much as we already do for allocation function lookup. Explicitly check access
for the function we actually select in one case that was previously missing,
but being caught behind the blanket diagnostics for all overload candidates.
This fixs PR7436.
llvm-svn: 106986
floating-point conversions or floating-integral conversions. We
really, really, really need to make isFloatingType() and friends not
apply to vector types.
llvm-svn: 106551
"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
introduced by using decls are hidden even if their template parameter lists
or return types differ from the "overriding" declaration.
Propagate using shadow declarations around more effectively when looking up
template-ids. Reperform lookup for template-ids in member expressions so that
access control is properly set up.
Fix some number of latent bugs involving template-ids with totally invalid
base types. You can only actually get these with a scope specifier, since
otherwise the template-id won't parse as a template-id.
Fixes PR7384.
llvm-svn: 106093
C++ semantics, eliminating an extension diagnostic that doesn't match
C++ semantics (ordered comparison with NULL) and tightening some
extwarns to errors in C++ to match GCC and maintain conformance in
SFINAE contexts. Fixes <rdar://problem/7941392>.
llvm-svn: 106050
Fix string concatenation to treat escapes in concatenated strings that
are wide because of other string chunks to process the escapes as wide
themselves. Before we would warn about and miscompile the attached testcase.
This fixes rdar://8040728 - miscompile + warning: hex escape sequence out of range
llvm-svn: 106012
being a subsequence of another standard conversion sequence. Instead
of requiring exact type equality for the second conversion step,
require type *similarity*, which is type equality with cv-qualifiers
removed at all levels. This appears to match the behavior of EDG and
VC++ (albeit not GCC), and feels more intuitive. Big thanks to John
for the line of reasoning that supports this change: since
cv-qualifiers are orthogonal to the second conversion step, we should
ignore them in the type comparison.
llvm-svn: 105678
VLA restrictions so that one can use VLAs in templates (even
accidentally), but not as part of a non-type template parameter (which
would be very bad).
llvm-svn: 104471
in several important ways:
- VLAs of non-POD types are not permitted.
- VLAs cannot be used in conjunction with C++ templates.
These restrictions are intended to keep VLAs out of the parts of the
C++ type system where they cause the most trouble. Fixes PR5678 and
<rdar://problem/8013618>.
llvm-svn: 104443
instance variables:
- Use isRecordType() rather than isa<RecordType>(), so that we see
through typedefs in ivar types.
- Mark the destructor as referenced
- Perform C++ access control on the destructor
llvm-svn: 104206
create a temporary copy of both the "true" and "false" results. Fixes
the Boost.Interprocess failures.
Daniel did all the hard work of tracking down the issue, I get to type
up the trivial fix for this horrible miscompile.
llvm-svn: 104184
involves extending implicit conversion sequences to model vector
conversions and vector splats, along with teaching the C++ conditional
operator-checking code about vector types.
Fixes <rdar://problem/7983501>.
llvm-svn: 104081
"used" (e.g., we will refer to the vtable in the generated code) and
when they are defined (i.e., because we've seen the key function
definition). Previously, we were effectively tracking "potential
definitions" rather than uses, so we were a bit too eager about emitting
vtables for classes without key functions.
The new scheme:
- For every use of a vtable, Sema calls MarkVTableUsed() to indicate
the use. For example, this occurs when calling a virtual member
function of the class, defining a constructor of that class type,
dynamic_cast'ing from that type to a derived class, casting
to/through a virtual base class, etc.
- For every definition of a vtable, Sema calls MarkVTableUsed() to
indicate the definition. This happens at the end of the translation
unit for classes whose key function has been defined (so we can
delay computation of the key function; see PR6564), and will also
occur with explicit template instantiation definitions.
- For every vtable defined/used, we mark all of the virtual member
functions of that vtable as defined/used, unless we know that the key
function is in another translation unit. This instantiates virtual
member functions when needed.
- At the end of the translation unit, Sema tells CodeGen (via the
ASTConsumer) which vtables must be defined (CodeGen will define
them) and which may be used (for which CodeGen will define the
vtables lazily).
From a language perspective, both the old and the new schemes are
permissible: we're allowed to instantiate virtual member functions
whenever we want per the standard. However, all other C++ compilers
were more lazy than we were, and our eagerness was both a performance
issue (we instantiated too much) and a portability problem (we broke
Boost test cases, which now pass).
Notes:
(1) There's a ton of churn in the tests, because the order in which
vtables get emitted to IR has changed. I've tried to isolate some of
the larger tests from these issues.
(2) Some diagnostics related to
implicitly-instantiated/implicitly-defined virtual member functions
have moved to the point of first use/definition. It's better this
way.
(3) I could use a review of the places where we MarkVTableUsed, to
see if I missed any place where the language effectively requires a
vtable.
Fixes PR7114 and PR6564.
llvm-svn: 103718
member function (default constructor, copy constructor, copy
assignment operator, destructor), emit a note showing where that
implicit definition was required.
llvm-svn: 103619
about the permitted scopes. Specifically:
1) Permit labels and gotos to appear after a prologue of variable initializations.
2) Permit indirect gotos to jump out of scopes that don't require cleanup.
3) Diagnose possible attempts to indirect-jump out of scopes that do require
cleanup.
This requires a substantial reinvention of the algorithm for checking indirect
goto. The current algorithm is Omega(M*N), with M = the number of unique
scopes being jumped from and N = the number of unique scopes being jumped to,
with an additional factor that is probably (worst-case) linear in the depth
of scopes. Thus the entire thing is likely cubic given some truly bizarre
ill-formed code; on well-formed code the additional factor collapses to
an amortized constant (when amortized over the entire function) and so
the algorithm is quadratic. Even this requires every label to appear in
its own scope, which would be very unusual for indirect-goto code (and
extremely unlikely for well-formed code); it is far more likely that
all labels will be in the same scope and so the algorithm becomes linear.
For such a marginal feature, I am fairly happy with this result.
(this is using JumpDiagnostic's definition of scope, where successive
variables in a block appear in their own scope)
llvm-svn: 103536
referenced unless we see one of them defined (or the key function
defined, if it as one) or if we need the vtable for something. Fixes
PR7114.
llvm-svn: 103497
"bottom-up" when implicit casts and comparisons are inserted, compute them
"top-down" when the full expression is finished. Makes it easier to
coordinate warnings and thus implement -Wconversion for signedness
conversions without double-warning with -Wsign-compare. Also makes it possible
to realize that a signedness conversion is okay because the context is
performing the inverse conversion. Also simplifies some logic that was
trying to calculate the ultimate comparison/result type and getting it wrong.
Also fixes a problem with the C++ explicit casts which are often "implemented"
in the AST with a series of implicit cast expressions.
llvm-svn: 103174
provide a note that shows where the copy-assignment operator was
needed. We used to have this, but I broke it during refactoring.
Finishes PR6999.
llvm-svn: 103127
implicitly-defined copy assignment operator, suppress the protected
access check. This eliminates the remaining failure in the
Boost.SmartPtr library (that was a product of the copy-assignment
generation rewrite) and, presumably, the Boost.TR1 library as well.
llvm-svn: 103010
assignment operators.
Previously, Sema provided type-checking and template instantiation for
copy assignment operators, then CodeGen would synthesize the actual
body of the copy constructor. Unfortunately, the two were not in sync,
and CodeGen might pick a copy-assignment operator that is different
from what Sema chose, leading to strange failures, e.g., link-time
failures when CodeGen called a copy-assignment operator that was not
instantiation, run-time failures when copy-assignment operators were
overloaded for const/non-const references and the wrong one was
picked, and run-time failures when by-value copy-assignment operators
did not have their arguments properly copy-initialized.
This implementation synthesizes the implicitly-defined copy assignment
operator bodies in Sema, so that the resulting ASTs encode exactly
what CodeGen needs to do; there is no longer any special code in
CodeGen to synthesize copy-assignment operators. The synthesis of the
body is relatively simple, and we generate one of three different
kinds of copy statements for each base or member:
- For a class subobject, call the appropriate copy-assignment
operator, after overload resolution has determined what that is.
- For an array of scalar types or an array of class types that have
trivial copy assignment operators, construct a call to
__builtin_memcpy.
- For an array of class types with non-trivial copy assignment
operators, synthesize a (possibly nested!) for loop whose inner
statement calls the copy constructor.
- For a scalar type, use built-in assignment.
This patch fixes at least a few tests cases in Boost.Spirit that were
failing because CodeGen picked the wrong copy-assignment operator
(leading to link-time failures), and I suspect a number of undiagnosed
problems will also go away with this change.
Some of the diagnostics we had previously have gotten worse with this
change, since we're going through generic code for our
type-checking. I will improve this in a subsequent patch.
llvm-svn: 102853
if *none* of the successors of the call expression is the exit block.
This matters when a call of bool type is the condition of (say) a while
loop in a function with no statements after the loop. This *can* happen
in C, but it's much more common in C++ because of overloaded operators.
Suppresses some substantial number of spurious -Wmissing-noreturn warnings.
llvm-svn: 102696
classes, since we only warn (not error) on offsetof() for non-POD
types. We store the base path within the OffsetOfExpr itself, then
evaluate the offsets within the constant evaluator.
llvm-svn: 102571
Amadini.
This change introduces a new expression node type, OffsetOfExpr, that
describes __builtin_offsetof. Previously, __builtin_offsetof was
implemented using a unary operator whose subexpression involved
various synthesized array-subscript and member-reference expressions,
which was ugly and made it very hard to instantiate as a
template. OffsetOfExpr represents the AST more faithfully, with proper
type source information and a more compact representation.
OffsetOfExpr also has support for dependent __builtin_offsetof
expressions; it can be value-dependent, but will never be
type-dependent (like sizeof or alignof). This commit introduces
template instantiation for __builtin_offsetof as well.
There are two major caveats to this patch:
1) CodeGen cannot handle the case where __builtin_offsetof is not a
constant expression, so it produces an error. So, to avoid
regressing in C, we retain the old UnaryOperator-based
__builtin_offsetof implementation in C while using the shiny new
OffsetOfExpr implementation in C++. The old implementation can go
away once we have proper CodeGen support for this case, which we
expect won't cause much trouble in C++.
2) __builtin_offsetof doesn't work well with non-POD class types,
particularly when the designated field is found within a base
class. I will address this in a subsequent patch.
Fixes PR5880 and a bunch of assertions when building Boost.Python
tests.
llvm-svn: 102542
we were relying on checking for abstract class types when an array
type was actually used to declare a variable, parameter, etc. However,
we need to check when the construct the array for, e.g., SFINAE
purposes (see DR337). Fixes problems with Boost's is_abstract type
trait.
llvm-svn: 102452
by using TypeSourceInfo, cleaning up the representation
somewhat. Teach getTypeOperand() to strip references and
cv-qualifiers, providing the semantic view of the type without
requiring any extra storage (the unmodified type remains within the
TypeSourceInfo). This fixes a bug found by Boost's call_traits test.
Finally, clean up semantic analysis, by splitting the ActOnCXXTypeid
routine into ActOnCXXTypeId (the parser action) and two BuildCXXTypeId
functions, which perform the semantic analysis for typeid(type) and
typeid(expression), respectively. We now perform less work at template
instantiation time (we don't look for std::type_info again) and can
give better diagnostics.
llvm-svn: 102393
thing. Audit all uses of Type::isStructure(), changing those calls to
isStructureOrClassType() as needed (which is alsmost
everywhere). Fixes the remaining failure in Boost.Utility/Swap.
llvm-svn: 102386
using declaration, look at its underlying declaration to determine the
lookup result kind (e.g., overloaded, unresolved). Fixes at least one
issue in Boost.Bimap.
llvm-svn: 102317
copy constructor, suppress user-defined conversions on the
argument. Otherwise, we can end up in a recursion loop where the
bind the argument of the copy constructor to another copy constructor call,
whose argument is then a copy constructor call...
Found by Boost.Regex which, alas, still isn't building.
llvm-svn: 102269
way that C does. Among other differences, elaborated type specifiers
are defined to skip "non-types", which, as you might imagine, does not
include typedefs. Rework our use of IDNS masks to capture the semantics
of different kinds of declarations better, and remove most current lookup
filters. Removing the last remaining filter is more complicated and will
happen in a separate patch.
Fixes PR 6885 as well some spectrum of unfiled bugs.
llvm-svn: 102164
address of overloaded function, instead of assuming that a nested name
specifier was used. A nested name specifier is not required for static
functions.
Fixes PR6886.
llvm-svn: 102107
method parameter, provide a note pointing at the parameter itself so
the user does not have to manually look for the function/method being
called and match up parameters to arguments. For example, we now get:
t.c:4:5: warning: incompatible pointer types passing 'long *' to
parameter of
type 'int *' [-pedantic]
f(long_ptr);
^~~~~~~~
t.c:1:13: note: passing argument to parameter 'x' here
void f(int *x);
^
llvm-svn: 102038
resolution ([over.ics.ref]), we take some shortcuts required by the
standard that effectively permit binding of a const volatile reference
to an rvalue. We have to treat lightly here to avoid infinite
recursion.
Fixes PR6177.
llvm-svn: 101712
temporary object. This is blindingly obvious from reading C++
[over.match.ctor]p1, but somehow I'd missed it and it took DR152 to
educate me. Adjust one test that was relying on this non-standard
behavior.
llvm-svn: 101688
resolution. There are two sources of problems involving user-defined
conversions that this change eliminates, along with providing simpler
interfaces for checking implicit conversions:
- It eliminates a case of infinite recursion found in Boost.
- It eliminates the search for the constructor needed to copy a temporary
generated by an implicit conversion from overload
resolution. Overload resolution assumes that, if it gets a value
of the parameter's class type (or a derived class thereof), there
is a way to copy if... even if there isn't. We now model this
properly.
llvm-svn: 101680
intended for redeclarations, fixing those that need it. Fixes PR6831.
This uncovered an issue where the C++ type-specifier-seq parsing logic
would try to perform name lookup on an identifier after it already had
a type-specifier, which could also lead to spurious ambiguity errors
(as in PR6831, but with a different test case).
llvm-svn: 101419
ASTContext::getTypeSize() rather than ASTContext::getIntWidth() for
the width of an integral type. The former includes padding for bools
(to the target's size) while the latter does not, so we woud end up
zero-extending bools to the target width when we shouldn't. Fixes a
crash-on-valid in the included test.
llvm-svn: 101372
that have reference or const scalar members, since those members can
never be initializer or modified. Fixes <rdar://problem/7804350>.
llvm-svn: 101316
ResolveAddressOfOverloadedFunction when asked to complain. Previously,
we had some weird handshake where ResolveAddressOfOverloadedFunction
expected its caller to handle some of the diagnostics but not others,
and yet there was no way for the caller to know which case we were
in. Eliminate this madness, fixing <rdar://problem/7765884>.
llvm-svn: 101312
separate count of "suppressed" errors. This way, semantic analysis
bits that depend on the error count to determine whether problems
occured (e.g., some template argument deduction failures, jump-scope
checking) will not get confused.
The actual problem here is that a missing #include (which is a fatal
error) could cause the jump-scope checker to run on invalid code,
which it is not prepared to do. Trivial fix for both
<rdar://problem/7775941> and <rdar://problem/7775709>.
llvm-svn: 101297
that adds parentheses from the main diagnostic down to a new
note. This way, when the fix-it represents a choice between two
options, each of the options is associted with a note. There is no
default option in such cases. For example:
/Users/dgregor/t.c:2:9: warning: & has lower precedence than ==; ==
will be
evaluated first [-Wparentheses]
if (x & y == 0) {
^~~~~~~~
/Users/dgregor/t.c:2:9: note: place parentheses around the &
expression to
evaluate it first
if (x & y == 0) {
^
( )
/Users/dgregor/t.c:2:9: note: place parentheses around the ==
expression to
silence this warning
if (x & y == 0) {
^
( )
llvm-svn: 101249
copying the type location information from the conversion-type-id into
the type location information for the function type. Do something
similar for constructors and destructors, by giving their "void"
return type source-location information.
In all of these cases, we previously left this type-source information
uninitialized, which led to various unfortunate crashes.
We still aren't tracking good source-location information for the
actual names. That's PR6357.
John, please check my sanity on this.
llvm-svn: 101088
sure to introduce them into the current Scope (when we have one) in
addition to the DeclContext for the class, so that they can be found
by name lookup for inline members of the class. Fixes PR6570.
llvm-svn: 101047
member. Use a better diagnostic for this case. Also fix a bug with nested
anonymous structs/unions for -Wreorder; this last was PR6575.
llvm-svn: 100923
when they're instantiated. Merge the note into the -Wreorder warning; it
doesn't really contribute much, and it was splitting a thought across diagnostics
anyway. Don't crash in the parser when a constructor's initializers end in a
comma and there's no body; the recovery here is still terrible, but anything's
better than a crash.
llvm-svn: 100922
Remove -faccess-control from -cc1; add -fno-access-control.
Make the driver pass -fno-access-control by default.
Update a bunch of tests to be correct under access control.
llvm-svn: 100880
the implicit template instantiations we need to perform. Otherwise, we
end up erroneously diagnosing static functions as used if they were
only used within an implicit template instantiation. Fixes a bunch of
spurious failures when building Clang with Clang.
llvm-svn: 100872
destination type for initialization, assignment, parameter-passing,
etc. The main issue fixed here is that we used rather confusing
wording for diagnostics such as
t.c:2:9: warning: initializing 'char const [2]' discards qualifiers,
expected 'char *' [-pedantic]
char *name = __func__;
^ ~~~~~~~~
We're not initializing a 'char const [2]', we're initializing a 'char
*' with an expression of type 'char const [2]'. Similar problems
existed for other diagnostics in this area, so I've normalized them all
with more precise descriptive text to say what we're
initializing/converting/assigning/etc. from and to. The warning for
the code above is now:
t.c:2:9: warning: initializing 'char *' from an expression of type
'char const [2]' discards qualifiers [-pedantic]
char *name = __func__;
^ ~~~~~~~~
Fixes <rdar://problem/7447179>.
llvm-svn: 100832
<tr1/hashtable> header, where a friend class template
std::tr1::__detail::_Map_base is declared with the wrong template
parameters. GCC doesn't catch the problem, so Clang does a little
back-flip to avoid diagnosing just this one instance of the problem.
llvm-svn: 100790
poor (and wrong) approximation of the actual rules governing when to
build a copy and when it can be elided.
The correct implementation is actually simpler than the
approximation. When we only enumerate constructors as part of
initialization (e.g., for direct initialization or when we're copying
from a class type or one of its derived classes), we don't create a
copy. When we enumerate all conversion functions, we do create a
copy. Before, we created some extra copies and missed some
others. The new test copy-initialization.cpp shows a case where we
missed creating a (required, non-elidable) copy as part of a
user-defined conversion, which resulted in a miscompile. This commit
also fixes PR6757, where the missing copy made us reject well-formed
code in the ternary operator.
This commit also cleans up our handling of copy elision in the case
where we create an extra copy of a temporary object, which became
necessary now that we produce the right copies. The code that seeks to
find the temporary object being copied has moved into
Expr::getTemporaryObject(); it used to have two different
not-quite-the-same implementations, one in Sema and one in CodeGen.
Note that we still do not attempt to perform the named return value
optimization, so we miss copy elisions for return values and throw
expressions.
llvm-svn: 100196
nested-name-specifier (e.g., "class T::foo") fails to find a tag
member in the scope nominated by the
nested-name-specifier. Previously, we gave a bland
error: 'Nested' does not name a tag member in the specified scope
which didn't actually say where we were looking, which was rather
horrible when the nested-name-specifier was instantiated. Now, we give
something a bit better:
error: no class named 'Nested' in 'NoDepBase<T>'
llvm-svn: 100060
Douglas Gregor