static void test_mapentry_check() {
upb_status s = UPB_STATUS_INIT;
upb_msgdef *m = upb_msgdef_new(&m);
upb_fielddef *f = upb_fielddef_new(&f);
upb_symtab *symtab = upb_symtab_new(&symtab);
upb_msgdef *subm = upb_msgdef_new(&subm);
upb_def *defs[2];
upb_msgdef_setfullname(m, "TestMessage", &s);
upb_fielddef_setname(f, "field1", &s);
upb_fielddef_setnumber(f, 1, &s);
upb_fielddef_setlabel(f, UPB_LABEL_OPTIONAL);
upb_fielddef_settype(f, UPB_TYPE_MESSAGE);
upb_fielddef_setsubdefname(f, ".MapEntry", &s);
upb_msgdef_addfield(m, f, &f, &s);
ASSERT(upb_ok(&s));
upb_msgdef_setfullname(subm, "MapEntry", &s);
upb_msgdef_setmapentry(subm, true);
defs[0] = upb_msgdef_upcast_mutable(m);
defs[1] = upb_msgdef_upcast_mutable(subm);
upb_symtab_add(symtab, defs, 2, NULL, &s);
/* Should not have succeeded: non-repeated field pointing to a MapEntry. */
ASSERT(!upb_ok(&s));
upb_fielddef_setlabel(f, UPB_LABEL_REPEATED);
upb_symtab_add(symtab, defs, 2, NULL, &s);
ASSERT(upb_ok(&s));
upb_symtab_unref(symtab, &symtab);
upb_msgdef_unref(subm, &subm);
upb_msgdef_unref(m, &m);
}
upb_success_t upb_decoder_decode(upb_decoder *d) {
assert(d->input);
if (sigsetjmp(d->exitjmp, 0)) {
assert(!upb_ok(&d->status));
return UPB_ERROR;
}
upb_dispatch_startmsg(&d->dispatcher);
// Prime the buf so we can hit the JIT immediately.
upb_trypullbuf(d);
upb_fhandlers *f = d->dispatcher.top->f;
while(1) {
upb_decoder_checkdelim(d);
#ifdef UPB_USE_JIT_X64
upb_decoder_enterjit(d);
upb_decoder_checkpoint(d);
#endif
if (!d->top_is_packed) f = upb_decode_tag(d);
if (!f) {
// Sucessful EOF. We may need to dispatch a top-level implicit frame.
if (d->dispatcher.top == d->dispatcher.stack + 1) {
assert(d->dispatcher.top->is_sequence);
upb_dispatch_endseq(&d->dispatcher);
}
return UPB_OK;
}
f->decode(d, f);
upb_decoder_checkpoint(d);
}
}
static bool initialize()
{
// Initialize upb state, decode descriptor.
upb_status status = UPB_STATUS_INIT;
upb_symtab *s = upb_symtab_new();
upb_string *fds_str = upb_strreadfile(MESSAGE_DESCRIPTOR_FILE);
if(fds_str == NULL) {
fprintf(stderr, "Couldn't read " MESSAGE_DESCRIPTOR_FILE ":"),
upb_printerr(&status);
return false;
}
upb_parsedesc(s, fds_str, &status);
upb_string_unref(fds_str);
if(!upb_ok(&status)) {
fprintf(stderr, "Error importing " MESSAGE_DESCRIPTOR_FILE ":");
upb_printerr(&status);
return false;
}
def = upb_dyncast_msgdef(upb_symtab_lookup(s, UPB_STRLIT(MESSAGE_NAME)));
if(!def) {
fprintf(stderr, "Error finding symbol '" UPB_STRFMT "'.\n",
UPB_STRARG(UPB_STRLIT(MESSAGE_NAME)));
return false;
}
upb_symtab_unref(s);
// Read the message data itself.
input_str = upb_strreadfile(MESSAGE_FILE);
if(input_str == NULL) {
fprintf(stderr, "Error reading " MESSAGE_FILE "\n");
return false;
}
upb_status_uninit(&status);
msg = upb_msg_new(def);
upb_stringsrc_init(&strsrc);
upb_handlers_init(&h, def);
upb_msg_regdhandlers(&h);
upb_decoder_init(&d, &h);
if (!BYREF) {
// Pretend the input string is stack-allocated, which will force its data
// to be copied instead of referenced. There is no good reason to do this,
// except to benchmark against proto2 more fairly, which in its open-source
// release does not support referencing the input string.
input_str->refcount.v = _UPB_STRING_REFCOUNT_STACK;
}
return true;
}
static void test_descriptor_flags() {
upb_msgdef *m = upb_msgdef_new(&m);
upb_msgdef *m2;
upb_status s = UPB_STATUS_INIT;
ASSERT(upb_msgdef_mapentry(m) == false);
upb_msgdef_setfullname(m, "TestMessage", &s);
ASSERT(upb_ok(&s));
upb_msgdef_setmapentry(m, true);
ASSERT(upb_msgdef_mapentry(m) == true);
m2 = upb_msgdef_dup(m, &m2);
ASSERT(upb_msgdef_mapentry(m2) == true);
upb_msgdef_unref(m, &m);
upb_msgdef_unref(m2, &m2);
}
static size_t run(int i)
{
upb_status status = UPB_STATUS_INIT;
i %= NUM_MESSAGES;
upb_msg_clear(msg[i], def);
upb_decoder_reset(&d, upb_stringsrc_bytesrc(&strsrc),
0, UPB_NONDELIMITED, msg[i]);
upb_decoder_decode(&d, &status);
if(!upb_ok(&status)) goto err;
return len;
err:
fprintf(stderr, "Decode error: %s", upb_status_getstr(&status));
return 0;
}
bool upb_def_freeze(upb_def *const* defs, int n, upb_status *s) {
// First perform validation, in two passes so we can check that we have a
// transitive closure without needing to search.
for (int i = 0; i < n; i++) {
upb_def *def = defs[i];
if (upb_def_isfrozen(def)) {
// Could relax this requirement if it's annoying.
upb_status_seterrliteral(s, "def is already frozen");
goto err;
} else if (def->type == UPB_DEF_FIELD) {
upb_status_seterrliteral(s, "standalone fielddefs can not be frozen");
goto err;
} else {
// Set now to detect transitive closure in the second pass.
def->came_from_user = true;
}
}
for (int i = 0; i < n; i++) {
upb_msgdef *m = upb_dyncast_msgdef_mutable(defs[i]);
upb_enumdef *e = upb_dyncast_enumdef_mutable(defs[i]);
if (m) {
upb_inttable_compact(&m->itof);
upb_msg_iter j;
uint32_t selector = UPB_STATIC_SELECTOR_COUNT;
for(upb_msg_begin(&j, m); !upb_msg_done(&j); upb_msg_next(&j)) {
upb_fielddef *f = upb_msg_iter_field(&j);
assert(f->msgdef == m);
if (!upb_validate_field(f, s)) goto err;
f->selector_base = selector + upb_handlers_selectorbaseoffset(f);
selector += upb_handlers_selectorcount(f);
}
m->selector_count = selector;
} else if (e) {
upb_inttable_compact(&e->iton);
}
}
// Validation all passed; freeze the defs.
return upb_refcounted_freeze((upb_refcounted*const*)defs, n, s);
err:
for (int i = 0; i < n; i++) {
defs[i]->came_from_user = false;
}
assert(!upb_ok(s));
return false;
}
static size_t run(int i)
{
(void)i;
upb_status status = UPB_STATUS_INIT;
upb_msg_clear(msg, def);
upb_stringsrc_reset(&strsrc, input_str);
upb_decoder_reset(&d, upb_stringsrc_bytesrc(&strsrc), msg);
upb_decoder_decode(&d, &status);
if(!upb_ok(&status)) goto err;
return upb_string_len(input_str);
err:
fprintf(stderr, "Decode error: ");
upb_printerr(&status);
return 0;
}
bool upb_def_freeze(upb_def *const* defs, int n, upb_status *s) {
// First perform validation, in two passes so we can check that we have a
// transitive closure without needing to search.
for (int i = 0; i < n; i++) {
upb_def *def = defs[i];
if (upb_def_isfrozen(def)) {
// Could relax this requirement if it's annoying.
upb_status_seterrmsg(s, "def is already frozen");
goto err;
} else if (def->type == UPB_DEF_FIELD) {
upb_status_seterrmsg(s, "standalone fielddefs can not be frozen");
goto err;
} else {
// Set now to detect transitive closure in the second pass.
def->came_from_user = true;
}
}
// Second pass of validation. Also assign selector bases and indexes, and
// compact tables.
for (int i = 0; i < n; i++) {
upb_msgdef *m = upb_dyncast_msgdef_mutable(defs[i]);
upb_enumdef *e = upb_dyncast_enumdef_mutable(defs[i]);
if (m) {
upb_inttable_compact(&m->itof);
assign_msg_indices(m, s);
} else if (e) {
upb_inttable_compact(&e->iton);
}
}
// Def graph contains FieldDefs between each MessageDef, so double the limit.
int maxdepth = UPB_MAX_MESSAGE_DEPTH * 2;
// Validation all passed; freeze the defs.
return upb_refcounted_freeze((upb_refcounted * const *)defs, n, s, maxdepth);
err:
for (int i = 0; i < n; i++) {
defs[i]->came_from_user = false;
}
assert(!upb_ok(s));
return false;
}
static bool initialize()
{
// Initialize upb state, decode descriptor.
upb_status status = UPB_STATUS_INIT;
upb_symtab *s = upb_symtab_new();
upb_load_descriptor_file_into_symtab(s, MESSAGE_DESCRIPTOR_FILE, &status);
if(!upb_ok(&status)) {
fprintf(stderr, "Error reading descriptor: %s\n",
upb_status_getstr(&status));
return false;
}
def = upb_dyncast_msgdef_const(upb_symtab_lookup(s, MESSAGE_NAME));
if(!def) {
fprintf(stderr, "Error finding symbol '%s'.\n", MESSAGE_NAME);
return false;
}
upb_symtab_unref(s);
// Read the message data itself.
char *str = upb_readfile(MESSAGE_FILE, &len);
if(str == NULL) {
fprintf(stderr, "Error reading " MESSAGE_FILE "\n");
return false;
}
upb_status_uninit(&status);
for (int i = 0; i < NUM_MESSAGES; i++)
msg[i] = upb_stdmsg_new(def);
upb_stringsrc_init(&strsrc);
upb_stringsrc_reset(&strsrc, str, len);
upb_handlers *h = upb_handlers_new();
upb_accessors_reghandlers(h, def);
if (!JIT) h->should_jit = false;
upb_decoder_init(&d, h);
upb_handlers_unref(h);
if (!BYREF) {
// TODO: use byref/byval accessors.
}
return true;
}
int main() {
upb_symtab *symtab = upb_symtab_new();
upb_symtab_add_descriptorproto(symtab);
upb_def *fds = upb_symtab_lookup(
symtab, UPB_STRLIT("google.protobuf.FileDescriptorSet"));
upb_stdio *in = upb_stdio_new();
upb_stdio_reset(in, stdin);
upb_stdio *out = upb_stdio_new();
upb_stdio_reset(out, stdout);
upb_decoder d;
upb_decoder_init(&d, upb_downcast_msgdef(fds));
upb_decoder_reset(&d, upb_stdio_bytesrc(in));
upb_textprinter *p = upb_textprinter_new();
upb_handlers handlers;
upb_handlers_init(&handlers);
upb_textprinter_reset(p, &handlers, upb_stdio_bytesink(out), false);
upb_src *src = upb_decoder_src(&d);
upb_src_sethandlers(src, &handlers);
upb_status status = UPB_STATUS_INIT;
upb_src_run(src, &status);
assert(upb_ok(&status));
upb_status_uninit(&status);
upb_stdio_free(in);
upb_stdio_free(out);
upb_decoder_uninit(&d);
upb_textprinter_free(p);
upb_def_unref(fds);
upb_symtab_unref(symtab);
// Prevent C library from holding buffers open, so Valgrind doesn't see
// memory leaks.
fclose(stdin);
fclose(stdout);
}
static bool initialize()
{
// Initialize upb state, decode descriptor.
upb_status status = UPB_STATUS_INIT;
upb_symtab *s = upb_symtab_new(&s);
upb_load_descriptor_file_into_symtab(s, MESSAGE_DESCRIPTOR_FILE, &status);
if(!upb_ok(&status)) {
fprintf(stderr, "Error reading descriptor: %s\n",
upb_status_getstr(&status));
return false;
}
def = upb_dyncast_msgdef_const(upb_symtab_lookup(s, MESSAGE_NAME, &def));
if(!def) {
fprintf(stderr, "Error finding symbol '%s'.\n", MESSAGE_NAME);
return false;
}
upb_symtab_unref(s, &s);
// Read the message data itself.
input_str = upb_readfile(MESSAGE_FILE, &input_len);
if(input_str == NULL) {
fprintf(stderr, "Error reading " MESSAGE_FILE "\n");
return false;
}
upb_handlers *handlers = upb_handlers_new();
// Cause all messages to be read, but do nothing when they are.
upb_handlerset hset = {NULL, NULL, value, startsubmsg, NULL, NULL, NULL};
upb_handlers_reghandlerset(handlers, def, &hset);
upb_decoder_init(&decoder);
plan = upb_decoderplan_new(handlers, JIT);
upb_decoder_resetplan(&decoder, plan, 0);
upb_handlers_unref(handlers);
upb_stringsrc_init(&stringsrc);
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;
}
/* 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;
}
void lupb_checkstatus(lua_State *L, upb_status *s) {
if (!upb_ok(s)) {
lua_pushstring(L, upb_status_errmsg(s));
lua_error(L);
}
}
int main(int argc, char *argv[]) {
if (argc < 3) {
fprintf(stderr, "Usage: stream_transcode <descfile> <msgname>\n");
return 1;
}
upb_symtab *symtab = upb_symtab_new();
size_t desc_len;
const char *desc = upb_readfile(argv[1], &desc_len);
if (!desc) {
fprintf(stderr, "Couldn't open descriptor file: %s\n", argv[1]);
return 1;
}
upb_status status = UPB_STATUS_INIT;
upb_load_descriptor_into_symtab(symtab, desc, desc_len, &status);
if (!upb_ok(&status)) {
fprintf(stderr, "Error parsing descriptor: %s", upb_status_getstr(&status));
return 1;
}
free((void*)desc);
const upb_def *md = upb_symtab_lookup(symtab, argv[2]);
if (!md) {
fprintf(stderr, "Descriptor did not contain message: %s\n", argv[2]);
return 1;
}
const upb_msgdef *m = upb_dyncast_msgdef_const(md);
if (!m) {
fprintf(stderr, "Def was not a msgdef.\n");
return 1;
}
upb_stdio in, out;
upb_stdio_init(&in);
upb_stdio_init(&out);
upb_stdio_reset(&in, stdin);
upb_stdio_reset(&out, stdout);
upb_handlers *handlers = upb_handlers_new();
upb_textprinter *p = upb_textprinter_new();
upb_textprinter_reset(p, upb_stdio_bytesink(&out), false);
upb_textprinter_reghandlers(handlers, m);
upb_decoder d;
upb_decoder_init(&d, handlers);
upb_decoder_reset(&d, upb_stdio_bytesrc(&in), 0, UPB_NONDELIMITED, p);
upb_status_clear(&status);
upb_decoder_decode(&d, &status);
if (!upb_ok(&status)) {
fprintf(stderr, "Error parsing input: %s", upb_status_getstr(&status));
}
upb_status_uninit(&status);
upb_stdio_uninit(&in);
upb_stdio_uninit(&out);
upb_decoder_uninit(&d);
upb_textprinter_free(p);
upb_def_unref(UPB_UPCAST(m));
upb_symtab_unref(symtab);
// Prevent C library from holding buffers open, so Valgrind doesn't see
// memory leaks.
fclose(stdin);
fclose(stdout);
}
static void check_upb_status(const upb_status* status, const char* msg) {
if (!upb_ok(status)) {
zend_error(E_ERROR, "%s: %s\n", msg, upb_status_errmsg(status));
}
}
/* 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;
}