// TODO(haberman): discard upb_handlers* objects that do not actually have any
// handlers set and cannot reach any upb_handlers* object that does. This is
// slightly tricky to do correctly.
static upb_handlers *newformsg(const upb_msgdef *m, const void *owner,
dfs_state *s) {
upb_handlers *h = upb_handlers_new(m, owner);
if (!h) return NULL;
if (!upb_inttable_insertptr(&s->tab, m, upb_value_ptr(h))) goto oom;
s->callback(s->closure, h);
// For each submessage field, get or create a handlers object and set it as
// the subhandlers.
upb_msg_iter i;
for(upb_msg_begin(&i, m); !upb_msg_done(&i); upb_msg_next(&i)) {
upb_fielddef *f = upb_msg_iter_field(&i);
if (!upb_fielddef_issubmsg(f)) continue;
const upb_msgdef *subdef = upb_downcast_msgdef(upb_fielddef_subdef(f));
upb_value subm_ent;
if (upb_inttable_lookupptr(&s->tab, subdef, &subm_ent)) {
upb_handlers_setsubhandlers(h, f, upb_value_getptr(subm_ent));
} else {
upb_handlers *sub_mh = newformsg(subdef, &sub_mh, s);
if (!sub_mh) goto oom;
upb_handlers_setsubhandlers(h, f, sub_mh);
upb_handlers_unref(sub_mh, &sub_mh);
}
}
return h;
oom:
upb_handlers_unref(h, owner);
return NULL;
}
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;
}
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;
}
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);
}
// Creates a new Handlers instance.
static Handlers* New() {
return static_cast<Handlers*>(upb_handlers_new());
}
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;
}
