void layout_deep_copy(MessageLayout* layout, void* to, void* from) {
upb_msg_field_iter it;
for (upb_msg_field_begin(&it, layout->msgdef);
!upb_msg_field_done(&it);
upb_msg_field_next(&it)) {
const upb_fielddef* field = upb_msg_iter_field(&it);
void* to_memory = slot_memory(layout, to, field);
uint32_t* to_oneof_case = slot_oneof_case(layout, to, field);
void* from_memory = slot_memory(layout, from, field);
uint32_t* from_oneof_case = slot_oneof_case(layout, from, field);
if (upb_fielddef_containingoneof(field)) {
if (*from_oneof_case == upb_fielddef_number(field)) {
*to_oneof_case = *from_oneof_case;
native_slot_deep_copy(upb_fielddef_type(field), to_memory, from_memory);
}
} else if (is_map_field(field)) {
DEREF(to_memory, VALUE) =
Map_deep_copy(DEREF(from_memory, VALUE));
} else if (upb_fielddef_label(field) == UPB_LABEL_REPEATED) {
DEREF(to_memory, VALUE) =
RepeatedField_deep_copy(DEREF(from_memory, VALUE));
} else {
if (field_contains_hasbit(layout, field)) {
if (!slot_is_hasbit_set(layout, from, field)) continue;
slot_set_hasbit(layout, to, field);
}
native_slot_deep_copy(upb_fielddef_type(field), to_memory, from_memory);
}
}
}
// narg is a lua table containing a list of defs to add.
static PyObject *PyUpb_SymbolTable_add_defs(PyObject *o, PyObject *defs) {
upb_symtab *s = Check_SymbolTable(o, NULL);
if (!PySequence_Check(defs)) return PyUpb_Error("Must be a sequence");
Py_ssize_t n = PySequence_Length(defs);
// Prevent stack overflow.
if (n > 2048) return PyUpb_Error("Too many defs");
upb_def *cdefs[n];
int i = 0;
for (i = 0; i < n; i++) {
PyObject *pydef = PySequence_GetItem(defs, i);
upb_def *def = Check_MessageDef(pydef, NULL);
cdefs[i++] = def;
upb_msgdef *md = upb_dyncast_msgdef(def);
if (!md) continue;
upb_msg_field_iter j;
for(upb_msg_field_begin(&j, md);
!upb_msg_field_done(&j);
upb_msg_field_next(&j)) {
upb_fielddef *f = upb_msg_iter_field(j);
upb_fielddef_setaccessor(f, PyUpb_AccessorForField(f));
}
upb_msgdef_layout(md);
}
upb_status status = UPB_STATUS_INIT;
upb_symtab_add(s, cdefs, n, &status);
PyUpb_CheckStatus(&status);
return Py_None;
}
static void onmreg(const void *c, upb_handlers *h) {
const upb_msgdef *m = upb_handlers_msgdef(h);
upb_msg_field_iter i;
UPB_UNUSED(c);
upb_handlers_setstartmsg(h, textprinter_startmsg, NULL);
upb_handlers_setendmsg(h, textprinter_endmsg, NULL);
for(upb_msg_field_begin(&i, m);
!upb_msg_field_done(&i);
upb_msg_field_next(&i)) {
upb_fielddef *f = upb_msg_iter_field(&i);
upb_handlerattr attr = UPB_HANDLERATTR_INITIALIZER;
upb_handlerattr_sethandlerdata(&attr, f);
switch (upb_fielddef_type(f)) {
case UPB_TYPE_INT32:
upb_handlers_setint32(h, f, textprinter_putint32, &attr);
break;
case UPB_TYPE_INT64:
upb_handlers_setint64(h, f, textprinter_putint64, &attr);
break;
case UPB_TYPE_UINT32:
upb_handlers_setuint32(h, f, textprinter_putuint32, &attr);
break;
case UPB_TYPE_UINT64:
upb_handlers_setuint64(h, f, textprinter_putuint64, &attr);
break;
case UPB_TYPE_FLOAT:
upb_handlers_setfloat(h, f, textprinter_putfloat, &attr);
break;
case UPB_TYPE_DOUBLE:
upb_handlers_setdouble(h, f, textprinter_putdouble, &attr);
break;
case UPB_TYPE_BOOL:
upb_handlers_setbool(h, f, textprinter_putbool, &attr);
break;
case UPB_TYPE_STRING:
case UPB_TYPE_BYTES:
upb_handlers_setstartstr(h, f, textprinter_startstr, &attr);
upb_handlers_setstring(h, f, textprinter_putstr, &attr);
upb_handlers_setendstr(h, f, textprinter_endstr, &attr);
break;
case UPB_TYPE_MESSAGE: {
const char *name =
upb_fielddef_istagdelim(f)
? shortname(upb_msgdef_fullname(upb_fielddef_msgsubdef(f)))
: upb_fielddef_name(f);
upb_handlerattr_sethandlerdata(&attr, name);
upb_handlers_setstartsubmsg(h, f, textprinter_startsubmsg, &attr);
upb_handlers_setendsubmsg(h, f, textprinter_endsubmsg, &attr);
break;
}
case UPB_TYPE_ENUM:
upb_handlers_setint32(h, f, textprinter_putenum, &attr);
break;
}
}
}
void layout_init(MessageLayout* layout,
void* storage) {
upb_msg_field_iter it;
for (upb_msg_field_begin(&it, layout->msgdef);
!upb_msg_field_done(&it);
upb_msg_field_next(&it)) {
const upb_fielddef* field = upb_msg_iter_field(&it);
void* memory = slot_memory(layout, storage, field);
uint32_t* oneof_case = slot_oneof_case(layout, storage, field);
if (upb_fielddef_containingoneof(field)) {
memset(memory, 0, NATIVE_SLOT_MAX_SIZE);
*oneof_case = ONEOF_CASE_NONE;
} else if (is_map_field(field)) {
VALUE map = Qnil;
const upb_fielddef* key_field = map_field_key(field);
const upb_fielddef* value_field = map_field_value(field);
VALUE type_class = field_type_class(value_field);
if (type_class != Qnil) {
VALUE args[3] = {
fieldtype_to_ruby(upb_fielddef_type(key_field)),
fieldtype_to_ruby(upb_fielddef_type(value_field)),
type_class,
};
map = rb_class_new_instance(3, args, cMap);
} else {
VALUE args[2] = {
fieldtype_to_ruby(upb_fielddef_type(key_field)),
fieldtype_to_ruby(upb_fielddef_type(value_field)),
};
map = rb_class_new_instance(2, args, cMap);
}
DEREF(memory, VALUE) = map;
} else if (upb_fielddef_label(field) == UPB_LABEL_REPEATED) {
VALUE ary = Qnil;
VALUE type_class = field_type_class(field);
if (type_class != Qnil) {
VALUE args[2] = {
fieldtype_to_ruby(upb_fielddef_type(field)),
type_class,
};
ary = rb_class_new_instance(2, args, cRepeatedField);
} else {
VALUE args[1] = { fieldtype_to_ruby(upb_fielddef_type(field)) };
ary = rb_class_new_instance(1, args, cRepeatedField);
}
DEREF(memory, VALUE) = ary;
} else {
native_slot_init(upb_fielddef_type(field), memory);
}
}
}
void layout_init(MessageLayout* layout,
void* storage) {
upb_msg_field_iter it;
for (upb_msg_field_begin(&it, layout->msgdef);
!upb_msg_field_done(&it);
upb_msg_field_next(&it)) {
layout_clear(layout, storage, upb_msg_iter_field(&it));
}
}
static PyObject *PyUpb_MessageDef_fields(PyObject *obj, PyObject *args) {
upb_msgdef *m = Check_MessageDef(obj, NULL);
PyObject *ret = PyList_New(0);
upb_msg_field_iter i;
for(upb_msg_field_begin(&i, m);
!upb_msg_field_done(&i);
upb_msg_field_next(&ii)) {
upb_fielddef *f = upb_msg_iter_field(&i);
PyList_Append(ret, PyUpb_FieldDef_GetOrCreate(f));
}
return ret;
}
VALUE layout_hash(MessageLayout* layout, void* storage) {
upb_msg_field_iter it;
st_index_t h = rb_hash_start(0);
VALUE hash_sym = rb_intern("hash");
for (upb_msg_field_begin(&it, layout->msgdef);
!upb_msg_field_done(&it);
upb_msg_field_next(&it)) {
const upb_fielddef* field = upb_msg_iter_field(&it);
VALUE field_val = layout_get(layout, storage, field);
h = rb_hash_uint(h, NUM2LONG(rb_funcall(field_val, hash_sym, 0)));
}
h = rb_hash_end(h);
return INT2FIX(h);
}
/*
* call-seq:
* Message.to_h => {}
*
* Returns the message as a Ruby Hash object, with keys as symbols.
*/
VALUE Message_to_h(VALUE _self) {
MessageHeader* self;
VALUE hash;
upb_msg_field_iter it;
TypedData_Get_Struct(_self, MessageHeader, &Message_type, self);
hash = rb_hash_new();
for (upb_msg_field_begin(&it, self->descriptor->msgdef);
!upb_msg_field_done(&it);
upb_msg_field_next(&it)) {
const upb_fielddef* field = upb_msg_iter_field(&it);
// For proto2, do not include fields which are not set.
if (upb_msgdef_syntax(self->descriptor->msgdef) == UPB_SYNTAX_PROTO2 &&
field_contains_hasbit(self->descriptor->layout, field) &&
!layout_has(self->descriptor->layout, Message_data(self), field)) {
continue;
}
VALUE msg_value = layout_get(self->descriptor->layout, Message_data(self),
field);
VALUE msg_key = ID2SYM(rb_intern(upb_fielddef_name(field)));
if (is_map_field(field)) {
msg_value = Map_to_h(msg_value);
} else if (upb_fielddef_label(field) == UPB_LABEL_REPEATED) {
msg_value = RepeatedField_to_ary(msg_value);
if (upb_msgdef_syntax(self->descriptor->msgdef) == UPB_SYNTAX_PROTO2 &&
RARRAY_LEN(msg_value) == 0) {
continue;
}
if (upb_fielddef_type(field) == UPB_TYPE_MESSAGE) {
for (int i = 0; i < RARRAY_LEN(msg_value); i++) {
VALUE elem = rb_ary_entry(msg_value, i);
rb_ary_store(msg_value, i, Message_to_h(elem));
}
}
} else if (msg_value != Qnil &&
upb_fielddef_type(field) == UPB_TYPE_MESSAGE) {
msg_value = Message_to_h(msg_value);
}
rb_hash_aset(hash, msg_key, msg_value);
}
return hash;
}
VALUE Message_to_h(VALUE _self) {
MessageHeader* self;
TypedData_Get_Struct(_self, MessageHeader, &Message_type, self);
VALUE hash = rb_hash_new();
upb_msg_field_iter it;
for (upb_msg_field_begin(&it, self->descriptor->msgdef);
!upb_msg_field_done(&it);
upb_msg_field_next(&it)) {
const upb_fielddef* field = upb_msg_iter_field(&it);
VALUE msg_value = layout_get(self->descriptor->layout, Message_data(self), field);
VALUE msg_key = ID2SYM(rb_intern(upb_fielddef_name(field)));
if (upb_fielddef_label(field) == UPB_LABEL_REPEATED) {
msg_value = RepeatedField_to_ary(msg_value);
}
rb_hash_aset(hash, msg_key, msg_value);
}
return hash;
}
void layout_mark(MessageLayout* layout, void* storage) {
upb_msg_field_iter it;
for (upb_msg_field_begin(&it, layout->msgdef);
!upb_msg_field_done(&it);
upb_msg_field_next(&it)) {
const upb_fielddef* field = upb_msg_iter_field(&it);
void* memory = slot_memory(layout, storage, field);
uint32_t* oneof_case = slot_oneof_case(layout, storage, field);
if (upb_fielddef_containingoneof(field)) {
if (*oneof_case == upb_fielddef_number(field)) {
native_slot_mark(upb_fielddef_type(field), memory);
}
} else if (upb_fielddef_label(field) == UPB_LABEL_REPEATED) {
rb_gc_mark(DEREF(memory, VALUE));
} else {
native_slot_mark(upb_fielddef_type(field), memory);
}
}
}
VALUE layout_eq(MessageLayout* layout, void* msg1, void* msg2) {
upb_msg_field_iter it;
for (upb_msg_field_begin(&it, layout->msgdef);
!upb_msg_field_done(&it);
upb_msg_field_next(&it)) {
const upb_fielddef* field = upb_msg_iter_field(&it);
void* msg1_memory = slot_memory(layout, msg1, field);
uint32_t* msg1_oneof_case = slot_oneof_case(layout, msg1, field);
void* msg2_memory = slot_memory(layout, msg2, field);
uint32_t* msg2_oneof_case = slot_oneof_case(layout, msg2, field);
if (upb_fielddef_containingoneof(field)) {
if (*msg1_oneof_case != *msg2_oneof_case ||
(*msg1_oneof_case == upb_fielddef_number(field) &&
!native_slot_eq(upb_fielddef_type(field),
msg1_memory,
msg2_memory))) {
return Qfalse;
}
} else if (is_map_field(field)) {
if (!Map_eq(DEREF(msg1_memory, VALUE),
DEREF(msg2_memory, VALUE))) {
return Qfalse;
}
} else if (upb_fielddef_label(field) == UPB_LABEL_REPEATED) {
if (!RepeatedField_eq(DEREF(msg1_memory, VALUE),
DEREF(msg2_memory, VALUE))) {
return Qfalse;
}
} else {
if (slot_is_hasbit_set(layout, msg1, field) !=
slot_is_hasbit_set(layout, msg2, field) ||
!native_slot_eq(upb_fielddef_type(field),
msg1_memory, msg2_memory)) {
return Qfalse;
}
}
}
return Qtrue;
}
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);
}
}
}
VALUE layout_inspect(MessageLayout* layout, void* storage) {
VALUE str = rb_str_new2("");
upb_msg_field_iter it;
bool first = true;
for (upb_msg_field_begin(&it, layout->msgdef);
!upb_msg_field_done(&it);
upb_msg_field_next(&it)) {
const upb_fielddef* field = upb_msg_iter_field(&it);
VALUE field_val = layout_get(layout, storage, field);
if (!first) {
str = rb_str_cat2(str, ", ");
} else {
first = false;
}
str = rb_str_cat2(str, upb_fielddef_name(field));
str = rb_str_cat2(str, ": ");
str = rb_str_append(str, rb_funcall(field_val, rb_intern("inspect"), 0));
}
return str;
}
/* Starts a depth-first traversal at "def", recursing into any subdefs
* (ie. submessage types). Adds duplicates of existing defs to addtab
* wherever necessary, so that the resulting symtab will be consistent once
* addtab is added.
*
* More specifically, if any def D is found in the DFS that:
*
* 1. can reach a def that is being replaced by something in addtab, AND
*
* 2. is not itself being replaced already (ie. this name doesn't already
* exist in addtab)
*
* ...then a duplicate (new copy) of D will be added to addtab.
*
* Returns true if this happened for any def reachable from "def."
*
* It is slightly tricky to do this correctly in the presence of cycles. If we
* detect that our DFS has hit a cycle, we might not yet know if any SCCs on
* our stack can reach a def in addtab or not. Once we figure this out, that
* answer needs to apply to *all* defs in these SCCs, even if we visited them
* already. So a straight up one-pass cycle-detecting DFS won't work.
*
* To work around this problem, we traverse each SCC (which we already
* computed, since these defs are frozen) as a single node. We first compute
* whether the SCC as a whole can reach any def in addtab, then we dup (or not)
* the entire SCC. This requires breaking the encapsulation of upb_refcounted,
* since that is where we get the data about what SCC we are in. */
static bool upb_resolve_dfs(const upb_def *def, upb_strtable *addtab,
const void *new_owner, upb_inttable *seen,
upb_status *s) {
upb_value v;
bool need_dup;
const upb_def *base;
const void* memoize_key;
/* Memoize results of this function for efficiency (since we're traversing a
* DAG this is not needed to limit the depth of the search).
*
* We memoize by SCC instead of by individual def. */
memoize_key = def->base.group;
if (upb_inttable_lookupptr(seen, memoize_key, &v))
return upb_value_getbool(v);
/* Visit submessages for all messages in the SCC. */
need_dup = false;
base = def;
do {
upb_value v;
const upb_msgdef *m;
assert(upb_def_isfrozen(def));
if (def->type == UPB_DEF_FIELD) continue;
if (upb_strtable_lookup(addtab, upb_def_fullname(def), &v)) {
need_dup = true;
}
/* For messages, continue the recursion by visiting all subdefs, but only
* ones in different SCCs. */
m = upb_dyncast_msgdef(def);
if (m) {
upb_msg_field_iter i;
for(upb_msg_field_begin(&i, m);
!upb_msg_field_done(&i);
upb_msg_field_next(&i)) {
upb_fielddef *f = upb_msg_iter_field(&i);
const upb_def *subdef;
if (!upb_fielddef_hassubdef(f)) continue;
subdef = upb_fielddef_subdef(f);
/* Skip subdefs in this SCC. */
if (def->base.group == subdef->base.group) continue;
/* |= to avoid short-circuit; we need its side-effects. */
need_dup |= upb_resolve_dfs(subdef, addtab, new_owner, seen, s);
if (!upb_ok(s)) return false;
}
}
} while ((def = (upb_def*)def->base.next) != base);
if (need_dup) {
/* Dup all defs in this SCC that don't already have entries in addtab. */
def = base;
do {
const char *name;
if (def->type == UPB_DEF_FIELD) continue;
name = upb_def_fullname(def);
if (!upb_strtable_lookup(addtab, name, NULL)) {
upb_def *newdef = upb_def_dup(def, new_owner);
if (!newdef) goto oom;
newdef->came_from_user = false;
if (!upb_strtable_insert(addtab, name, upb_value_ptr(newdef)))
goto oom;
}
} while ((def = (upb_def*)def->base.next) != base);
}
upb_inttable_insertptr(seen, memoize_key, upb_value_bool(need_dup));
return need_dup;
oom:
upb_status_seterrmsg(s, "out of memory");
return false;
}
static bool upb_msglayout_init(const upb_msgdef *m,
upb_msglayout *l,
upb_msgfactory *factory) {
upb_msg_field_iter it;
upb_msg_oneof_iter oit;
size_t hasbit;
size_t submsg_count = 0;
const upb_msglayout **submsgs;
upb_msglayout_field *fields;
for (upb_msg_field_begin(&it, m);
!upb_msg_field_done(&it);
upb_msg_field_next(&it)) {
const upb_fielddef* f = upb_msg_iter_field(&it);
if (upb_fielddef_issubmsg(f)) {
submsg_count++;
}
}
memset(l, 0, sizeof(*l));
fields = upb_gmalloc(upb_msgdef_numfields(m) * sizeof(*fields));
submsgs = upb_gmalloc(submsg_count * sizeof(*submsgs));
if ((!fields && upb_msgdef_numfields(m)) ||
(!submsgs && submsg_count)) {
/* OOM. */
upb_gfree(fields);
upb_gfree(submsgs);
return false;
}
l->field_count = upb_msgdef_numfields(m);
l->fields = fields;
l->submsgs = submsgs;
/* Allocate data offsets in three stages:
*
* 1. hasbits.
* 2. regular fields.
* 3. oneof fields.
*
* OPT: There is a lot of room for optimization here to minimize the size.
*/
/* Allocate hasbits and set basic field attributes. */
submsg_count = 0;
for (upb_msg_field_begin(&it, m), hasbit = 0;
!upb_msg_field_done(&it);
upb_msg_field_next(&it)) {
const upb_fielddef* f = upb_msg_iter_field(&it);
upb_msglayout_field *field = &fields[upb_fielddef_index(f)];
field->number = upb_fielddef_number(f);
field->descriptortype = upb_fielddef_descriptortype(f);
field->label = upb_fielddef_label(f);
if (upb_fielddef_issubmsg(f)) {
const upb_msglayout *sub_layout =
upb_msgfactory_getlayout(factory, upb_fielddef_msgsubdef(f));
field->submsg_index = submsg_count++;
submsgs[field->submsg_index] = sub_layout;
}
if (upb_fielddef_haspresence(f) && !upb_fielddef_containingoneof(f)) {
field->presence = (hasbit++);
} else {
field->presence = 0;
}
}
/* Account for space used by hasbits. */
l->size = div_round_up(hasbit, 8);
/* Allocate non-oneof fields. */
for (upb_msg_field_begin(&it, m); !upb_msg_field_done(&it);
upb_msg_field_next(&it)) {
const upb_fielddef* f = upb_msg_iter_field(&it);
size_t field_size = upb_msg_fielddefsize(f);
size_t index = upb_fielddef_index(f);
if (upb_fielddef_containingoneof(f)) {
/* Oneofs are handled separately below. */
continue;
}
fields[index].offset = upb_msglayout_place(l, field_size);
}
/* Allocate oneof fields. Each oneof field consists of a uint32 for the case
* and space for the actual data. */
for (upb_msg_oneof_begin(&oit, m); !upb_msg_oneof_done(&oit);
upb_msg_oneof_next(&oit)) {
const upb_oneofdef* o = upb_msg_iter_oneof(&oit);
upb_oneof_iter fit;
size_t case_size = sizeof(uint32_t); /* Could potentially optimize this. */
size_t field_size = 0;
uint32_t case_offset;
uint32_t data_offset;
/* Calculate field size: the max of all field sizes. */
for (upb_oneof_begin(&fit, o);
!upb_oneof_done(&fit);
upb_oneof_next(&fit)) {
const upb_fielddef* f = upb_oneof_iter_field(&fit);
field_size = UPB_MAX(field_size, upb_msg_fielddefsize(f));
}
/* Align and allocate case offset. */
case_offset = upb_msglayout_place(l, case_size);
data_offset = upb_msglayout_place(l, field_size);
for (upb_oneof_begin(&fit, o);
!upb_oneof_done(&fit);
upb_oneof_next(&fit)) {
const upb_fielddef* f = upb_oneof_iter_field(&fit);
fields[upb_fielddef_index(f)].offset = data_offset;
fields[upb_fielddef_index(f)].presence = ~case_offset;
}
}
/* Size of the entire structure should be a multiple of its greatest
* alignment. TODO: track overall alignment for real? */
l->size = align_up(l->size, 8);
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);
}
}
MessageLayout* create_layout(const upb_msgdef* msgdef) {
MessageLayout* layout = ALLOC(MessageLayout);
int nfields = upb_msgdef_numfields(msgdef);
upb_msg_field_iter it;
upb_msg_oneof_iter oit;
size_t off = 0;
layout->fields = ALLOC_N(MessageField, nfields);
size_t hasbit = 0;
for (upb_msg_field_begin(&it, msgdef);
!upb_msg_field_done(&it);
upb_msg_field_next(&it)) {
const upb_fielddef* field = upb_msg_iter_field(&it);
if (upb_fielddef_haspresence(field)) {
layout->fields[upb_fielddef_index(field)].hasbit = hasbit++;
} else {
layout->fields[upb_fielddef_index(field)].hasbit =
MESSAGE_FIELD_NO_HASBIT;
}
}
if (hasbit != 0) {
off += (hasbit + 8 - 1) / 8;
}
for (upb_msg_field_begin(&it, msgdef);
!upb_msg_field_done(&it);
upb_msg_field_next(&it)) {
const upb_fielddef* field = upb_msg_iter_field(&it);
size_t field_size;
if (upb_fielddef_containingoneof(field)) {
// Oneofs are handled separately below.
continue;
}
// Allocate |field_size| bytes for this field in the layout.
field_size = 0;
if (upb_fielddef_label(field) == UPB_LABEL_REPEATED) {
field_size = sizeof(VALUE);
} else {
field_size = native_slot_size(upb_fielddef_type(field));
}
// Align current offset up to |size| granularity.
off = align_up_to(off, field_size);
layout->fields[upb_fielddef_index(field)].offset = off;
layout->fields[upb_fielddef_index(field)].case_offset =
MESSAGE_FIELD_NO_CASE;
off += field_size;
}
// Handle oneofs now -- we iterate over oneofs specifically and allocate only
// one slot per oneof.
//
// We assign all value slots first, then pack the 'case' fields at the end,
// since in the common case (modern 64-bit platform) these are 8 bytes and 4
// bytes respectively and we want to avoid alignment overhead.
//
// Note that we reserve 4 bytes (a uint32) per 'case' slot because the value
// space for oneof cases is conceptually as wide as field tag numbers. In
// practice, it's unlikely that a oneof would have more than e.g. 256 or 64K
// members (8 or 16 bits respectively), so conceivably we could assign
// consecutive case numbers and then pick a smaller oneof case slot size, but
// the complexity to implement this indirection is probably not worthwhile.
for (upb_msg_oneof_begin(&oit, msgdef);
!upb_msg_oneof_done(&oit);
upb_msg_oneof_next(&oit)) {
const upb_oneofdef* oneof = upb_msg_iter_oneof(&oit);
upb_oneof_iter fit;
// Always allocate NATIVE_SLOT_MAX_SIZE bytes, but share the slot between
// all fields.
size_t field_size = NATIVE_SLOT_MAX_SIZE;
// Align the offset.
off = align_up_to(off, field_size);
// Assign all fields in the oneof this same offset.
for (upb_oneof_begin(&fit, oneof);
!upb_oneof_done(&fit);
upb_oneof_next(&fit)) {
const upb_fielddef* field = upb_oneof_iter_field(&fit);
layout->fields[upb_fielddef_index(field)].offset = off;
}
off += field_size;
}
// Now the case fields.
for (upb_msg_oneof_begin(&oit, msgdef);
!upb_msg_oneof_done(&oit);
upb_msg_oneof_next(&oit)) {
const upb_oneofdef* oneof = upb_msg_iter_oneof(&oit);
upb_oneof_iter fit;
size_t field_size = sizeof(uint32_t);
// Align the offset.
off = (off + field_size - 1) & ~(field_size - 1);
// Assign all fields in the oneof this same offset.
for (upb_oneof_begin(&fit, oneof);
!upb_oneof_done(&fit);
upb_oneof_next(&fit)) {
const upb_fielddef* field = upb_oneof_iter_field(&fit);
layout->fields[upb_fielddef_index(field)].case_offset = off;
}
off += field_size;
}
layout->size = off;
layout->msgdef = msgdef;
upb_msgdef_ref(layout->msgdef, &layout->msgdef);
return layout;
}
/* TODO(haberman): we need a lot more testing of error conditions.
* The came_from_user stuff in particular is not tested. */
bool upb_symtab_add(upb_symtab *s, upb_def *const*defs, int n, void *ref_donor,
upb_status *status) {
int i;
upb_strtable_iter iter;
upb_def **add_defs = NULL;
upb_strtable addtab;
upb_inttable seen;
assert(!upb_symtab_isfrozen(s));
if (!upb_strtable_init(&addtab, UPB_CTYPE_PTR)) {
upb_status_seterrmsg(status, "out of memory");
return false;
}
/* Add new defs to our "add" set. */
for (i = 0; i < n; i++) {
upb_def *def = defs[i];
const char *fullname;
upb_fielddef *f;
if (upb_def_isfrozen(def)) {
upb_status_seterrmsg(status, "added defs must be mutable");
goto err;
}
assert(!upb_def_isfrozen(def));
fullname = upb_def_fullname(def);
if (!fullname) {
upb_status_seterrmsg(
status, "Anonymous defs cannot be added to a symtab");
goto err;
}
f = upb_dyncast_fielddef_mutable(def);
if (f) {
if (!upb_fielddef_containingtypename(f)) {
upb_status_seterrmsg(status,
"Standalone fielddefs must have a containing type "
"(extendee) name set");
goto err;
}
} else {
if (upb_strtable_lookup(&addtab, fullname, NULL)) {
upb_status_seterrf(status, "Conflicting defs named '%s'", fullname);
goto err;
}
/* We need this to back out properly, because if there is a failure we
* need to donate the ref back to the caller. */
def->came_from_user = true;
upb_def_donateref(def, ref_donor, s);
if (!upb_strtable_insert(&addtab, fullname, upb_value_ptr(def)))
goto oom_err;
}
}
/* Add standalone fielddefs (ie. extensions) to the appropriate messages.
* If the appropriate message only exists in the existing symtab, duplicate
* it so we have a mutable copy we can add the fields to. */
for (i = 0; i < n; i++) {
upb_def *def = defs[i];
upb_fielddef *f = upb_dyncast_fielddef_mutable(def);
const char *msgname;
upb_value v;
upb_msgdef *m;
if (!f) continue;
msgname = upb_fielddef_containingtypename(f);
/* We validated this earlier in this function. */
assert(msgname);
/* If the extendee name is absolutely qualified, move past the initial ".".
* TODO(haberman): it is not obvious what it would mean if this was not
* absolutely qualified. */
if (msgname[0] == '.') {
msgname++;
}
if (upb_strtable_lookup(&addtab, msgname, &v)) {
/* Extendee is in the set of defs the user asked us to add. */
m = upb_value_getptr(v);
} else {
/* Need to find and dup the extendee from the existing symtab. */
const upb_msgdef *frozen_m = upb_symtab_lookupmsg(s, msgname);
if (!frozen_m) {
upb_status_seterrf(status,
"Tried to extend message %s that does not exist "
"in this SymbolTable.",
msgname);
goto err;
}
m = upb_msgdef_dup(frozen_m, s);
if (!m) goto oom_err;
if (!upb_strtable_insert(&addtab, msgname, upb_value_ptr(m))) {
upb_msgdef_unref(m, s);
goto oom_err;
}
}
if (!upb_msgdef_addfield(m, f, ref_donor, status)) {
goto err;
}
}
/* Add dups of any existing def that can reach a def with the same name as
* anything in our "add" set. */
if (!upb_inttable_init(&seen, UPB_CTYPE_BOOL)) goto oom_err;
upb_strtable_begin(&iter, &s->symtab);
for (; !upb_strtable_done(&iter); upb_strtable_next(&iter)) {
upb_def *def = upb_value_getptr(upb_strtable_iter_value(&iter));
upb_resolve_dfs(def, &addtab, s, &seen, status);
if (!upb_ok(status)) goto err;
}
upb_inttable_uninit(&seen);
/* Now using the table, resolve symbolic references for subdefs. */
upb_strtable_begin(&iter, &addtab);
for (; !upb_strtable_done(&iter); upb_strtable_next(&iter)) {
const char *base;
upb_def *def = upb_value_getptr(upb_strtable_iter_value(&iter));
upb_msgdef *m = upb_dyncast_msgdef_mutable(def);
upb_msg_field_iter j;
if (!m) continue;
/* Type names are resolved relative to the message in which they appear. */
base = upb_msgdef_fullname(m);
for(upb_msg_field_begin(&j, m);
!upb_msg_field_done(&j);
upb_msg_field_next(&j)) {
upb_fielddef *f = upb_msg_iter_field(&j);
const char *name = upb_fielddef_subdefname(f);
if (name && !upb_fielddef_subdef(f)) {
/* Try the lookup in the current set of to-be-added defs first. If not
* there, try existing defs. */
upb_def *subdef = upb_resolvename(&addtab, base, name);
if (subdef == NULL) {
subdef = upb_resolvename(&s->symtab, base, name);
}
if (subdef == NULL) {
upb_status_seterrf(
status, "couldn't resolve name '%s' in message '%s'", name, base);
goto err;
} else if (!upb_fielddef_setsubdef(f, subdef, status)) {
goto err;
}
}
}
}
/* We need an array of the defs in addtab, for passing to upb_def_freeze. */
add_defs = malloc(sizeof(void*) * upb_strtable_count(&addtab));
if (add_defs == NULL) goto oom_err;
upb_strtable_begin(&iter, &addtab);
for (n = 0; !upb_strtable_done(&iter); upb_strtable_next(&iter)) {
add_defs[n++] = upb_value_getptr(upb_strtable_iter_value(&iter));
}
if (!upb_def_freeze(add_defs, n, status)) goto err;
/* This must be delayed until all errors have been detected, since error
* recovery code uses this table to cleanup defs. */
upb_strtable_uninit(&addtab);
/* TODO(haberman) we don't properly handle errors after this point (like
* OOM in upb_strtable_insert() below). */
for (i = 0; i < n; i++) {
upb_def *def = add_defs[i];
const char *name = upb_def_fullname(def);
upb_value v;
bool success;
if (upb_strtable_remove(&s->symtab, name, &v)) {
const upb_def *def = upb_value_getptr(v);
upb_def_unref(def, s);
}
success = upb_strtable_insert(&s->symtab, name, upb_value_ptr(def));
UPB_ASSERT_VAR(success, success == true);
}
free(add_defs);
return true;
oom_err:
upb_status_seterrmsg(status, "out of memory");
err: {
/* For defs the user passed in, we need to donate the refs back. For defs
* we dup'd, we need to just unref them. */
upb_strtable_begin(&iter, &addtab);
for (; !upb_strtable_done(&iter); upb_strtable_next(&iter)) {
upb_def *def = upb_value_getptr(upb_strtable_iter_value(&iter));
bool came_from_user = def->came_from_user;
def->came_from_user = false;
if (came_from_user) {
upb_def_donateref(def, s, ref_donor);
} else {
upb_def_unref(def, s);
}
}
}
upb_strtable_uninit(&addtab);
free(add_defs);
assert(!upb_ok(status));
return false;
}
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);
}