binutils-gdb/include/gcc-cp-fe.def

1051 lines
45 KiB
Modula-2
Raw Normal View History

Sync libiberty/ & include/ with GCC Note this brings in the interface files for libcc1/G++ as well, which we will be needing in GDB soon anyway. That commit renamed a method in the C interface and that required a small update to GDB's compile/ code, which I've included that in this patch to keep the tree building. include/ChangeLog: 2017-09-15 Pedro Alves <palves@redhat.com> * ansidecl.h (DISABLE_COPY_AND_ASSIGN): New macro. 2017-09-12 Jiong Wang <jiong.wang@arm.com> * dwarf2.def (DW_CFA_AARCH64_negate_ra_state): New DW_CFA_DUP. * dwarf2.h (DW_CFA_DUP): New define. 2017-08-21 Richard Biener <rguenther@suse.de> * simple-object.h (simple_object_copy_lto_debug_sections): New function. 2017-05-18 Martin Liska <mliska@suse.cz> * ansidecl.h: Define CONSTEXPR macro. 2017-05-24 Nathan Sidwell <nathan@acm.org> * libiberty.h (ASTRDUP): Adjust cast to avoid warning. 2017-01-30 Alexandre Oliva <aoliva@redhat.com> Introduce C++ support in libcc1. * gcc-c-fe.def (int_type_v0): Rename from... (int_type): ... this. Introduce new version. (float_type_v0): Rename from... (float_type): ... this. Introduce new version. (char_type): New. * gcc-c-interface.h (gcc_c_api_version): Add GCC_C_FE_VERSION_1. (gcc_type_array): Move... * gcc-interface.h: ... here. * gcc-cp-fe.def: New. * gcc-cp-interface.h: New. 2016-04-29 Oleg Endo <olegendo@gcc.gnu.org> * longlong.h (umul_ppmm): Remove SHMEDIA checks. (__umulsidi3, count_leading_zeros): Remove SHMEDIA implementations. 2017-09-15 Yao Qi <yao.qi@linaro.org> Pedro Alves <palves@redhat.com> * ansidecl.h (DISABLE_COPY_AND_ASSIGN): New macro. 2017-09-12 Jiong Wang <jiong.wang@arm.com> * dwarf2.def (DW_CFA_AARCH64_negate_ra_state): New DW_CFA_DUP. * dwarf2.h (DW_CFA_DUP): New define. 2017-08-21 Richard Biener <rguenther@suse.de> * simple-object.h (simple_object_copy_lto_debug_sections): New function. 2017-05-18 Martin Liska <mliska@suse.cz> * ansidecl.h: Define CONSTEXPR macro. 2017-05-24 Nathan Sidwell <nathan@acm.org> * libiberty.h (ASTRDUP): Adjust cast to avoid warning. 2017-01-30 Alexandre Oliva <aoliva@redhat.com> Introduce C++ support in libcc1. * gcc-c-fe.def (int_type_v0): Rename from... (int_type): ... this. Introduce new version. (float_type_v0): Rename from... (float_type): ... this. Introduce new version. (char_type): New. * gcc-c-interface.h (gcc_c_api_version): Add GCC_C_FE_VERSION_1. (gcc_type_array): Move... * gcc-interface.h: ... here. * gcc-cp-fe.def: New. * gcc-cp-interface.h: New. 2016-04-29 Oleg Endo <olegendo@gcc.gnu.org> * longlong.h (umul_ppmm): Remove SHMEDIA checks. (__umulsidi3, count_leading_zeros): Remove SHMEDIA implementations. libiberty/ChangeLog: 2017-09-15 Nathan Sidwell <nathan@acm.org> PR demangler/82195 * cp-demangle.c (d_name): Add 'toplevel' parm. Pass to ... (d_local_name): ... here. Parse trailing function args on nested local_name. (d_encoding, d_special_name, d_class_enum_type): Adjust d_name calls. * testsuite/demangle-expected: Add tests. 2017-09-15 Richard Biener <rguenther@suse.de> PR lto/81968 * simple-object-elf.c (simple_object_elf_copy_lto_debug_sections): Iterate marking dependent sections necessary. 2017-09-15 Nathan Sidwell <nathan@acm.org> * cp-demangle.c (is_fnqual_component_type): Reimplement using FNQUAL_COMPONENT_CASE. (d_encoding): Hold bare_function_type in local var. (d_local_name): Build name in both cases and build result once. Collapse switch-if to single conditional. (d_local_name): * testsuite/demangle-expected: Realign blank lines with tests. 2017-09-12 Jiong Wang <jiong.wang@arm.com> * dwarfnames.c (DW_CFA_DUP): New define. gdb/ChangeLog: 2017-09-15 Pedro Alves <palves@redhat.com> * compile/compile-c-types.c (convert_enum, convert_int) (convert_float): Adjust to refer to int_type_v0 and float_type_v0.
2017-09-15 18:40:33 +02:00
/* Interface between GCC C++ FE and GDB -*- c -*-
Copyright (C) 2014-2020 Free Software Foundation, Inc.
Sync libiberty/ & include/ with GCC Note this brings in the interface files for libcc1/G++ as well, which we will be needing in GDB soon anyway. That commit renamed a method in the C interface and that required a small update to GDB's compile/ code, which I've included that in this patch to keep the tree building. include/ChangeLog: 2017-09-15 Pedro Alves <palves@redhat.com> * ansidecl.h (DISABLE_COPY_AND_ASSIGN): New macro. 2017-09-12 Jiong Wang <jiong.wang@arm.com> * dwarf2.def (DW_CFA_AARCH64_negate_ra_state): New DW_CFA_DUP. * dwarf2.h (DW_CFA_DUP): New define. 2017-08-21 Richard Biener <rguenther@suse.de> * simple-object.h (simple_object_copy_lto_debug_sections): New function. 2017-05-18 Martin Liska <mliska@suse.cz> * ansidecl.h: Define CONSTEXPR macro. 2017-05-24 Nathan Sidwell <nathan@acm.org> * libiberty.h (ASTRDUP): Adjust cast to avoid warning. 2017-01-30 Alexandre Oliva <aoliva@redhat.com> Introduce C++ support in libcc1. * gcc-c-fe.def (int_type_v0): Rename from... (int_type): ... this. Introduce new version. (float_type_v0): Rename from... (float_type): ... this. Introduce new version. (char_type): New. * gcc-c-interface.h (gcc_c_api_version): Add GCC_C_FE_VERSION_1. (gcc_type_array): Move... * gcc-interface.h: ... here. * gcc-cp-fe.def: New. * gcc-cp-interface.h: New. 2016-04-29 Oleg Endo <olegendo@gcc.gnu.org> * longlong.h (umul_ppmm): Remove SHMEDIA checks. (__umulsidi3, count_leading_zeros): Remove SHMEDIA implementations. 2017-09-15 Yao Qi <yao.qi@linaro.org> Pedro Alves <palves@redhat.com> * ansidecl.h (DISABLE_COPY_AND_ASSIGN): New macro. 2017-09-12 Jiong Wang <jiong.wang@arm.com> * dwarf2.def (DW_CFA_AARCH64_negate_ra_state): New DW_CFA_DUP. * dwarf2.h (DW_CFA_DUP): New define. 2017-08-21 Richard Biener <rguenther@suse.de> * simple-object.h (simple_object_copy_lto_debug_sections): New function. 2017-05-18 Martin Liska <mliska@suse.cz> * ansidecl.h: Define CONSTEXPR macro. 2017-05-24 Nathan Sidwell <nathan@acm.org> * libiberty.h (ASTRDUP): Adjust cast to avoid warning. 2017-01-30 Alexandre Oliva <aoliva@redhat.com> Introduce C++ support in libcc1. * gcc-c-fe.def (int_type_v0): Rename from... (int_type): ... this. Introduce new version. (float_type_v0): Rename from... (float_type): ... this. Introduce new version. (char_type): New. * gcc-c-interface.h (gcc_c_api_version): Add GCC_C_FE_VERSION_1. (gcc_type_array): Move... * gcc-interface.h: ... here. * gcc-cp-fe.def: New. * gcc-cp-interface.h: New. 2016-04-29 Oleg Endo <olegendo@gcc.gnu.org> * longlong.h (umul_ppmm): Remove SHMEDIA checks. (__umulsidi3, count_leading_zeros): Remove SHMEDIA implementations. libiberty/ChangeLog: 2017-09-15 Nathan Sidwell <nathan@acm.org> PR demangler/82195 * cp-demangle.c (d_name): Add 'toplevel' parm. Pass to ... (d_local_name): ... here. Parse trailing function args on nested local_name. (d_encoding, d_special_name, d_class_enum_type): Adjust d_name calls. * testsuite/demangle-expected: Add tests. 2017-09-15 Richard Biener <rguenther@suse.de> PR lto/81968 * simple-object-elf.c (simple_object_elf_copy_lto_debug_sections): Iterate marking dependent sections necessary. 2017-09-15 Nathan Sidwell <nathan@acm.org> * cp-demangle.c (is_fnqual_component_type): Reimplement using FNQUAL_COMPONENT_CASE. (d_encoding): Hold bare_function_type in local var. (d_local_name): Build name in both cases and build result once. Collapse switch-if to single conditional. (d_local_name): * testsuite/demangle-expected: Realign blank lines with tests. 2017-09-12 Jiong Wang <jiong.wang@arm.com> * dwarfnames.c (DW_CFA_DUP): New define. gdb/ChangeLog: 2017-09-15 Pedro Alves <palves@redhat.com> * compile/compile-c-types.c (convert_enum, convert_int) (convert_float): Adjust to refer to int_type_v0 and float_type_v0.
2017-09-15 18:40:33 +02:00
This file is part of GCC.
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 3 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program. If not, see <http://www.gnu.org/licenses/>. */
/* Push namespace NAME as the current binding level, to which
newly-introduced decls will be bound. An empty string identifies
the global namespace, whereas NULL identifies an anonymous
namespace. A namespace named NAME is created in the current scope,
if needed.
If the newly-created namespace is to be an inline namespace, see
make_namespace_inline. */
GCC_METHOD1 (int /* bool */, push_namespace,
const char *) /* Argument NAME. */
/* Push TYPE as the current binding level, making its members visible
for name lookup. The current scope before the call must be the
scope in which the class was declared. This should be used if the
definition of a class is already finished, but one wishes to define
a nested class, or to enter the scope of one of its member
functions. */
GCC_METHOD1 (int /* bool */, push_class,
gcc_type) /* Argument TYPE. */
/* Push FUNCTION_DECL as the current (empty) binding level (see
reactivate_decl). The current enclosing scope before the call must
be the scope in which the function was declared. */
GCC_METHOD1 (int /* bool */, push_function,
gcc_decl) /* Argument FUNCTION_DECL. */
/* Make DECL visible (again?) within SCOPE. When SCOPE is NULL, it
means the current scope; if it is not NULL, it must name a function
that is currently active, even if not at the top of the binding
chain.
This function can be used to make e.g. a global function or
variable visible in a namespace or local scope (overriding another
enclosing definition of the same name), but its most common
expected use of this primitive, that gives it its name, is to make
declarations visible again after reentering a function scope,
because when a function is entered with push_function, that does
NOT make any of the declarations nested in it visible for name
lookup.
There is a reason/excuse for that: unlike namespaces and classes,
G++ doesn't ever have to reenter function scopes, so its name
resolution infrastructure is not prepared to do that. But wait,
there is also a good use for this apparent limitation: a function
may contain multiple scopes (blocks), and the name may be bound to
different symbols in each of these scopes. With this interface, as
we reenter a function scope, we may choose which symbols to make
visible for the code snippet, or, if there could be template
functions in local scopes, for unresolved names in nested template
class default arguments, or in nested template function signatures.
As for making a local declaration visible for the code snippet,
there are two possibilities: a) introduce it upfront, while
entering the scope for the user expression (see the enter_scope
callback, called by g++ when encountering the push_user_expression
pragma), which might save some scope switching and reactivate_decl
(though this can't be helped if some declarations have to be
introduced and discarded, because of multiple definitions of the
same name in different scopes within a function: they have to be
defined in discriminator order); or b) introduce it when its name
is looked up, entering the scope, introducing the declaration,
leaving the scope, and then reactivating the declaration in its
local scope.
Here's some more detail on how reactivate_decl works. Say there's
a function foo whose body looks like this:
{
{
// point 1
class c {} o __attribute__ ((__used__)); // c , o
}
struct c {
void f() {
// point 2
}
} o __attribute__ ((__used__)); // c_0, o_0
{
class c {} p __attribute__ ((__used__)); // c_1, p
// point 3
o.f();
}
}
When we are about to define class c at point 1, we enter the
function foo scope, and since no symbols are visible at point 1, we
proceed to declare class c. We may then define the class right
away, or, if we leave the function scope, and we later wish to
define it, or to define object o, we can reenter the scope and just
use the previously-obtained gcc_decl to define the class, without
having to reactivate the declaration.
Now, if we are to set up the binding context for point 2, we have
to define c_0::f, and in order to do so, we have to declare and
define c_0. Before we can declare c_0, we MUST at least declare c.
As a general rule, before we can declare or define any local name
with a discriminator, we have to at least declare any other
occurrences of the same name in the same enclosing entity with
lower or absent discriminator.
So, we declare c, then we leave the function scope and reenter it
so as to declare c_0 (also with name "c", which is why we have to
leave and reenter the function scope, otherwise we would get an
error because of the duplicate definition; g++ will assign a
discriminator because it still remembers there was an earlier
declaration of c_0 within the function, it's just no longer in
scope), then we can define c_0, including its member function f.
Likewise, if we wish to define o_0, we have to define o first. If
we wish to declare (and maybe then define) c_1, we have to at least
declare (c and then) c_0 first.
Then, as we set up the binding context to compile a code snippet at
point 3, we may choose to activate c_1, o_0 and p upfront,
declaring and discarding c, c_0 and o, and then reentering the
funciton scope to declare c_1, o_0 and p; or we can wait for oracle
lookups of c, o or p. If c is looked up, and the debugger resolves
c in the scope to c_1, it is expected to enter the function scope
from the top level, declare c, leave it, reenter it, declare c_0,
leave it, reenter it, declare c_1, leave it, and then reactivate
c_1 in the function scope. If c_1 is needed as a complete type,
the definition may be given right after the declaration, or the
scope will have to be reentered in order to define the class.
. If the code snippet is at point 2, we don't need to (re)activate
any declaration: nothing from any local scope is visible. Just
entering the scope of the class containing member function f
reactivates the names of its members, including the class name
itself. */
GCC_METHOD2 (int /* bool */, reactivate_decl,
gcc_decl, /* Argument DECL. */
gcc_decl) /* Argument SCOPE. */
/* Pop the namespace last entered with push_namespace, or class last
entered with push_class, or function last entered with
push_function, restoring the binding level in effect before the
matching push_* call. */
GCC_METHOD0 (int /* bool */, pop_binding_level)
/* Return the NAMESPACE_DECL, TYPE_DECL or FUNCTION_DECL of the
binding level that would be popped by pop_scope. */
GCC_METHOD0 (gcc_decl, get_current_binding_level_decl)
/* Make the current binding level an inline namespace. It must be a
namespace to begin with. It is safe to call this more than once
for the same namespace, but after the first call, subsequent ones
will not return a success status. */
GCC_METHOD0 (int /* bool */, make_namespace_inline)
/* Add USED_NS to the namespaces used by the current binding level.
Use get_current_binding_level_decl to obtain USED_NS's
gcc_decl. */
GCC_METHOD1 (int /* bool */, add_using_namespace,
gcc_decl) /* Argument USED_NS. */
/* Introduce a namespace alias declaration, as in:
namespace foo = [... ::] bar;
After this call, namespace TARGET will be visible as ALIAS within
the current namespace. Get the declaration for TARGET by calling
get_current_binding_level_decl after pushing into it. */
GCC_METHOD2 (int /* bool */, add_namespace_alias,
const char *, /* Argument ALIAS. */
gcc_decl) /* Argument TARGET. */
/* Introduce a using declaration, as in:
using foo::bar;
The TARGET decl names the qualifying scope (foo:: above) and the
identifier (bar), but that does not mean that only TARGET will be
brought into the current scope: all bindings of TARGET's identifier
in the qualifying scope will be brought in.
FLAGS should specify GCC_CP_SYMBOL_USING. If the current scope is
a class scope, visibility flags must be supplied.
Even when TARGET is template dependent, we don't need to specify
whether or not it is a typename: the supplied declaration (that
could be a template-dependent type converted to declaration by
get_type_decl) indicates so. */
GCC_METHOD2 (int /* bool */, add_using_decl,
enum gcc_cp_symbol_kind, /* Argument FLAGS. */
gcc_decl) /* Argument TARGET. */
/* Create a new "decl" in GCC, and bind it in the current binding
level. A decl is a declaration, basically a kind of symbol.
NAME is the name of the new symbol. SYM_KIND is the kind of
symbol being requested. SYM_TYPE is the new symbol's C++ type;
except for labels, where this is not meaningful and should be
zero. If SUBSTITUTION_NAME is not NULL, then a reference to this
decl in the source will later be substituted with a dereference
of a variable of the given name. Otherwise, for symbols having
an address (e.g., functions), ADDRESS is the address. FILENAME
and LINE_NUMBER refer to the symbol's source location. If this
is not known, FILENAME can be NULL and LINE_NUMBER can be 0.
This function returns the new decl.
Use this function to register typedefs, functions and variables to
namespace and local binding levels, and typedefs, member functions
(static or not), and static data members to class binding levels.
Class members must have their access controls specified with
GCC_CP_ACCESS_* flags in SYM_KIND.
Note that, since access controls are disabled, we have no means to
express private, protected and public.
There are various flags that can be set in SYM_KIND to specify
additional semantics. Look for GCC_CP_FLAGs in the definition of
enum gcc_cp_symbol_kind in gcc-cp-interface.h.
In order to define member functions, pass GCC_CP_SYMBOL_FUNCTION in
SYM_KIND, and a function_type for static member functions or a
method type for non-static member functions, including constructors
and destructors. Use build_function_type to create a function
type; for a method type, start by creating a function type without
any compiler-introduced artificial arguments (the implicit this
pointer, and the __in_chrg added to constructors and destructors,
and __vtt_parm added to the former), and then use build_method_type
to create the method type out of the class type and the function
type.
For operator functions, set GCC_CP_FLAG_SPECIAL_FUNCTION in
SYM_KIND, in addition to any other applicable flags, and pass as
NAME a string starting with the two-character mangling for operator
name: "ps" for unary plus, "mL" for multiply and assign, *=; etc.
Use "cv" for type converstion operators (the target type portion
may be omitted, as it is taken from the return type in SYM_TYPE).
For operator"", use "li" followed by the identifier (the mangled
name mandates digits specifying the length of the identifier; if
present, they determine the end of the identifier, otherwise, the
identifier extents to the end of the string, so that "li3_Kme" and
"li_Km" are equivalent).
Constructors and destructors need special care, because for each
constructor and destructor there may be multiple clones defined
internally by the compiler. With build_decl, you can introduce the
base declaration of a constructor or a destructor, setting
GCC_CP_FLAG_SPECIAL_FUNCTION the flag and using names starting with
capital "C" or "D", respectively, followed by a digit (see below),
a blank, or NUL ('\0'). DO NOT supply an ADDRESS or a
SUBSTITUTION_NAME to build_decl, it would be meaningless (and
rejected) for the base declaration; use define_cdtor_clone to
introduce the address of each clone. For constructor templates,
declare the template with build_decl, and then, for each
specialization, introduce it with
build_function_template_specialization, and then define the
addresses of each of its clones with define_cdtor_clone.
NAMEs for GCC_CP_FLAG_SPECIAL_FUNCTION:
NAME meaning
C? constructor base declaration (? may be 1, 2, 4, blank or NUL)
D? destructor base declaration (? may be 0, 1, 2, 4, blank or NUL)
nw operator new
na operator new[]
dl operator delete
da operator delete[]
ps operator + (unary)
ng operator - (unary)
ad operator & (unary)
de operator * (unary)
co operator ~
pl operator +
mi operator -
ml operator *
dv operator /
rm operator %
an operator &
or operator |
eo operator ^
aS operator =
pL operator +=
mI operator -=
mL operator *=
dV operator /=
rM operator %=
aN operator &=
oR operator |=
eO operator ^=
ls operator <<
rs operator >>
lS operator <<=
rS operator >>=
eq operator ==
ne operator !=
lt operator <
gt operator >
le operator <=
ge operator >=
nt operator !
aa operator &&
oo operator ||
pp operator ++
mm operator --
cm operator ,
pm operator ->*
pt operator ->
cl operator ()
ix operator []
qu operator ?
cv operator <T> (conversion operator)
li<id> operator "" <id>
FIXME: How about attributes? */
GCC_METHOD7 (gcc_decl, build_decl,
const char *, /* Argument NAME. */
enum gcc_cp_symbol_kind, /* Argument SYM_KIND. */
gcc_type, /* Argument SYM_TYPE. */
const char *, /* Argument SUBSTITUTION_NAME. */
gcc_address, /* Argument ADDRESS. */
const char *, /* Argument FILENAME. */
unsigned int) /* Argument LINE_NUMBER. */
/* Supply the ADDRESS of one of the multiple clones of constructor or
destructor CDTOR. The clone is specified by NAME, using the
following name mangling conventions:
C1 in-charge constructor
C2 not-in-charge constructor
C4 unified constructor
D0 deleting destructor
D1 in-charge destructor
D2 not-in-charge destructor
D4 unified destructor
The following information is not necessary to use the API.
C1 initializes an instance of the class (rather than of derived
classes), including virtual base classes, whereas C2 initializes a
sub-object (of the given class type) of an instance of some derived
class (or a full object that doesn't have any virtual base
classes).
D0 and D1 destruct an instance of the class, including virtual base
classes, but only the former calls operator delete to release the
object's storage at the end; D2 destructs a sub-object (of the
given class type) of an instance of a derived class (or a full
object that doesn't have any virtual base classes).
The [CD]4 manglings (and symbol definitions) are non-standard, but
GCC uses them in some cases: rather than assuming they are
in-charge or not-in-charge, they test the implicit argument that
the others ignore to tell how to behave. These are used instead of
cloning when we just can't use aliases. */
GCC_METHOD3 (gcc_decl, define_cdtor_clone,
const char *, /* Argument NAME. */
gcc_decl, /* Argument CDTOR. */
gcc_address) /* Argument ADDRESS. */
/* Return the type associated with the given declaration. This is
most useful to obtain the type associated with a forward-declared
class, because it is the gcc_type, rather than the gcc_decl, that
has to be used to build other types, but build_decl returns a
gcc_decl rather than a gcc_type. This call can in theory be used
to obtain the type from any other declaration; it is supposed to
return the same type that was supplied when the declaration was
created. */
GCC_METHOD1 (gcc_type, get_decl_type,
gcc_decl) /* Argument DECL. */
/* Return the declaration for a type. */
GCC_METHOD1 (gcc_decl, get_type_decl,
gcc_type) /* Argument TYPE. */
/* Declare DECL as a friend of the current class scope, if TYPE is
NULL, or of TYPE itself otherwise. DECL may be a function or a
class, be they template generics, template specializations or not
templates. TYPE must be a class type (not a template generic).
The add_friend call cannot introduce a declaration; even if the
friend is first declared as a friend in the source code, the
declaration belongs in the enclosing namespace, so it must be
introduced in that namespace, and the resulting declaration can
then be made a friend.
DECL cannot, however, be a member of a template class generic,
because we have no means to introduce their declarations. This
interface has no notion of definitions for template generics. As a
consequence, users of this interface must introduce each friend
template member specialization separately, i.e., instead of:
template <typename T> friend struct X<T>::M;
they must be declared as if they were:
friend struct X<onetype>::M;
friend struct X<anothertype>::M;
... for each specialization of X.
Specializations of a template can have each others' members as
friends:
template <typename T> class foo {
int f();
template <typename U> friend int foo<U>::f();
};
It wouldn't always be possible to define all specializations of a
template class before introducing the friend declarations in their
expanded, per-specialization form.
In order to simplify such friend declarations, and to enable
incremental friend declarations as template specializations are
introduced, add_friend can be called after the befriending class is
fully defined, passing it a non-NULL TYPE argument naming the
befriending class type. */
GCC_METHOD2 (int /* bool */, add_friend,
gcc_decl, /* Argument DECL. */
gcc_type) /* Argument TYPE. */
/* Return the type of a pointer to a given base type. */
GCC_METHOD1 (gcc_type, build_pointer_type,
gcc_type) /* Argument BASE_TYPE. */
/* Return the type of a reference to a given base type. */
GCC_METHOD2 (gcc_type, build_reference_type,
gcc_type, /* Argument BASE_TYPE. */
enum gcc_cp_ref_qualifiers) /* Argument RQUALS. */
/* Create a new pointer-to-member type. MEMBER_TYPE is the data
member type, while CLASS_TYPE is the class type containing the data
member. For pointers to member functions, MEMBER_TYPE must be a
method type, and CLASS_TYPE must be specified even though it might
be possible to extract it from the method type. */
GCC_METHOD2 (gcc_type, build_pointer_to_member_type,
gcc_type, /* Argument CLASS_TYPE. */
gcc_type) /* Argument MEMBER_TYPE. */
/* Start a template parameter list scope and enters it, so that
subsequent build_type_template_parameter and
build_value_template_parameter calls create template parameters in
the list. The list is closed by a build_decl call with
GCC_CP_SYMBOL_FUNCTION or GCC_CP_SYMBOL_CLASS, that, when the scope
is a template parameter list, declares a template function or a
template class with the then-closed parameter list. The scope in
which the new declaration is to be introduced by build_decl must be
entered before calling start_template_decl, and build_decl returns
to that scope, from the template parameter list scope, before
introducing the declaration. */
GCC_METHOD0 (int /* bool */, start_template_decl)
/* Build a typename template-parameter (e.g., the T in template
<typename T = X>). Either PACK_P should be nonzero, to indicate an
argument pack (the last argument in a variadic template argument
list, as in template <typename... T>), or DEFAULT_TYPE may be
non-NULL to set the default type argument (e.g. X) for the template
parameter. FILENAME and LINE_NUMBER may specify the source
location in which the template parameter was declared. */
GCC_METHOD5 (gcc_type, build_type_template_parameter,
const char *, /* Argument ID. */
int /* bool */, /* Argument PACK_P. */
gcc_type, /* Argument DEFAULT_TYPE. */
const char *, /* Argument FILENAME. */
unsigned int) /* Argument LINE_NUMBER. */
/* Build a template template-parameter (e.g., the T in template
<template <[...]> class T = X>). DEFAULT_TEMPL may be non-NULL to
set the default type-template argument (e.g. X) for the template
template parameter. FILENAME and LINE_NUMBER may specify the
source location in which the template parameter was declared. */
GCC_METHOD5 (gcc_utempl, build_template_template_parameter,
const char *, /* Argument ID. */
int /* bool */, /* Argument PACK_P. */
gcc_utempl, /* Argument DEFAULT_TEMPL. */
const char *, /* Argument FILENAME. */
unsigned int) /* Argument LINE_NUMBER. */
/* Build a value template-parameter (e.g., the V in template <typename
T, T V> or in template <int V = X>). DEFAULT_VALUE may be non-NULL
to set the default value argument for the template parameter (e.g.,
X). FILENAME and LINE_NUMBER may specify the source location in
which the template parameter was declared. */
GCC_METHOD5 (gcc_decl, build_value_template_parameter,
gcc_type, /* Argument TYPE. */
const char *, /* Argument ID. */
gcc_expr, /* Argument DEFAULT_VALUE. */
const char *, /* Argument FILENAME. */
unsigned int) /* Argument LINE_NUMBER. */
/* Build a template-dependent typename (e.g., typename T::bar or
typename T::template bart<X>). ENCLOSING_TYPE should be the
template-dependent nested name specifier (e.g., T), ID should be
the name of the member of the ENCLOSING_TYPE (e.g., bar or bart),
and TARGS should be non-NULL and specify the template arguments
(e.g. <X>) iff ID is to name a class template.
In this and other calls, a template-dependent nested name specifier
may be a template class parameter (build_type_template_parameter),
a specialization (returned by build_dependent_type_template_id) of
a template template parameter (returned by
build_template_template_parameter) or a member type thereof
(returned by build_dependent_typename itself). */
GCC_METHOD3 (gcc_type, build_dependent_typename,
gcc_type, /* Argument ENCLOSING_TYPE. */
const char *, /* Argument ID. */
const struct gcc_cp_template_args *) /* Argument TARGS. */
/* Build a template-dependent class template (e.g., T::template bart).
ENCLOSING_TYPE should be the template-dependent nested name
specifier (e.g., T), ID should be the name of the class template
member of the ENCLOSING_TYPE (e.g., bart). */
GCC_METHOD2 (gcc_utempl, build_dependent_class_template,
gcc_type, /* Argument ENCLOSING_TYPE. */
const char *) /* Argument ID. */
/* Build a template-dependent type template-id (e.g., T<A>).
TEMPLATE_DECL should be a template template parameter (e.g., the T
in template <template <[...]> class T = X>), and TARGS should
specify the template arguments (e.g. <A>). */
GCC_METHOD2 (gcc_type, build_dependent_type_template_id,
gcc_utempl, /* Argument TEMPLATE_DECL. */
const struct gcc_cp_template_args *) /* Argument TARGS. */
/* Build a template-dependent expression (e.g., S::val or S::template
mtf<X>, or unqualified f or template tf<X>).
ENCLOSING_SCOPE should be a template-dependent nested name
specifier (e.g., T), a resolved namespace or class decl, or NULL
for unqualified names; ID should be the name of the member of the
ENCLOSING_SCOPE (e.g., val or mtf) or unqualified overloaded
function; and TARGS should list template arguments (e.g. <X>) when
mtf or tf are to name a template function, or be NULL otherwise.
Unqualified names and namespace- or class-qualified names can only
resolve to overloaded functions, to be used in contexts that
involve overload resolution that cannot be resolved because of
template-dependent argument or return types, such as call
expressions with template-dependent arguments, conversion
expressions to function types with template-dependent argument
types or the like. Other cases of unqualified or
non-template-dependent-qualified names should NOT use this
function, and use decl_expr to convert the appropriate function or
object declaration to an expression.
If ID is the name of a special member function, FLAGS should be
GCC_CP_SYMBOL_FUNCTION|GCC_CP_FLAG_SPECIAL_FUNCTION, and ID should
be one of the encodings for special member functions documented in
build_decl. Otherwise, FLAGS should be GCC_CP_SYMBOL_MASK, which
suggests the symbol kind is not known (though we know it is not a
type).
If ID denotes a conversion operator, CONV_TYPE should name the
target type of the conversion. Otherwise, CONV_TYPE must be
NULL. */
GCC_METHOD5 (gcc_expr, build_dependent_expr,
gcc_decl, /* Argument ENCLOSING_SCOPE. */
enum gcc_cp_symbol_kind, /* Argument FLAGS. */
const char *, /* Argument NAME. */
gcc_type, /* Argument CONV_TYPE. */
const struct gcc_cp_template_args *) /* Argument TARGS. */
/* Build a gcc_expr for the value VALUE in type TYPE. */
GCC_METHOD2 (gcc_expr, build_literal_expr,
gcc_type, /* Argument TYPE. */
unsigned long) /* Argument VALUE. */
/* Build a gcc_expr that denotes DECL, the declaration of a variable
or function in namespace scope, or of a static member variable or
function. Use QUALIFIED_P to build the operand of unary & so as to
compute a pointer-to-member, rather than a regular pointer. */
GCC_METHOD2 (gcc_expr, build_decl_expr,
gcc_decl, /* Argument DECL. */
int /* bool */) /* Argument QUALIFIED_P. */
/* Build a gcc_expr that denotes the unary operation UNARY_OP applied
to the gcc_expr OPERAND. For non-expr operands, see
unary_type_expr. Besides the UNARY_OP encodings used for operator
names, we support "pp_" for preincrement, and "mm_" for
predecrement, "nx" for noexcept, "tw" for throw, "tr" for rethrow
(pass NULL as the operand), "te" for typeid, "sz" for sizeof, "az"
for alignof, "dl" for delete, "gsdl" for ::delete, "da" for
delete[], "gsda" for ::delete[], "sp" for pack expansion, "sZ" for
sizeof...(function argument pack). */
GCC_METHOD2 (gcc_expr, build_unary_expr,
const char *, /* Argument UNARY_OP. */
gcc_expr) /* Argument OPERAND. */
/* Build a gcc_expr that denotes the binary operation BINARY_OP
applied to gcc_exprs OPERAND1 and OPERAND2. Besides the BINARY_OP
encodings used for operator names, we support "ds" for the operator
token ".*" and "dt" for the operator token ".". When using
operators that take a name as their second operand ("." and "->")
use decl_expr to convert the gcc_decl of the member name to a
gcc_expr, if the member name wasn't created with
e.g. build_dependent_expr. */
GCC_METHOD3 (gcc_expr, build_binary_expr,
const char *, /* Argument BINARY_OP. */
gcc_expr, /* Argument OPERAND1. */
gcc_expr) /* Argument OPERAND2. */
/* Build a gcc_expr that denotes the ternary operation TERNARY_OP
applied to gcc_exprs OPERAND1, OPERAND2 and OPERAND3. The only
supported TERNARY_OP is "qu", for the "?:" operator. */
GCC_METHOD4 (gcc_expr, build_ternary_expr,
const char *, /* Argument TERNARY_OP. */
gcc_expr, /* Argument OPERAND1. */
gcc_expr, /* Argument OPERAND2. */
gcc_expr) /* Argument OPERAND3. */
/* Build a gcc_expr that denotes the unary operation UNARY_OP applied
to the gcc_type OPERAND. Supported unary operations taking types
are "ti" for typeid, "st" for sizeof, "at" for alignof, and "sZ"
for sizeof...(template argument pack). */
GCC_METHOD2 (gcc_expr, build_unary_type_expr,
const char *, /* Argument UNARY_OP. */
gcc_type) /* Argument OPERAND. */
/* Build a gcc_expr that denotes the binary operation BINARY_OP
applied to gcc_type OPERAND1 and gcc_expr OPERAND2. Use this for
all kinds of (single-argument) type casts ("dc", "sc", "cc", "rc"
for dynamic, static, const and reinterpret casts, respectively;
"cv" for functional or C-style casts). */
GCC_METHOD3 (gcc_expr, build_cast_expr,
const char *, /* Argument BINARY_OP. */
gcc_type, /* Argument OPERAND1. */
gcc_expr) /* Argument OPERAND2. */
/* Build a gcc_expr that denotes the conversion of an expression list
VALUES to TYPE, with ("tl") or without ("cv") braces, or a braced
initializer list of unspecified type (e.g., a component of another
braced initializer list; pass "il" for CONV_OP, and NULL for
TYPE). */
GCC_METHOD3 (gcc_expr, build_expression_list_expr,
const char *, /* Argument CONV_OP. */
gcc_type, /* Argument TYPE. */
const struct gcc_cp_function_args *) /* Argument VALUES. */
/* Build a gcc_expr that denotes a new ("nw") or new[] ("na")
expression of TYPE, with or without a GLOBAL_NS qualifier (prefix
the NEW_OP with "gs"), with or without PLACEMENT, with or without
INITIALIZER. If it's not a placement new, PLACEMENT must be NULL
(rather than a zero-length placement arg list). If there's no
specified initializer, INITIALIZER must be NULL; a zero-length arg
list stands for a default initializer. */
GCC_METHOD4 (gcc_expr, build_new_expr,
const char *, /* Argument NEW_OP. */
const struct gcc_cp_function_args *, /* Argument PLACEMENT. */
gcc_type, /* Argument TYPE. */
const struct gcc_cp_function_args *) /* Argument INITIALIZER. */
/* Return a call expression that calls CALLABLE with arguments ARGS.
CALLABLE may be a function, a callable object, a pointer to
function, an unresolved expression, an unresolved overload set, an
object expression combined with a member function overload set or a
pointer-to-member. If QUALIFIED_P, CALLABLE will be interpreted as
a qualified name, preventing virtual function dispatch. */
GCC_METHOD3 (gcc_expr, build_call_expr,
gcc_expr, /* Argument CALLABLE. */
int /* bool */, /* Argument QUALIFIED_P. */
const struct gcc_cp_function_args *) /* Argument ARGS. */
/* Return the type of the gcc_expr OPERAND.
Use this for decltype.
For decltype (auto), pass a NULL OPERAND.
Note: for template-dependent expressions, the result is NULL,
because the type is only computed when template argument
substitution is performed. */
GCC_METHOD1 (gcc_type, get_expr_type,
gcc_expr) /* Argument OPERAND. */
/* Introduce a specialization of a template function.
TEMPLATE_DECL is the template function, and TARGS are the arguments
for the specialization. ADDRESS is the address of the
specialization. FILENAME and LINE_NUMBER specify the source
location associated with the template function specialization. */
GCC_METHOD5 (gcc_decl, build_function_template_specialization,
gcc_decl, /* Argument TEMPLATE_DECL. */
const struct gcc_cp_template_args *, /* Argument TARGS. */
gcc_address, /* Argument ADDRESS. */
const char *, /* Argument FILENAME. */
unsigned int) /* Argument LINE_NUMBER. */
/* Specialize a template class as an incomplete type. A definition
can be supplied later, with start_class_type.
TEMPLATE_DECL is the template class, and TARGS are the arguments
for the specialization. FILENAME and LINE_NUMBER specify the
source location associated with the template class
specialization. */
GCC_METHOD4 (gcc_decl, build_class_template_specialization,
gcc_decl, /* Argument TEMPLATE_DECL. */
const struct gcc_cp_template_args *, /* Argument TARGS. */
const char *, /* Argument FILENAME. */
unsigned int) /* Argument LINE_NUMBER. */
/* Start defining a 'class', 'struct' or 'union' type, entering its
own binding level. Initially it has no fields.
TYPEDECL is the forward-declaration of the type, returned by
build_decl. BASE_CLASSES indicate the base classes of class NAME.
FILENAME and LINE_NUMBER specify the source location associated
with the class definition, should they be different from those of
the forward declaration. */
GCC_METHOD4 (gcc_type, start_class_type,
gcc_decl, /* Argument TYPEDECL. */
const struct gcc_vbase_array *,/* Argument BASE_CLASSES. */
const char *, /* Argument FILENAME. */
unsigned int) /* Argument LINE_NUMBER. */
/* Create a new closure class type, record it as the
DISCRIMINATOR-numbered closure type in the current scope (or
associated with EXTRA_SCOPE, if non-NULL), and enter the closure
type's own binding level. This primitive would sort of combine
build_decl and start_class_type, if they could be used to introduce
a closure type. Initially it has no fields.
FILENAME and LINE_NUMBER specify the source location associated
with the class. EXTRA_SCOPE, if non-NULL, must be a PARM_DECL of
the current function, or a FIELD_DECL of the current class. If it
is NULL, the current scope must be a function. */
GCC_METHOD5 (gcc_type, start_closure_class_type,
int, /* Argument DISCRIMINATOR. */
gcc_decl, /* Argument EXTRA_SCOPE. */
enum gcc_cp_symbol_kind, /* Argument FLAGS. */
const char *, /* Argument FILENAME. */
unsigned int) /* Argument LINE_NUMBER. */
/* Add a non-static data member to the most-recently-started
unfinished struct or union type. FIELD_NAME is the field's name.
FIELD_TYPE is the type of the field. BITSIZE and BITPOS indicate
where in the struct the field occurs. */
GCC_METHOD5 (gcc_decl, build_field,
const char *, /* Argument FIELD_NAME. */
gcc_type, /* Argument FIELD_TYPE. */
enum gcc_cp_symbol_kind, /* Argument FIELD_FLAGS. */
unsigned long, /* Argument BITSIZE. */
unsigned long) /* Argument BITPOS. */
/* After all the fields have been added to a struct, class or union,
the struct or union type must be "finished". This does some final
cleanups in GCC, and pops to the binding level that was in effect
before the matching start_class_type or
start_closure_class_type. */
GCC_METHOD1 (int /* bool */, finish_class_type,
unsigned long) /* Argument SIZE_IN_BYTES. */
/* Create a new 'enum' type, and record it in the current binding
level. The new type initially has no associated constants.
NAME is the enum name. FILENAME and LINE_NUMBER specify its source
location. */
GCC_METHOD5 (gcc_type, start_enum_type,
const char *, /* Argument NAME. */
gcc_type, /* Argument UNDERLYING_INT_TYPE. */
enum gcc_cp_symbol_kind, /* Argument FLAGS. */
const char *, /* Argument FILENAME. */
unsigned int) /* Argument LINE_NUMBER. */
/* Add a new constant to an enum type. NAME is the constant's name
and VALUE is its value. Returns a gcc_decl for the constant. */
GCC_METHOD3 (gcc_decl, build_enum_constant,
gcc_type, /* Argument ENUM_TYPE. */
const char *, /* Argument NAME. */
unsigned long) /* Argument VALUE. */
/* After all the constants have been added to an enum, the type must
be "finished". This does some final cleanups in GCC. */
GCC_METHOD1 (int /* bool */, finish_enum_type,
gcc_type) /* Argument ENUM_TYPE. */
/* Create a new function type. RETURN_TYPE is the type returned by
the function, and ARGUMENT_TYPES is a vector, of length NARGS, of
the argument types. IS_VARARGS is true if the function is
varargs. */
GCC_METHOD3 (gcc_type, build_function_type,
gcc_type, /* Argument RETURN_TYPE. */
const struct gcc_type_array *,/* Argument ARGUMENT_TYPES. */
int /* bool */) /* Argument IS_VARARGS. */
/* Create a variant of a function type with an exception
specification. FUNCTION_TYPE is a function or method type.
EXCEPT_TYPES is an array with the list of exception types. Zero as
the array length implies throw() AKA noexcept(true); NULL as the
pointer to gcc_type_array implies noexcept(false), which is almost
equivalent (but distinguishable by the compiler) to an unspecified
exception list. */
GCC_METHOD2 (gcc_type, build_exception_spec_variant,
gcc_type, /* Argument FUNCTION_TYPE. */
const struct gcc_type_array *)/* Argument EXCEPT_TYPES. */
/* Create a new non-static member function type. FUNC_TYPE is the
method prototype, without the implicit THIS pointer, added as a
pointer to the QUALS-qualified CLASS_TYPE. If CLASS_TYPE is NULL,
this creates a cv-qualified (member) function type not associated
with any specific class, as needed to support "typedef void f(int)
const;", which can later be used to declare member functions and
pointers to member functions. */
GCC_METHOD4 (gcc_type, build_method_type,
gcc_type, /* Argument CLASS_TYPE. */
gcc_type, /* Argument FUNC_TYPE. */
enum gcc_cp_qualifiers, /* Argument QUALS. */
enum gcc_cp_ref_qualifiers) /* Argument RQUALS. */
/* Return a declaration for the (INDEX - 1)th argument of
FUNCTION_DECL, i.e., for the first argument, use zero as the index.
If FUNCTION_DECL is a non-static member function, use -1 to get the
implicit THIS parameter. */
GCC_METHOD2 (gcc_decl, get_function_parameter_decl,
gcc_decl, /* Argument FUNCTION_DECL. */
int) /* Argument INDEX. */
/* Return a lambda expr that constructs an instance of CLOSURE_TYPE.
Only lambda exprs without any captures can be correctly created
through these mechanisms; that's all we need to support lambdas
expressions in default parameters, the only kind that may have to
be introduced through this interface. */
GCC_METHOD1 (gcc_expr, build_lambda_expr,
gcc_type) /* Argument CLOSURE_TYPE. */
/* Return an integer type with the given properties. If BUILTIN_NAME
is non-NULL, it must name a builtin integral type with the given
signedness and size, and that is the type that will be returned. */
GCC_METHOD3 (gcc_type, get_int_type,
int /* bool */, /* Argument IS_UNSIGNED. */
unsigned long, /* Argument SIZE_IN_BYTES. */
const char *) /* Argument BUILTIN_NAME. */
/* Return the 'char' type, a distinct type from both 'signed char' and
'unsigned char' returned by int_type. */
GCC_METHOD0 (gcc_type, get_char_type)
/* Return a floating point type with the given properties. If BUILTIN_NAME
is non-NULL, it must name a builtin integral type with the given
signedness and size, and that is the type that will be returned. */
GCC_METHOD2 (gcc_type, get_float_type,
unsigned long, /* Argument SIZE_IN_BYTES. */
const char *) /* Argument BUILTIN_NAME. */
/* Return the 'void' type. */
GCC_METHOD0 (gcc_type, get_void_type)
/* Return the 'bool' type. */
GCC_METHOD0 (gcc_type, get_bool_type)
/* Return the std::nullptr_t type. */
GCC_METHOD0 (gcc_type, get_nullptr_type)
/* Return the nullptr constant. */
GCC_METHOD0 (gcc_expr, get_nullptr_constant)
/* Create a new array type. If NUM_ELEMENTS is -1, then the array
is assumed to have an unknown length. */
GCC_METHOD2 (gcc_type, build_array_type,
gcc_type, /* Argument ELEMENT_TYPE. */
int) /* Argument NUM_ELEMENTS. */
/* Create a new array type. NUM_ELEMENTS is a template-dependent
expression. */
GCC_METHOD2 (gcc_type, build_dependent_array_type,
gcc_type, /* Argument ELEMENT_TYPE. */
gcc_expr) /* Argument NUM_ELEMENTS. */
/* Create a new variably-sized array type. UPPER_BOUND_NAME is the
name of a local variable that holds the upper bound of the array;
it is one less than the array size. */
GCC_METHOD2 (gcc_type, build_vla_array_type,
gcc_type, /* Argument ELEMENT_TYPE. */
const char *) /* Argument UPPER_BOUND_NAME. */
/* Return a qualified variant of a given base type. QUALIFIERS says
which qualifiers to use; it is composed of or'd together
constants from 'enum gcc_cp_qualifiers'. */
GCC_METHOD2 (gcc_type, build_qualified_type,
gcc_type, /* Argument UNQUALIFIED_TYPE. */
enum gcc_cp_qualifiers) /* Argument QUALIFIERS. */
/* Build a complex type given its element type. */
GCC_METHOD1 (gcc_type, build_complex_type,
gcc_type) /* Argument ELEMENT_TYPE. */
/* Build a vector type given its element type and number of
elements. */
GCC_METHOD2 (gcc_type, build_vector_type,
gcc_type, /* Argument ELEMENT_TYPE. */
int) /* Argument NUM_ELEMENTS. */
/* Build a constant. NAME is the constant's name and VALUE is its
value. FILENAME and LINE_NUMBER refer to the type's source
location. If this is not known, FILENAME can be NULL and
LINE_NUMBER can be 0. */
GCC_METHOD5 (int /* bool */, build_constant,
gcc_type, /* Argument TYPE. */
const char *, /* Argument NAME. */
unsigned long, /* Argument VALUE. */
const char *, /* Argument FILENAME. */
unsigned int) /* Argument LINE_NUMBER. */
/* Emit an error and return an error type object. */
GCC_METHOD1 (gcc_type, error,
const char *) /* Argument MESSAGE. */
/* Declare a static_assert with the given CONDITION and ERRORMSG at
FILENAME:LINE_NUMBER. */
GCC_METHOD4 (int /* bool */, add_static_assert,
gcc_expr, /* Argument CONDITION. */
const char *, /* Argument ERRORMSG. */
const char *, /* Argument FILENAME. */
unsigned int) /* Argument LINE_NUMBER. */
#if 0
/* FIXME: We don't want to expose the internal implementation detail
that default parms are stored in function types, and it's not clear
how this or other approaches would interact with the type sharing
of e.g. ctor clones, so we're leaving this out, since default args
are not even present in debug information anyway. Besides, the set
of default args for a function may grow within its scope, and vary
independently in other scopes. */
/* Create a modified version of a function type that has default
values for some of its arguments. The returned type should ONLY be
used to define functions or methods, never to declare parameters,
variables, types or the like.
DEFAULTS must have at most as many N_ELEMENTS as there are
arguments without default values in FUNCTION_TYPE. Say, if
FUNCTION_TYPE has an argument list such as (T1, T2, T3, T4 = V0)
and DEFAULTS has 2 elements (V1, V2), the returned type will have
the following argument list: (T1, T2 = V1, T3 = V2, T4 = V0).
Any NULL expressions in DEFAULTS will be marked as deferred, and
they should be filled in with set_deferred_function_default_args. */
GCC_METHOD2 (gcc_type, add_function_default_args,
gcc_type, /* Argument FUNCTION_TYPE. */
const struct gcc_cp_function_args *) /* Argument DEFAULTS. */
/* Fill in the first deferred default args in FUNCTION_DECL with the
expressions given in DEFAULTS. This can be used when the
declaration of a parameter is needed to create a default
expression, such as taking the size of an earlier parameter, or
building a lambda expression in the parameter's context. */
GCC_METHOD2 (int /* bool */, set_deferred_function_default_args,
gcc_decl, /* Argument FUNCTION_DECL. */
const struct gcc_cp_function_args *) /* Argument DEFAULTS. */
#endif
/* When you add entry points, add them at the end, so that the new API
version remains compatible with the old version.
The following conventions have been observed as to naming entry points:
- build_* creates (and maybe records) something and returns it;
- add_* creates and records something, but doesn't return it;
- get_* obtains something without creating it;
- start_* marks the beginning of a compound (type, list, ...);
- finish_* completes the compound when needed.
Entry points that return an int (bool) and don't have a return value
specification return nonzero (true) on success and zero (false) on
failure. This is in line with libcc1's conventions of returning a
zero-initialized value in case of e.g. a transport error. */