// Returns the effective closure type for this handler (which will propagate
// from outer frames if this frame has no START* handler). Not implemented for
// UPB_HANDLER_STRING at the moment since this is not needed. Returns NULL is
// the effective closure type is unspecified (either no handler was registered
// to specify it or the handler that was registered did not specify the closure
// type).
const void *effective_closure_type(upb_handlers *h, const upb_fielddef *f,
upb_handlertype_t type) {
assert(type != UPB_HANDLER_STRING);
const void *ret = h->top_closure_type;
upb_selector_t sel;
if (upb_fielddef_isseq(f) &&
type != UPB_HANDLER_STARTSEQ &&
type != UPB_HANDLER_ENDSEQ &&
h->table[sel = getsel(h, f, UPB_HANDLER_STARTSEQ)].func) {
ret = upb_handlerattr_returnclosuretype(&h->table[sel].attr);
}
if (type == UPB_HANDLER_STRING &&
h->table[sel = getsel(h, f, UPB_HANDLER_STARTSTR)].func) {
ret = upb_handlerattr_returnclosuretype(&h->table[sel].attr);
}
// The effective type of the submessage; not used yet.
// if (type == SUBMESSAGE &&
// h->table[sel = getsel(h, f, UPB_HANDLER_STARTSUBMSG)].func) {
// ret = upb_handlerattr_returnclosuretype(&h->table[sel].attr);
// }
return ret;
}
/**
* lupb_msg_index
*
* Handles:
* msg.foo
* msg["foo"]
* msg[field_descriptor] # (for extensions) (TODO)
*/
static int lupb_msg_index(lua_State *L) {
lupb_msg *lmsg = lupb_msg_check(L, 1);
const upb_fielddef *f = lupb_msg_checkfield(L, lmsg, 2);
const upb_msglayout *l = lmsg->lmsgclass->layout;
if (in_userval(f)) {
lupb_uservalgeti(L, 1, lupb_fieldindex(f));
if (lua_isnil(L, -1)) {
/* Check if we need to lazily create wrapper. */
if (upb_fielddef_isseq(f)) {
/* TODO(haberman) */
} else if (upb_fielddef_issubmsg(f)) {
/* TODO(haberman) */
} else {
UPB_ASSERT(upb_fielddef_isstring(f));
if (upb_msg_has(lmsg->msg, f, l)) {
upb_msgval val = upb_msg_get(lmsg->msg, f, l);
lua_pop(L, 1);
lua_pushlstring(L, val.str.ptr, val.str.len);
lupb_uservalseti(L, 1, lupb_fieldindex(f), -1);
}
}
}
} else {
lupb_pushmsgval(L, upb_fielddef_type(f), upb_msg_get(lmsg->msg, f, l));
}
return 1;
}
/**
* lupb_msg_newindex()
*
* Handles:
* msg.foo = bar
* msg["foo"] = bar
* msg[field_descriptor] = bar # (for extensions) (TODO)
*/
static int lupb_msg_newindex(lua_State *L) {
lupb_msg *lmsg = lupb_msg_check(L, 1);
const upb_fielddef *f = lupb_msg_checkfield(L, lmsg, 2);
upb_fieldtype_t type = upb_fielddef_type(f);
upb_msgval msgval;
/* Typecheck and get msgval. */
if (upb_fielddef_isseq(f)) {
msgval = lupb_array_typecheck(L, 3, 1, f);
} else if (upb_fielddef_ismap(f)) {
msgval = lupb_map_typecheck(L, 3, 1, f);
} else {
const lupb_msgclass *lmsgclass = NULL;
if (type == UPB_TYPE_MESSAGE) {
lmsgclass = lupb_msg_getsubmsgclass(L, 1, f);
}
msgval = lupb_tomsgval(L, type, 3, lmsgclass);
}
/* Set in upb_msg and userval (if necessary). */
upb_msg_set(lmsg->msg, f, msgval, lmsg->lmsgclass->layout);
if (in_userval(f)) {
lupb_uservalseti(L, 1, lupb_fieldindex(f), 3);
}
return 0; /* 1 for chained assignments? */
}
static uint8_t upb_msg_fielddefsize(const upb_fielddef *f) {
if (upb_fielddef_isseq(f)) {
return sizeof(void*);
} else {
return upb_msgval_sizeof2(upb_fielddef_type(f));
}
}
uint32_t upb_handlers_selectorcount(const upb_fielddef *f) {
uint32_t ret = 1;
if (upb_fielddef_isseq(f)) ret += 2; // STARTSEQ/ENDSEQ
if (upb_fielddef_isstring(f)) ret += 2; // [STRING]/STARTSTR/ENDSTR
if (upb_fielddef_issubmsg(f)) {
// ENDSUBMSG (STARTSUBMSG is at table beginning)
ret += 0;
if (upb_fielddef_lazy(f)) {
// STARTSTR/ENDSTR/STRING (for lazy)
ret += 3;
}
}
return ret;
}
static void add_handlers_for_message(const void *closure, upb_handlers *h) {
const upb_msgdef* msgdef = upb_handlers_msgdef(h);
Descriptor* desc = ruby_to_Descriptor(get_def_obj((void*)msgdef));
upb_msg_field_iter i;
// If this is a mapentry message type, set up a special set of handlers and
// bail out of the normal (user-defined) message type handling.
if (upb_msgdef_mapentry(msgdef)) {
add_handlers_for_mapentry(msgdef, h, desc);
return;
}
// Ensure layout exists. We may be invoked to create handlers for a given
// message if we are included as a submsg of another message type before our
// class is actually built, so to work around this, we just create the layout
// (and handlers, in the class-building function) on-demand.
if (desc->layout == NULL) {
desc->layout = create_layout(desc->msgdef);
}
for (upb_msg_field_begin(&i, desc->msgdef);
!upb_msg_field_done(&i);
upb_msg_field_next(&i)) {
const upb_fielddef *f = upb_msg_iter_field(&i);
size_t offset = desc->layout->fields[upb_fielddef_index(f)].offset +
sizeof(MessageHeader);
if (upb_fielddef_containingoneof(f)) {
size_t oneof_case_offset =
desc->layout->fields[upb_fielddef_index(f)].case_offset +
sizeof(MessageHeader);
add_handlers_for_oneof_field(h, f, offset, oneof_case_offset);
} else if (is_map_field(f)) {
add_handlers_for_mapfield(h, f, offset, desc);
} else if (upb_fielddef_isseq(f)) {
add_handlers_for_repeated_field(h, f, offset);
} else {
add_handlers_for_singular_field(h, f, offset);
}
}
}
static void putmsg(VALUE msg_rb, const Descriptor* desc,
upb_sink *sink, int depth) {
MessageHeader* msg;
upb_msg_field_iter i;
upb_status status;
upb_sink_startmsg(sink);
// Protect against cycles (possible because users may freely reassign message
// and repeated fields) by imposing a maximum recursion depth.
if (depth > ENCODE_MAX_NESTING) {
rb_raise(rb_eRuntimeError,
"Maximum recursion depth exceeded during encoding.");
}
TypedData_Get_Struct(msg_rb, MessageHeader, &Message_type, msg);
for (upb_msg_field_begin(&i, desc->msgdef);
!upb_msg_field_done(&i);
upb_msg_field_next(&i)) {
upb_fielddef *f = upb_msg_iter_field(&i);
uint32_t offset =
desc->layout->fields[upb_fielddef_index(f)].offset +
sizeof(MessageHeader);
if (upb_fielddef_containingoneof(f)) {
uint32_t oneof_case_offset =
desc->layout->fields[upb_fielddef_index(f)].case_offset +
sizeof(MessageHeader);
// For a oneof, check that this field is actually present -- skip all the
// below if not.
if (DEREF(msg, oneof_case_offset, uint32_t) !=
upb_fielddef_number(f)) {
continue;
}
// Otherwise, fall through to the appropriate singular-field handler
// below.
}
if (is_map_field(f)) {
VALUE map = DEREF(msg, offset, VALUE);
if (map != Qnil) {
putmap(map, f, sink, depth);
}
} else if (upb_fielddef_isseq(f)) {
VALUE ary = DEREF(msg, offset, VALUE);
if (ary != Qnil) {
putary(ary, f, sink, depth);
}
} else if (upb_fielddef_isstring(f)) {
VALUE str = DEREF(msg, offset, VALUE);
if (RSTRING_LEN(str) > 0) {
putstr(str, f, sink);
}
} else if (upb_fielddef_issubmsg(f)) {
putsubmsg(DEREF(msg, offset, VALUE), f, sink, depth);
} else {
upb_selector_t sel = getsel(f, upb_handlers_getprimitivehandlertype(f));
#define T(upbtypeconst, upbtype, ctype, default_value) \
case upbtypeconst: { \
ctype value = DEREF(msg, offset, ctype); \
if (value != default_value) { \
upb_sink_put##upbtype(sink, sel, value); \
} \
} \
break;
switch (upb_fielddef_type(f)) {
T(UPB_TYPE_FLOAT, float, float, 0.0)
T(UPB_TYPE_DOUBLE, double, double, 0.0)
T(UPB_TYPE_BOOL, bool, uint8_t, 0)
case UPB_TYPE_ENUM:
T(UPB_TYPE_INT32, int32, int32_t, 0)
T(UPB_TYPE_UINT32, uint32, uint32_t, 0)
T(UPB_TYPE_INT64, int64, int64_t, 0)
T(UPB_TYPE_UINT64, uint64, uint64_t, 0)
case UPB_TYPE_STRING:
case UPB_TYPE_BYTES:
case UPB_TYPE_MESSAGE:
rb_raise(rb_eRuntimeError, "Internal error.");
}
#undef T
}
}
upb_sink_endmsg(sink, &status);
}
static bool in_userval(const upb_fielddef *f) {
return lupb_istypewrapped(upb_fielddef_type(f)) || upb_fielddef_isseq(f) ||
upb_fielddef_ismap(f);
}
uint32_t upb_handlers_selectorbaseoffset(const upb_fielddef *f) {
return upb_fielddef_isseq(f) ? 2 : 0;
}
bool upb_handlers_getselector(const upb_fielddef *f, upb_handlertype_t type,
upb_selector_t *s) {
switch (type) {
case UPB_HANDLER_INT32:
case UPB_HANDLER_INT64:
case UPB_HANDLER_UINT32:
case UPB_HANDLER_UINT64:
case UPB_HANDLER_FLOAT:
case UPB_HANDLER_DOUBLE:
case UPB_HANDLER_BOOL:
if (!upb_fielddef_isprimitive(f) ||
upb_handlers_getprimitivehandlertype(f) != type)
return false;
*s = f->selector_base;
break;
case UPB_HANDLER_STRING:
if (upb_fielddef_isstring(f)) {
*s = f->selector_base;
} else if (upb_fielddef_issubmsg(f)) {
*s = f->selector_base + 3;
} else {
return false;
}
break;
case UPB_HANDLER_STARTSTR:
if (upb_fielddef_isstring(f) || upb_fielddef_issubmsg(f)) {
*s = f->selector_base + 1;
} else {
return false;
}
break;
case UPB_HANDLER_ENDSTR:
if (upb_fielddef_isstring(f) || upb_fielddef_issubmsg(f)) {
*s = f->selector_base + 2;
} else {
return false;
}
break;
case UPB_HANDLER_STARTSEQ:
if (!upb_fielddef_isseq(f)) return false;
*s = f->selector_base - 2;
break;
case UPB_HANDLER_ENDSEQ:
if (!upb_fielddef_isseq(f)) return false;
*s = f->selector_base - 1;
break;
case UPB_HANDLER_STARTSUBMSG:
if (!upb_fielddef_issubmsg(f)) return false;
// Selectors for STARTSUBMSG are at the beginning of the table so that the
// selector can also be used as an index into the "sub" array of
// subhandlers. The indexes for the two into these two tables are the
// same, except that in the handler table the static selectors come first.
*s = f->index_ + UPB_STATIC_SELECTOR_COUNT;
break;
case UPB_HANDLER_ENDSUBMSG:
if (!upb_fielddef_issubmsg(f)) return false;
*s = f->selector_base;
break;
}
assert(*s < upb_fielddef_containingtype(f)->selector_count);
return true;
}
bool upb_handlers_freeze(upb_handlers *const*handlers, int n, upb_status *s) {
// TODO: verify we have a transitive closure.
for (int i = 0; i < n; i++) {
upb_handlers *h = handlers[i];
if (!upb_ok(&h->status_)) {
upb_status_seterrf(s, "handlers for message %s had error status: %s",
upb_msgdef_fullname(upb_handlers_msgdef(h)),
upb_status_errmsg(&h->status_));
return false;
}
// Check that there are no closure mismatches due to missing Start* handlers
// or subhandlers with different type-level types.
upb_msg_iter j;
for(upb_msg_begin(&j, h->msg); !upb_msg_done(&j); upb_msg_next(&j)) {
const upb_fielddef *f = upb_msg_iter_field(&j);
if (upb_fielddef_isseq(f)) {
if (!checkstart(h, f, UPB_HANDLER_STARTSEQ, s))
return false;
}
if (upb_fielddef_isstring(f)) {
if (!checkstart(h, f, UPB_HANDLER_STARTSTR, s))
return false;
}
if (upb_fielddef_issubmsg(f)) {
bool hashandler = false;
if (upb_handlers_gethandler(h, getsel(h, f, UPB_HANDLER_STARTSUBMSG)) ||
upb_handlers_gethandler(h, getsel(h, f, UPB_HANDLER_ENDSUBMSG))) {
hashandler = true;
}
if (upb_fielddef_isseq(f) &&
(upb_handlers_gethandler(h, getsel(h, f, UPB_HANDLER_STARTSEQ)) ||
upb_handlers_gethandler(h, getsel(h, f, UPB_HANDLER_ENDSEQ)))) {
hashandler = true;
}
if (hashandler && !upb_handlers_getsubhandlers(h, f)) {
// For now we add an empty subhandlers in this case. It makes the
// decoder code generator simpler, because it only has to handle two
// cases (submessage has handlers or not) as opposed to three
// (submessage has handlers in enclosing message but no subhandlers).
//
// This makes parsing less efficient in the case that we want to
// notice a submessage but skip its contents (like if we're testing
// for submessage presence or counting the number of repeated
// submessages). In this case we will end up parsing the submessage
// field by field and throwing away the results for each, instead of
// skipping the whole delimited thing at once. If this is an issue we
// can revisit it, but do remember that this only arises when you have
// handlers (startseq/startsubmsg/endsubmsg/endseq) set for the
// submessage but no subhandlers. The uses cases for this are
// limited.
upb_handlers *sub = upb_handlers_new(upb_fielddef_msgsubdef(f), &sub);
upb_handlers_setsubhandlers(h, f, sub);
upb_handlers_unref(sub, &sub);
}
// TODO(haberman): check type of submessage.
// This is slightly tricky; also consider whether we should check that
// they match at setsubhandlers time.
}
}
}
if (!upb_refcounted_freeze((upb_refcounted*const*)handlers, n, s,
UPB_MAX_HANDLER_DEPTH)) {
return false;
}
return true;
}
static bool doset(upb_handlers *h, int32_t sel, const upb_fielddef *f,
upb_handlertype_t type, upb_func *func,
upb_handlerattr *attr) {
assert(!upb_handlers_isfrozen(h));
if (sel < 0) {
upb_status_seterrmsg(&h->status_,
"incorrect handler type for this field.");
return false;
}
if (h->table[sel].func) {
upb_status_seterrmsg(&h->status_,
"cannot change handler once it has been set.");
return false;
}
upb_handlerattr set_attr = UPB_HANDLERATTR_INITIALIZER;
if (attr) {
set_attr = *attr;
}
// Check that the given closure type matches the closure type that has been
// established for this context (if any).
const void *closure_type = upb_handlerattr_closuretype(&set_attr);
const void **context_closure_type;
if (type == UPB_HANDLER_STRING) {
context_closure_type = returntype(h, f, UPB_HANDLER_STARTSTR);
} else if (f && upb_fielddef_isseq(f) &&
type != UPB_HANDLER_STARTSEQ &&
type != UPB_HANDLER_ENDSEQ) {
context_closure_type = returntype(h, f, UPB_HANDLER_STARTSEQ);
} else {
context_closure_type = &h->top_closure_type;
}
if (closure_type && *context_closure_type &&
closure_type != *context_closure_type) {
// TODO(haberman): better message for debugging.
upb_status_seterrmsg(&h->status_, "closure type does not match");
return false;
}
if (closure_type)
*context_closure_type = closure_type;
// If this is a STARTSEQ or STARTSTR handler, check that the returned pointer
// matches any pre-existing expectations about what type is expected.
if (type == UPB_HANDLER_STARTSEQ || type == UPB_HANDLER_STARTSTR) {
const void *return_type = upb_handlerattr_returnclosuretype(&set_attr);
const void *table_return_type =
upb_handlerattr_returnclosuretype(&h->table[sel].attr);
if (return_type && table_return_type && return_type != table_return_type) {
upb_status_seterrmsg(&h->status_, "closure return type does not match");
return false;
}
if (table_return_type && !return_type)
upb_handlerattr_setreturnclosuretype(&set_attr, table_return_type);
}
h->table[sel].func = (upb_func*)func;
h->table[sel].attr = set_attr;
return true;
}
static void newhandlers_callback(const void *closure, upb_handlers *h) {
UPB_UNUSED(closure);
upb_handlers_setstartmsg(h, startmsg, NULL);
upb_handlers_setendmsg(h, endmsg, NULL);
const upb_msgdef *m = upb_handlers_msgdef(h);
upb_msg_field_iter i;
for(upb_msg_field_begin(&i, m);
!upb_msg_field_done(&i);
upb_msg_field_next(&i)) {
const upb_fielddef *f = upb_msg_iter_field(&i);
bool packed = upb_fielddef_isseq(f) && upb_fielddef_isprimitive(f) &&
upb_fielddef_packed(f);
upb_handlerattr attr;
upb_wiretype_t wt =
packed ? UPB_WIRE_TYPE_DELIMITED
: upb_pb_native_wire_types[upb_fielddef_descriptortype(f)];
// Pre-encode the tag for this field.
new_tag(h, f, wt, &attr);
if (packed) {
upb_handlers_setstartseq(h, f, encode_startdelimfield, &attr);
upb_handlers_setendseq(h, f, encode_enddelimfield, &attr);
}
#define T(upper, lower, upbtype) \
case UPB_DESCRIPTOR_TYPE_##upper: \
if (packed) { \
upb_handlers_set##upbtype(h, f, encode_packed_##lower, &attr); \
} else { \
upb_handlers_set##upbtype(h, f, encode_scalar_##lower, &attr); \
} \
break;
switch (upb_fielddef_descriptortype(f)) {
T(DOUBLE, double, double);
T(FLOAT, float, float);
T(INT64, int64, int64);
T(INT32, int32, int32);
T(FIXED64, fixed64, uint64);
T(FIXED32, fixed32, uint32);
T(BOOL, bool, bool);
T(UINT32, uint32, uint32);
T(UINT64, uint64, uint64);
T(ENUM, enum, int32);
T(SFIXED32, sfixed32, int32);
T(SFIXED64, sfixed64, int64);
T(SINT32, sint32, int32);
T(SINT64, sint64, int64);
case UPB_DESCRIPTOR_TYPE_STRING:
case UPB_DESCRIPTOR_TYPE_BYTES:
upb_handlers_setstartstr(h, f, encode_startstr, &attr);
upb_handlers_setendstr(h, f, encode_enddelimfield, &attr);
upb_handlers_setstring(h, f, encode_strbuf, &attr);
break;
case UPB_DESCRIPTOR_TYPE_MESSAGE:
upb_handlers_setstartsubmsg(h, f, encode_startdelimfield, &attr);
upb_handlers_setendsubmsg(h, f, encode_enddelimfield, &attr);
break;
case UPB_DESCRIPTOR_TYPE_GROUP: {
// Endgroup takes a different tag (wire_type = END_GROUP).
upb_handlerattr attr2;
new_tag(h, f, UPB_WIRE_TYPE_END_GROUP, &attr2);
upb_handlers_setstartsubmsg(h, f, encode_startgroup, &attr);
upb_handlers_setendsubmsg(h, f, encode_endgroup, &attr2);
upb_handlerattr_uninit(&attr2);
break;
}
}
#undef T
upb_handlerattr_uninit(&attr);
}
}