecl/FieldDecl.
Objc AST's can now inherit from Decl without getting instance variables and types that are C specific. For now, I am keeping NextDeclarator, since I believe it may be useful to ObjC. If not, it can be moved later.
llvm-svn: 41934
void func() {
int xx = xx; // incorrectly diagnosed 'xx' as an undeclared identifier.
}
This smallish bug resulted in a largish fix. Here are some highlights:
- Needed to make sure ParseDeclarator is called *before* parsing any
initializer. Removed the "Init" argument to ParseDeclarator.
- Added AddInitializerToDecl() to the Action & Sema classes.
In Sema, this hook is responsible for validating the initializer and
installing it into the respective decl.
- Moved several semantic checks from ParseDeclarator() to
FinalizeDeclaratorGroup(). Previously, this hook was only responsible for
reversing a list. Now it plays a much larger semantic role.
All of the above changes ended up simplifying ParseDeclarator(), which
is goodness...
llvm-svn: 41877
- Instantiate the node in Sema::ParseField(), based on the type of the TagDecl.
- Add Sema::ObjcAddInstanceVariable(), responsible for adorning/adding the ObjcIvarDecl.
llvm-svn: 41864
as types. That said, the AST nodes ObjcInterfaceDecl, ObjcInterfaceType, and ObjcClassDecl are *very*
preliminary.
The good news is we no longer need -parse-noop (aka MinimalActions) to parse cocoa.m.
llvm-svn: 41752
- ArrayType::getBaseType(), and
- ConstantArrayType::getMaximumElements().
Wanted to do this cleanup before adding structure support, which will add more complexity.
llvm-svn: 41715
- Fixed many bugs, enhanced test case considerably, added a diagnostic, etc.
- Refactored CheckInitList() into CheckVariableInitList()/CheckConstantInitList().
- Added CheckInitExpr().
- Support for multi-dimensional arrays looking good.
llvm-svn: 41690
- Added Expr::isConstantExpr().
- Added type checking for InitListExpr elements.
- Added diagnostic for trying to initialize a variable sized object.
llvm-svn: 41674
Step 1: Start instantiating InitListExpr's.
Step 2: Call newly added function Sema::CheckInitializer() from Sema::ParseDeclarator().
Step 3: Give InitListExpr's a preliminary type.
Step 4: Start emitting diagnostics for simple assignments.
Note:
As a result of step 1, the CodeGen/mandel.c test asserts "Unimplemented agg expr!", which is expected.
As a result of step 4, the test below now fails. This isn't expected and needs to be investigated (it appears type checking for C++ references is flawed in some way).
******************** TEST 'Sema/cxx-references.cpp' FAILED! ********************
Command:
clang -fsyntax-only Sema/cxx-references.cpp
Output:
Sema/cxx-references.cpp:8:12: warning: incompatible pointer types assigning 'int &*' to 'int *'
int *p = &r;
^~
Sema/cxx-references.cpp:10:20: error: incompatible types assigning 'int (int)' to 'int (&)(int)'
int (&rg)(int) = g;
^
Sema/cxx-references.cpp:13:18: error: incompatible types assigning 'int [3]' to 'int (&)[3]'
int (&ra)[3] = a;
^
Sema/cxx-references.cpp:16:14: error: incompatible types assigning 'int *' to 'int *&'
int *& P = Q;
^
4 diagnostics generated.
******************** TEST 'Sema/cxx-references.cpp' FAILED! ********************
llvm-svn: 41671
routine was causing more trouble than it was worth. Anders/Chris noticed that it could return an error code
without emiting a diagnostic (which results in an silent invalid decl, which should *never* happen). In addition,
this routine didn't work well for typedefs and field decls. Lastly, it didn't consider that initializers aren't
in place yet.
Added Type::getAsConstantArrayType(), Type::getAsVariableArrayType(), Type::getAsVariablyModifiedType(),
and Type::isVariablyModifiedType();
Modified Sema::ParseDeclarator() and Sema::ParseField() to use the new predicates. Also added a FIXME for
the initializer omission. Also added a missing test for "static" @ file scope.
llvm-svn: 41647
Moved several array constraints checks from Sema::VerifyConstantArrayType() to
Sema::GetTypeForDeclarator(). VerifyConstantArrayType() is now very simple, and
could be removed eventually.
Now, we get the following (correct) messages for BlockVarDecls:-)
[dylan:~/llvm/tools/clang] admin% ../../Debug/bin/clang x.c -pedantic
x.c:4:20: error: size of array has non-integer type 'float'
int size_not_int[f];
^
x.c:5:21: error: array size is negative
int negative_size[1-2];
^~~
x.c:6:17: warning: zero size arrays are an extension
int zero_size[0];
^
3 diagnostics generated.
llvm-svn: 41624
[dylan:~/llvm/tools/clang] admin% cat tentative_decls.c
// incorrectly generates redefinition error
extern int array[3];
int array[3];
// incorrectly generates a redefinition error
extern void nup(int a[3]);
void nup(int a[3]) {}
It turns out that this exposed a fairly major flaw in the type system,
array types were never getting uniqued! This is because all array types
contained an expression, which aren't unique.
To solve this, we now have 2 array types, ConstantArrayType and
VariableArrayType. ConstantArrayType's are unique, VAT's aren't.
This is a fairly extensive set of fundamental changes. Fortunately,
all the tests pass. Nevertheless, there may be some collateral damage:-)
If so, let me know!
llvm-svn: 41592
