bool upb_handlers_setsubhandlers(upb_handlers *h, const upb_fielddef *f,
const upb_handlers *sub) {
assert(sub);
assert(!upb_handlers_isfrozen(h));
assert(upb_fielddef_issubmsg(f));
if (SUBH_F(h, f)) return false; // Can't reset.
if (UPB_UPCAST(upb_handlers_msgdef(sub)) != upb_fielddef_subdef(f)) {
return false;
}
SUBH_F(h, f) = sub;
upb_ref2(sub, h);
return true;
}
/* Output a symbolic value from the enum if found, else just print as int32. */
static bool textprinter_putenum(void *closure, const void *handler_data,
int32_t val) {
upb_textprinter *p = closure;
const upb_fielddef *f = handler_data;
const upb_enumdef *enum_def = upb_downcast_enumdef(upb_fielddef_subdef(f));
const char *label = upb_enumdef_iton(enum_def, val);
if (label) {
indent(p);
putf(p, "%s: %s", upb_fielddef_name(f), label);
endfield(p);
} else {
if (!textprinter_putint32(closure, handler_data, val))
return false;
}
return true;
}
// Allocates a new map_handlerdata_t given the map entry message definition. If
// the offset of the field within the parent message is also given, that is
// added to the handler data as well. Note that this is called *twice* per map
// field: once in the parent message handler setup when setting the startsubmsg
// handler and once in the map entry message handler setup when setting the
// key/value and endmsg handlers. The reason is that there is no easy way to
// pass the handlerdata down to the sub-message handler setup.
static map_handlerdata_t* new_map_handlerdata(
size_t ofs,
const upb_msgdef* mapentry_def,
Descriptor* desc) {
const upb_fielddef* key_field;
const upb_fielddef* value_field;
map_handlerdata_t* hd = ALLOC(map_handlerdata_t);
hd->ofs = ofs;
key_field = upb_msgdef_itof(mapentry_def, MAP_KEY_FIELD);
assert(key_field != NULL);
hd->key_field_type = upb_fielddef_type(key_field);
value_field = upb_msgdef_itof(mapentry_def, MAP_VALUE_FIELD);
assert(value_field != NULL);
hd->value_field_type = upb_fielddef_type(value_field);
hd->value_field_subdef = upb_fielddef_subdef(value_field);
return hd;
}
static bool upb_validate_field(upb_fielddef *f, upb_status *s) {
if (upb_fielddef_name(f) == NULL || upb_fielddef_number(f) == 0) {
upb_status_seterrmsg(s, "fielddef must have name and number set");
return false;
}
if (!f->type_is_set_) {
upb_status_seterrmsg(s, "fielddef type was not initialized");
return false;
}
if (upb_fielddef_lazy(f) &&
upb_fielddef_descriptortype(f) != UPB_DESCRIPTOR_TYPE_MESSAGE) {
upb_status_seterrmsg(s,
"only length-delimited submessage fields may be lazy");
return false;
}
if (upb_fielddef_hassubdef(f)) {
if (f->subdef_is_symbolic) {
upb_status_seterrf(s,
"field '%s' has not been resolved", upb_fielddef_name(f));
return false;
}
const upb_def *subdef = upb_fielddef_subdef(f);
if (subdef == NULL) {
upb_status_seterrf(s,
"field %s.%s is missing required subdef",
msgdef_name(f->msg.def), upb_fielddef_name(f));
return false;
} else if (!upb_def_isfrozen(subdef) && !subdef->came_from_user) {
upb_status_seterrf(s,
"subdef of field %s.%s is not frozen or being frozen",
msgdef_name(f->msg.def), upb_fielddef_name(f));
return false;
} else if (upb_fielddef_default_is_symbolic(f)) {
upb_status_seterrf(s,
"enum field %s.%s has not been resolved",
msgdef_name(f->msg.def), upb_fielddef_name(f));
return false;
}
}
return true;
}
static void check_repeated_field_type(VALUE val, const upb_fielddef* field) {
assert(upb_fielddef_label(field) == UPB_LABEL_REPEATED);
if (!RB_TYPE_P(val, T_DATA) || !RTYPEDDATA_P(val) ||
RTYPEDDATA_TYPE(val) != &RepeatedField_type) {
rb_raise(rb_eTypeError, "Expected repeated field array");
}
RepeatedField* self = ruby_to_RepeatedField(val);
if (self->field_type != upb_fielddef_type(field)) {
rb_raise(rb_eTypeError, "Repeated field array has wrong element type");
}
if (self->field_type == UPB_TYPE_MESSAGE ||
self->field_type == UPB_TYPE_ENUM) {
if (self->field_type_class !=
get_def_obj(upb_fielddef_subdef(field))) {
rb_raise(rb_eTypeError,
"Repeated field array has wrong message/enum class");
}
}
}
bool upb_fielddef_resolveenumdefault(upb_fielddef *f, upb_status *s) {
if (!upb_fielddef_default_is_symbolic(f)) return true;
str_t *str = upb_value_getptr(f->defaultval);
const upb_enumdef *e = upb_downcast_enumdef(upb_fielddef_subdef(f));
assert(str); // Points to either a real default or the empty string.
assert(e);
if (str->len == 0) {
// The "default default" for an enum is the first defined value.
upb_value_setint32(&f->defaultval, e->defaultval);
} else {
int32_t val = 0;
if (!upb_enumdef_ntoi(e, str->str, &val)) {
upb_status_seterrf(s, "enum default not found in enum (%s)", str->str);
return false;
}
upb_value_setint32(&f->defaultval, val);
}
f->default_is_string = false;
freestr(str);
return true;
}
static void test_cycles() {
bool ok;
upb_symtab *s = load_test_proto(&s);
const upb_msgdef *m;
const upb_fielddef *f;
const upb_def *def;
const upb_def *def2;
/* Test cycle detection by making a cyclic def's main refcount go to zero
* and then be incremented to one again. */
def = upb_symtab_lookup(s, "A");
upb_def_ref(def, &def);
ASSERT(def);
ASSERT(upb_def_isfrozen(def));
upb_symtab_unref(s, &s);
/* Message A has only one subfield: "optional B b = 1". */
m = upb_downcast_msgdef(def);
f = upb_msgdef_itof(m, 1);
ASSERT(f);
ASSERT(upb_fielddef_hassubdef(f));
ASSERT(upb_msgdef_ntofz(m, "b") == f);
ASSERT(upb_msgdef_ntof(m, "b", 1) == f);
def2 = upb_fielddef_subdef(f);
ASSERT(upb_downcast_msgdef(def2));
ok = strcmp(upb_def_fullname(def2), "B") == 0;
ASSERT(ok);
upb_def_ref(def2, &def2);
upb_def_unref(def, &def);
/* We know "def" is still alive because it's reachable from def2. */
ok = strcmp(upb_def_fullname(def), "A") == 0;
ASSERT(ok);
upb_def_unref(def2, &def2);
}
static bool upb_validate_field(upb_fielddef *f, upb_status *s) {
if (upb_fielddef_name(f) == NULL || upb_fielddef_number(f) == 0) {
upb_status_seterrliteral(s, "fielddef must have name and number set");
return false;
}
if (!f->type_is_set_) {
upb_status_seterrliteral(s, "fielddef type was not initialized");
return false;
}
if (upb_fielddef_hassubdef(f)) {
if (f->subdef_is_symbolic) {
upb_status_seterrf(s,
"field '%s' has not been resolved", upb_fielddef_name(f));
return false;
}
const upb_def *subdef = upb_fielddef_subdef(f);
if (subdef == NULL) {
upb_status_seterrf(s,
"field %s.%s is missing required subdef",
msgdef_name(f->msgdef), upb_fielddef_name(f));
return false;
} else if (!upb_def_isfrozen(subdef) && !subdef->came_from_user) {
upb_status_seterrf(s,
"subdef of field %s.%s is not frozen or being frozen",
msgdef_name(f->msgdef), upb_fielddef_name(f));
return false;
} else if (upb_fielddef_default_is_symbolic(f)) {
upb_status_seterrf(s,
"enum field %s.%s has not been resolved",
msgdef_name(f->msgdef), upb_fielddef_name(f));
return false;
}
}
return true;
}
upb_def *upb_fielddef_subdef_mutable(upb_fielddef *f) {
return (upb_def*)upb_fielddef_subdef(f);
}
const upb_enumdef *upb_fielddef_enumsubdef(const upb_fielddef *f) {
const upb_def *def = upb_fielddef_subdef(f);
return def ? upb_dyncast_enumdef(def) : NULL;
}
const upb_msgdef *upb_fielddef_msgsubdef(const upb_fielddef *f) {
const upb_def *def = upb_fielddef_subdef(f);
return def ? upb_dyncast_msgdef(def) : NULL;
}
/* 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;
}
/* 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;
}
