copies with memcpy instead of memmove. This matches what GCC does and if it
causes a problem with a particular libc we can always fix it with a target
hook.
llvm-svn: 65699
- Move the 'LabelMap' from Sema to Scope. To avoid layering problems, the second element is now a 'StmtTy *', which makes the LabelMap a bit more verbose to deal with.
- Add 'ActiveScope' to Sema. Managed by ActOnStartOfFunctionDef(), ObjCActOnStartOfMethodDef(), ActOnBlockStmtExpr().
- Changed ActOnLabelStmt(), ActOnGotoStmt(), ActOnAddrLabel(), and ActOnFinishFunctionBody() to use the new ActiveScope.
- Added FIXME to workaround in ActOnFinishFunctionBody() (for dealing with C++ nested functions).
llvm-svn: 65694
As far as I know, this catches all cases of jumping into the scope of a
variable with a variably modified type (excluding statement
expressions) in C. This is missing some stuff we probably want to check
(other kinds of variably modified declarations, statement expressions,
indirect gotos/addresses of labels in a scope, ObjC @try/@finally, cleanup
attribute), the diagnostics aren't very good, and it's not particularly
efficient, but it's a decent start.
This patch is a slightly modified version of the patch I attached to
PR3259, and it fixes that bug. I was sort of planning on improving
it, but I think it's okay as-is, especially since it looks like CodeGen
doesn't have any use for this sort of data structure. The only
significant change I can think of from the version I attached to PR3259
is that this version skips running the checking code when a function
doesn't contain any labels.
This patch doesn't cover case statements, which also need similar
checking; I'm not sure how we should deal with that. Extending the goto
checking to also check case statements wouldn't be too hard; it's just a
matter of keeping track of the scope of the closest switch and checking that
the scope of every case is the same as the scope of the switch. That said,
it would likely be a performance hit to run this check on every
function (it's an extra pass over the entire function), so we probably want
some other solution.
llvm-svn: 65678
array types. Semantic checking for the construction of these types has
been factored out of GetTypeForDeclarator and into separate
subroutines (BuildPointerType, BuildReferenceType,
BuildArrayType). We'll be doing the same thing for all other types
(and declarations and expressions).
As part of this, moved the type-instantiation functions into a class
in an anonymous namespace.
llvm-svn: 65663
we ensure that things added to the module can be found even when they
are not in GlobalDeclMap. The later is for increased flexibility,
should someone want to do something tricky like extern "Ada" in the
same module.
llvm-svn: 65657
building nested member expressions. This location is used to determine the range
of the entire expression, and the expression itself already has its location
inherited from its Base.
This fixes <rdar://problem/6629829>.
llvm-svn: 65650
stubs for those types we don't yet know how to instantiate (everything
that isn't a template parameter!).
We now instantiate default arguments for template type parameters when
needed. This will be our testbed while I fill out the remaining
type-instantiation logic.
llvm-svn: 65649
in C89 mode. This makes it enabled by default instead of only enabled with
-pedantic. Clang defaults to c99 mode, so people will see this more often
than with GCC, but they can always use -std=c89 if they really want c89.
llvm-svn: 65647
normal expression, and change Evaluate and IRGen to evaluate it like a
normal expression. This simplifies the code significantly, and fixes
PR3396.
llvm-svn: 65622
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
nicely sugared type that shows how the user wrote the actual
specialization. This sugared type won't actually show up until we
start doing instantiations.
llvm-svn: 65577
know how to recover from an error, we can attach a hint to the
diagnostic that states how to modify the code, which can be one of:
- Insert some new code (a text string) at a particular source
location
- Remove the code within a given range
- Replace the code within a given range with some new code (a text
string)
Right now, we use these hints to annotate diagnostic information. For
example, if one uses the '>>' in a template argument in C++98, as in
this code:
template<int I> class B { };
B<1000 >> 2> *b1;
we'll warn that the behavior will change in C++0x. The fix is to
insert parenthese, so we use code insertion annotations to illustrate
where the parentheses go:
test.cpp:10:10: warning: use of right-shift operator ('>>') in template
argument will require parentheses in C++0x
B<1000 >> 2> *b1;
^
( )
Use of these annotations is partially implemented for HTML
diagnostics, but it's not (yet) producing valid HTML, which may be
related to PR2386, so it has been #if 0'd out.
In this future, we could consider hooking this mechanism up to the
rewriter to actually try to fix these problems during compilation (or,
after a compilation whose only errors have fixes). For now, however, I
suggest that we use these code modification hints whenever we can, so
that we get better diagnostics now and will have better coverage when
we find better ways to use this information.
This also fixes PR3410 by placing the complaint about missing tokens
just after the previous token (rather than at the location of the next
token).
llvm-svn: 65570
- For types whose native representation is a pointer.
- Use to replace ExprConstant.cpp:HasPointerEvalType,
CodeGenFunction::isObjCPointerType.
llvm-svn: 65569
Needed to make isPropertyReadonly() non-const (for this fix to compile). I imagine there's a way to retain the const-ness, however I have more important fish to fry.
llvm-svn: 65562
The code for looking up local/private method in Sema::ActOnInstanceMessage() was not handling categories properly. Sema::ActOnClassMessage() didn't have this bug.
Created a helper with the correct logic and changed both methods to use it.
llvm-svn: 65532
Also changed ObjCInterfaceDecl::lookupClassMethod() to look through a categories protocols.
Test/patch submitted by Jean-Daniel Dupas (thanks!).
llvm-svn: 65526
pretty sure we want to keep constant expression verification outside of
Evaluate. Because of that, the short-circuit evaluation doesn't
generally make sense, and the comma warning doesn't make sense in its
current form.
llvm-svn: 65525
is a rather big change, but I think this is the direction we want to go;
the code is significantly shorter now, and it doesn't duplicate Evaluate
code. There shouldn't be any visible changes as far as I know.
There has been some movement towards putting ICE handling into
Evaluate (for example, VerifyIntegerConstantExpression uses Evaluate
instead of isICE). This patch is sort of the opposite of the approach,
making ICE handling work without Evaluate being aware of it. I think
this approach is better because it separates the code that does the
constant evaluation from code that's calculating a rather
arbitrary predicate.
The one thing I don't really like about this patch is that
the handling of commas in C99 complicates it signficantly. (Seriously,
what was the standards committee thinking when they wrote that
part?) I think I've come up with a decent approach, but it doesn't feel
ideal. I might add some way to check for evaluated commas from Evaluate
in a subsequent patch; that said, it might not be worth bothering.
llvm-svn: 65524
anymore. If we want to reuse bits and pieces to add strict checking for
constant initializers, we can dig them out of SVN history; the existing
code won't be useful as-is.
llvm-svn: 65502
vector<vector<double>> Matrix;
In C++98/03, this token always means "right shift". However, if we're in
a context where we know that it can't mean "right shift", provide a
friendly reminder to put a space between the two >'s and then treat it
as two >'s as part of recovery.
In C++0x, this token is always broken into two '>' tokens.
llvm-svn: 65484
expressions of the form: 'short x = (y != 10);' While we handle 'int x = (y !=
10)' lazily, the cast to another integer type currently loses the symbolic
constraint. Eager evaluation of the constraint causes the paths to bifurcate and
eagerly evaluate 'y != 10' to a constant of 1 or 0. This should address
<rdar://problem/6619921> until we have a better (more lazy approach) for
handling promotions/truncations of symbolic integer values.
llvm-svn: 65480
decls. Test and document the semantic location of class template
specialization definitions that occur within a scope enclosing the
scope of the class template.
llvm-svn: 65478
specializations. In particular:
- Make sure class template specializations have a "template<>"
header, and complain if they don't.
- Make sure class template specializations are declared/defined
within a valid context. (e.g., you can't declare a specialization
std::vector<MyType> in the global namespace).
llvm-svn: 65476
std::vector<int>::allocator_type
When we parse a template-id that names a type, it will become either a
template-id annotation (which is a parsed representation of a
template-id that has not yet been through semantic analysis) or a
typename annotation (where semantic analysis has resolved the
template-id to an actual type), depending on the context. We only
produce a type in contexts where we know that we only need type
information, e.g., in a type specifier. Otherwise, we create a
template-id annotation that can later be "upgraded" by transforming it
into a typename annotation when the parser needs a type. This occurs,
for example, when we've parsed "std::vector<int>" above and then see
the '::' after it. However, it means that when writing something like
this:
template<> class Outer::Inner<int> { ... };
We have two tokens to represent Outer::Inner<int>: one token for the
nested name specifier Outer::, and one template-id annotation token
for Inner<int>, which will be passed to semantic analysis to define
the class template specialization.
Most of the churn in the template tests in this patch come from an
improvement in our error recovery from ill-formed template-ids.
llvm-svn: 65467
only from a function definition (that does not have a prototype) are
only used to determine the compatible with other declarations of that
same function. In particular, when referencing the function we pretend
as if it does not have a prototype. Implement this behavior, which
fixes PR3626.
llvm-svn: 65460
external declarations to also support external variable
declarations. Unified the code for these two cases into two new
subroutines.
Note that we fail to diagnose cases like the one Neil pointed
out, where a visible non-external declaration hides an external
declaration by the same name. That will require some reshuffling of
name lookup.
llvm-svn: 65385
- For autorelease pool tracking, keep information about the stack of pools
separate from their contents. Also, keep track of the number of times an
autorelease pool will send the "release" message to an object when the pool is
destroyed.
- Update CFRefCount::Update to return a new state instead of a reference count
binding. This will allow us to implement more complicated semantics with
autorelease pools.
llvm-svn: 65384
- Only handles cases with @try with no @catch blocks, and there are a
number of problems with the implementation. Nevertheless, this is
good enough to handled @synchronized correctly, and some other
basic uses.
llvm-svn: 65378
that declaration to global scope so that it can be found from other
scopes. This allows us to diagnose redeclaration errors for external
declarations across scopes. We also warn when name lookup finds such
an out-of-scope declaration. This is part of <rdar://problem/6127293>;
we'll also need to do the same thing for variables.
llvm-svn: 65373
analyzer for array subscript expressions involving bases that are vectors. This
solution is probably a hack: it gets the lvalue of the vector instead of an
rvalue like all other types. This should be reviewed (big FIXME in
GRExprEngine).
llvm-svn: 65366
- When we are declaring a function in local scope, we can merge with
a visible declaration from an outer scope if that declaration
refers to an entity with linkage. This behavior now works in C++
and properly ignores entities without linkage.
- Diagnose the use of "static" on a function declaration in local
scope.
- Diagnose the declaration of a static function after a non-static
declaration of the same function.
- Propagate the storage specifier to a function declaration from a
prior declaration (PR3425)
- Don't name-mangle "main"
llvm-svn: 65360
assertion when the ivars and method list was reset into the existing
interface. To fix this, mark decls as invalid when they are redefined,
and don't insert ivars/methods into invalid decls.
llvm-svn: 65340
I don't think casting super makes any sense (since it won't effect method lookup).
Will discuss with other offline and decide what to do.
llvm-svn: 65317
- Implement instance/class overloading in ObjCContainerDecl (removing a FIXME). This involved hacking NamedDecl::declarationReplaces(), which took awhile to figure out (didn't realize replace was the default).
- Changed Sema::ActOnInstanceMessage() to remove redundant warnings when dealing with protocols. For now, I've omitted the "protocol" term in the diagnostic. It simplifies the code flow and wan't always 100% accurate (e.g. "Foo<Prot>" looks in the class interface, not just the protocol).
- Changed several test cases to jive with the above changes.
llvm-svn: 65292
helper isConstantInitializer) to check whether an initializer is
constant. This passes tests, but it's possible that it'll cause
regressions with real-world code.
Future work:
1. The diagnostics obtained this way are lower quality at the moment;
some work both here and in Evaluate is needed for accurate diagnostics.
2. We probably need some extra code when we're in -pedantic mode so we
can strictly enforce the rules in C99 6.6p7.
3. Dead code cleanup (this should wait until after 2, because we might
want to re-use some of the code).
llvm-svn: 65265
I know, these follow the exact same rules as pointers, so I just made
them use the same codepath. Someone more familiar with ObjC should
double-check this, though.
llvm-svn: 65261
Found while researching <rdar://problem/6497631> Message lookup is sometimes different than gcc's.
Will never be seen in user code. Needed to pass dejagnu testsuite.
llvm-svn: 65244
Should clang have a config.h or should we use the config.h of llvm or using the preprocessor is OK? I did a quick fix here, but having a guideline on how to handle non portable function would be great (or ask ted to stop breaking the windows build :)).
llvm-svn: 65233
Move two key ObjC typechecks from Sema::CheckPointerTypesForAssignment() to ASTContext::mergeTypes().
This allows us to take advantage of the recursion in ASTContext::mergeTypes(), removing some bogus warnings.
This test case I've added includes an example where we still warn (and GCC doesn't). Need to talk with folks and decide what to do. At this point, the major bogosities should be fixed.
llvm-svn: 65231
handle method names that contain 'new', 'copy', etc., but those words might be
the substring of larger words such as 'newsgroup' and 'photocopy' that do not
indicate the allocation of objects. This should address the issues discussed in
<rdar://problem/6552389>.
llvm-svn: 65224
to being allocated from the same bumpptr that the MacroInfo objects
themselves are.
This speeds up -Eonly cocoa.h pth by ~4%, fsyntax-only is barely measurable.
llvm-svn: 65195
with new/delete. With disable-free, this reduces the number of 4/8 byte
mallocs from 4793/1541 to 865/456 and also drops other sizes as well.
This is a very small perf win, nothing major.
llvm-svn: 65171
vanilla reverse-BFS followed by a forward-DFS instead of resulting to strange
histrionics (whose purpose I can no longer remember) in the reverse-BFS stage.
This fixes an assertion failure in BugReporter due to edge cases where no root
was being hit in the reverse-BFS phase.
llvm-svn: 65160
exactly one decl with a specific name in a specific context. This
avoids a bunch of malloc traffic and shrinks StoredDeclsMap to hold
one pointer instead of 3 words (for a std::vector).
This speeds up -fsyntax-only on cocoa.h with PTH by ~7.3%.
llvm-svn: 65103
This prevents emitting diagnostics which are almost certainly useless.
(Note that the test is checking that we emit only one diagnostic.)
llvm-svn: 65101
transitions and then generate a subsequent node that removes the dead symbol
bindings. This should drastically improve caching in the simulation graph when
retain-counted objects are being tracked.
llvm-svn: 65082
information about types. We often print diagnostics where we say
"foo_t" is bad, but the user doesn't know how foo_t is declared
(because it is a typedef). Fix this by expanding sugar when present
in a diagnostic (and not one of a few special cases, like vectors).
Before:
t.m:5:2: error: invalid operands to binary expression ('typeof(P)' and 'typeof(F)')
MAX(P, F);
^~~~~~~~~
t.m:1:78: note: instantiated from:
#define MAX(A,B) ({ __typeof__(A) __a = (A); __typeof__(B) __b = (B); __a < __b ? __b : __a; })
^
After:
t.m:5:2: error: invalid operands to binary expression ('typeof(P)' (aka 'struct mystruct') and 'typeof(F)' (aka 'float'))
MAX(P, F);
^~~~~~~~~
t.m:1:78: note: instantiated from:
#define MAX(A,B) ({ __typeof__(A) __a = (A); __typeof__(B) __b = (B); __a < __b ? __b : __a; })
^
llvm-svn: 65081
- PR3463, PR3398, <rdar://problem/6553401> crash on relocatable
symbol addresses as constants in static locals.
- There are many more scenarious we could handle (like arithmetic on
such an int) but this is the main use case.
llvm-svn: 65074
only occur for pointer types; they are also possible for integer types
now.
- No intended functionality change, IntExprEvaluate doesn't return
LValue results yet.
llvm-svn: 65066
manual setting of the Result.
- Idiom now enforces that result will always have correct width and
type; this exposed three new bugs:
o Enum constant decl value can have different width than type
(PR3173).
o EvaluateInteger should not run an IntExprEvaluator over
non-integral expressions.
o FloatExprEvaluate was not handling casts correctly (it was
evaluating the cast in the IntExprEvaluator!).
llvm-svn: 65053
- Handles assignment to Result with appropriate type.
- Simplifies & encapsulates most direct handling of the Result value;
prep for allowing IntExprEvaluator to deal with LValue APValues.
- No intended functionality change.
llvm-svn: 65038
"after" group instead of the system group makes it so #include <limits.h>
picks up the *system* limits.h file before clang's. This causes a failure
on linux and is definitely not what we want.
llvm-svn: 65026
situation where a tentative decl was emitted *after* the actual
initialization. This occurs in some rare situations with static decls.
- PR3613.
- I'm not particularly happy with this fix, but I don't see a simpler
or more elegant solution yet.
llvm-svn: 65018
- Fix some grammar.
- Fix a bug where a "reference count incremented" diagnostic would not be shown
if the previous typestate was "Released" (only happens in GC mode).
llvm-svn: 64971
(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
we used to not account for escapes in strings with
string concat. Before:
t.m:5:20: warning: field width should have type 'int', but argument has type 'unsigned int'
printf("\n\n" "\n\n%*d", (unsigned) 1, 1);
^ ~~~~~~~~~~~~
after:
t.m:5:23: warning: field width should have type 'int', but argument has type 'unsigned int'
printf("\n\n" "\n\n%*d", (unsigned) 1, 1);
^ ~~~~~~~~~~~~
llvm-svn: 64941
escapes in the string for subtoken positioning. This gives
us working examples like:
t.m:5:16: warning: field width should have type 'int', but argument has type 'unsigned int'
printf("\n\n%*d", (unsigned) 1, 1);
^ ~~~~~~~~~~~~
where before the caret pointed two spaces to the left.
llvm-svn: 64940
retain/releases performed via [... release] and CFRetain(). The former are
no-ops in GC. The checker already handled this, but now we emit nice diagnostics
to the user telling them that these are no-ops.
llvm-svn: 64937
We now emit:
t.m:6:15: warning: field width should have type 'int', but argument has type 'unsigned int'
printf(STR, (unsigned) 1, 1);
^ ~~~~~~~~~~~~
t.m:3:18: note: instantiated from:
#define STR "abc%*ddef"
^
which has the correct location in the string literal in the note line.
llvm-svn: 64936
and escaped newlines don't throw off the offset computation.
On this testcase:
printf("abc\
def"
"%*d", (unsigned) 1, 1);
Before:
t.m:5:5: warning: field width should have type 'int', but argument has type 'unsigned int'
def"
^
after:
t.m:6:12: warning: field width should have type 'int', but argument has type 'unsigned int'
"%*d", (unsigned) 1, 1);
^ ~~~~~~~~~~~~
llvm-svn: 64930
First step, handle diagnostics in StringLiteral's that are due to token pasting.
For example, we now handle:
id str2 = @"foo"
"bar"
@"baz"
" b\0larg"; // expected-warning {{literal contains NUL character}}
Correctly:
test/SemaObjC/exprs.m:17:15: warning: CFString literal contains NUL character
" b\0larg"; // expected-warning {{literal contains NUL character}}
~~~^~~~~~~
There are several other related issues still to be done.
llvm-svn: 64924
us whether there was an error in trying to parse a type-name (type-id
in C++). This allows propagation of errors further in the compiler,
suppressing more bogus error messages.
llvm-svn: 64922
any named parameters, e.g., this is accepted in C:
void f(...) __attribute__((overloadable));
although this would be rejected:
void f(...);
To do this, moved the checking of the "ellipsis without any named
arguments" condition from the parser into Sema (where it belongs anyway).
llvm-svn: 64902
to do in this area, since there are other places that reference
FunctionDecls.
Don't allow "overloadable" functions (in C) to be declared without a
prototype.
llvm-svn: 64897
the various PPTokens that are pasted together to make it. In the course
of working on this, I discovered ParseObjCStringLiteral which needs some
work. I'll tackle it next.
llvm-svn: 64892
When comparing if one Range is "less" than another, compare the actual APSInt
numeric values instead of their pointer addresses. This ensures that the
ImmutableSet in RangeSet always has a consistent ordering between Ranges. This
is critical for generating the same digest/hash for the contents of the sets.
This was a serious performance bug because it would often cause state caching
to be disabled along complicated paths.
Along the way:
- Put Range and RangeSet in the "anonymous namespace" and mark them hidden
llvm-svn: 64890
back to the summary used when evaluating the statement associated with a
simulation node. This is now being used to help improve the checker's
diagnostics. To get things started, the checker now emits a path diagnostic
indicating that 'autorelease' is a no-op in GC mode.
Some of these changes are exposing further grossness in the interface between
BugReporter and the ExplodedGraph::Trim facilities. These really need to be
cleaned up one day.
llvm-svn: 64881
1. Return of _Complex long double used wrong type.
2. va_arg of types passed in two SSE registers didn't account for
extra space in register save area.
Down to 18 failures on gcc/compat/x86_64. Combined 32/64 results are:
--
=== gcc Summary ===
# of expected passes 1292
# of unexpected failures 34
# of unsupported tests 2
--
llvm-svn: 64880
- This idiom ensures that the result will have the right width and
type.
- Tested on most of x86_64/llvm-test to satisfy my paranoia.
- This fixes at least the following bugs:
o UnaryTypeTraitExpr wasn't setting the width correctly.
o Arithmetic on _Bool wasn't setting the width correctly.
And probably a number more.
llvm-svn: 64864
keep searching for C++ headers when in C++ mode). In theory clang
should be able to find all of its own headers now. If not, the
CPATH or C_INCLUDE_PATH environment variables can be specified to
add a include path.
llvm-svn: 64862
specialization of class templates, e.g.,
template<typename T> class X;
template<> class X<int> { /* blah */ };
Each specialization is a different *Decl node (naturally), and can
have different members. We keep track of forward declarations and
definitions as for other class/struct/union types.
This is only the basic framework: we still have to deal with checking
the template headers properly, improving recovery when there are
failures, handling nested name specifiers, etc.
llvm-svn: 64848
- Renamed to getDeclAlignInBytes since most other query functions
work in bits.
- Fun to track down as isIntegerConstantExpr was getting it right,
but Evaluate() was getting it wrong. Maybe we should assert they
compute the same thing when they succeed?
llvm-svn: 64828
Zhongxing Xu. The resultant code is less than 1/2 the size of the original.
Key highlights:
- All CouldBeXXX methods have been removed. Checking for feasibility is now just
done in the AddXXX methods.
- RangeSets now represent "all possible values" explicitly as the range set {
[min, max] } instead of the empty set. The empty set now represents "no
feasible values". This change consolidated much of the core algorithm to only
have one code path instead of alternate paths that considered the empty set to
represent "all possible falues."
llvm-svn: 64787
Chris Lattner