Support automatically setting the calling convention to
STDMETHODCALLTYPE if it is part of an object interface.
Write out the calling convention in write_type_v.
In particular, don't allow them after cpp_quote and without a statement.
Update the rules for importlib, library definition, dispinterface
definition, interface definition, coclass definition and module
definition to optionally allow a semicolon postfix.
Call pop_import from the parser instead of the lexer to ensure that
pop_import is only called after the last statement in the imported file
has been parsed.
This is because write_string_tfs has support for the full range of
string types, write_simple_pointer doesn't have access to the var
attributes which are needed to properly detect strings, and
write_string_tfs sets the typestring offset to after the pointer is
written and write_remoting_arg depends on this.
is_ptr cannot be used because it follows the chain of types into the
type which has the context_handle attribute, which is typically "void *"
and so causes these context handles to be incorrectly detected as
context handles. Instead, we can use is_aliaschain_ptr to follow the
chain of aliases without following pointers and the absence of the
context_handle attribute indicates that it must be present on a type
after following a pointer.
Fix the is_string_type function used for detecting strings by only
examining aliases instead of both aliases and pointers. This is due to
the requirement that pointers to strings be handled as pointers and so
not detected as strings.
Structures containing enum fields without the v1_enum attribute must be
treated as complex because of the range check and memory/wire size
differences. This includes a test which didn't pass before and now
passes on Wine and Windows.
It doesn't matter whether or not the array is declared as a pointer or
is declared using array subscripts - the array is still allocated by the
unmarshalling function and so needs to be freed.
Fix a typo that caused varying arrays never to be freed.
Fix another typo with the check for conformant arrays being performed
twice, redundantly.
Otherwise, writing a type at top-level could cause the pointer format
string to be used in non-top-level places which may cause memory
corruption during freeing.
write_embedded_types doesn't descend into all types contained by the
array or structure, so create a new function, type_has_pointers, for the
purpose of descending into all types and returning whether or not
pointers are present.
Therefore, needs_freeing would have to duplicate a lot of
write_remoting_arg in order to get the detection right. Because of
this, it is easier and will cause less problems in the future to
simply move the logic of needs_freeing into write_remoting_arg in the
appropriate handling code for the detected type.
