Protocol Buffers - Google's data interchange format (grpc依赖) https://developers.google.com/protocol-buffers/
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

425 lines
15 KiB

upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
/*
* upb - a minimalist implementation of protocol buffers.
*
* Copyright (c) 2011-2012 Google Inc. See LICENSE for details.
* Author: Josh Haberman <jhaberman@gmail.com>
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
*/
#include "upb/handlers.h"
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
#include <stdlib.h>
#include <string.h>
#include "upb/sink.h"
// Defined for the sole purpose of having a unique pointer value for
// UPB_NO_CLOSURE.
char _upb_noclosure;
static const upb_fieldhandler *getfh(
const upb_handlers *h, upb_selector_t selector) {
assert(selector < upb_handlers_msgdef(h)->selector_count);
upb_fieldhandler* fhbase = (void*)&h->fh_base;
return &fhbase[selector];
}
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
static upb_fieldhandler *getfh_mutable(upb_handlers *h,
upb_selector_t selector) {
return (upb_fieldhandler*)getfh(h, selector);
}
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
bool upb_handlers_isfrozen(const upb_handlers *h) {
return upb_refcounted_isfrozen(upb_upcast(h));
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
}
uint32_t upb_handlers_selectorbaseoffset(const upb_fielddef *f) {
return upb_fielddef_isseq(f) ? 2 : 0;
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
}
uint32_t upb_handlers_selectorcount(const upb_fielddef *f) {
uint32_t ret = 1;
if (upb_fielddef_isstring(f)) ret += 2; // [STARTSTR]/STRING/ENDSTR
if (upb_fielddef_isseq(f)) ret += 2; // STARTSEQ/ENDSEQ
if (upb_fielddef_issubmsg(f)) ret += 1; // [STARTSUBMSG]/ENDSUBMSG
return ret;
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
}
upb_handlertype_t upb_handlers_getprimitivehandlertype(const upb_fielddef *f) {
switch (upb_fielddef_type(f)) {
case UPB_TYPE_INT32:
case UPB_TYPE_ENUM:
return UPB_HANDLER_INT32;
case UPB_TYPE_INT64:
return UPB_HANDLER_INT64;
case UPB_TYPE_UINT32:
return UPB_HANDLER_UINT32;
case UPB_TYPE_UINT64:
return UPB_HANDLER_UINT64;
case UPB_TYPE_FLOAT:
return UPB_HANDLER_FLOAT;
case UPB_TYPE_DOUBLE:
return UPB_HANDLER_DOUBLE;
case UPB_TYPE_BOOL:
return UPB_HANDLER_BOOL;
default: assert(false); return -1; // Invalid input.
}
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
}
bool upb_getselector(
const upb_fielddef *f, upb_handlertype_t type, upb_selector_t *s) {
// If the type checks in this function are a hot-spot, we can introduce a
// separate function that calculates the selector assuming that the type
// is correct (may even want to make it inline for the upb_sink fast-path.
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)) return false;
*s = f->selector_base;
break;
case UPB_HANDLER_STARTSTR:
if (!upb_fielddef_isstring(f)) return false;
*s = f->selector_base + 1;
break;
case UPB_HANDLER_ENDSTR:
if (!upb_fielddef_isstring(f)) return false;
*s = f->selector_base + 2;
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;
*s = f->selector_base;
break;
case UPB_HANDLER_ENDSUBMSG:
if (!upb_fielddef_issubmsg(f)) return false;
*s = f->selector_base + 1;
break;
}
assert(*s < upb_fielddef_msgdef(f)->selector_count);
return true;
}
void upb_handlers_ref(const upb_handlers *h, const void *owner) {
upb_refcounted_ref(upb_upcast(h), owner);
}
void upb_handlers_unref(const upb_handlers *h, const void *owner) {
upb_refcounted_unref(upb_upcast(h), owner);
}
void upb_handlers_donateref(
const upb_handlers *h, const void *from, const void *to) {
upb_refcounted_donateref(upb_upcast(h), from, to);
}
void upb_handlers_checkref(const upb_handlers *h, const void *owner) {
upb_refcounted_checkref(upb_upcast(h), owner);
}
static void do_cleanup(upb_handlers* h, const upb_fielddef *f,
upb_handlertype_t type) {
upb_selector_t selector;
if (!upb_getselector(f, type, &selector)) return;
upb_fieldhandler *fh = getfh_mutable(h, selector);
if (fh->cleanup) fh->cleanup(fh->data);
fh->cleanup = NULL;
fh->data = NULL;
}
static void freehandlers(upb_refcounted *r) {
upb_handlers *h = (upb_handlers*)r;
upb_msg_iter i;
for(upb_msg_begin(&i, h->msg); !upb_msg_done(&i); upb_msg_next(&i)) {
upb_fielddef *f = upb_msg_iter_field(&i);
for (upb_handlertype_t type = 0; type < UPB_HANDLER_MAX; type++)
do_cleanup(h, f, type);
}
upb_msgdef_unref(h->msg, h);
free(h);
}
static void visithandlers(const upb_refcounted *r, upb_refcounted_visit *visit,
void *closure) {
const upb_handlers *h = (const upb_handlers*)r;
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
upb_msg_iter i;
for(upb_msg_begin(&i, h->msg); !upb_msg_done(&i); upb_msg_next(&i)) {
upb_fielddef *f = upb_msg_iter_field(&i);
if (!upb_fielddef_issubmsg(f)) continue;
const upb_handlers *sub = upb_handlers_getsubhandlers(h, f);
if (sub) visit(r, upb_upcast(sub), closure);
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
}
}
upb_handlers *upb_handlers_new(const upb_msgdef *md, const upb_frametype *ft,
const void *owner) {
assert(upb_msgdef_isfrozen(md));
static const struct upb_refcounted_vtbl vtbl = {visithandlers, freehandlers};
size_t fhandlers_size = sizeof(upb_fieldhandler) * md->selector_count;
upb_handlers *h = calloc(sizeof(*h) - sizeof(void*) + fhandlers_size, 1);
if (!h) return NULL;
h->msg = md;
h->ft = ft;
upb_msgdef_ref(h->msg, h);
if (!upb_refcounted_init(upb_upcast(h), &vtbl, owner)) goto oom;
// calloc() above initialized all handlers to NULL.
return h;
oom:
freehandlers(upb_upcast(h));
return NULL;
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
}
bool upb_handlers_freeze(upb_handlers *const*handlers, int n, upb_status *s) {
// TODO: verify we have a transitive closure.
return upb_refcounted_freeze((upb_refcounted*const*)handlers, n, s);
}
const upb_msgdef *upb_handlers_msgdef(const upb_handlers *h) { return h->msg; }
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
const upb_frametype *upb_handlers_frametype(const upb_handlers *h) {
return h->ft;
}
void upb_handlers_setstartmsg(upb_handlers *h, upb_startmsg_handler *handler) {
assert(!upb_handlers_isfrozen(h));
h->startmsg = handler;
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
}
upb_startmsg_handler *upb_handlers_getstartmsg(const upb_handlers *h) {
return h->startmsg;
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
}
void upb_handlers_setendmsg(upb_handlers *h, upb_endmsg_handler *handler) {
assert(!upb_handlers_isfrozen(h));
h->endmsg = handler;
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
}
upb_endmsg_handler *upb_handlers_getendmsg(const upb_handlers *h) {
return h->endmsg;
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
}
// For now we stuff the subhandlers pointer into the fieldhandlers*
// corresponding to the UPB_HANDLER_STARTSUBMSG handler.
static const upb_handlers **subhandlersptr_sel(upb_handlers *h,
upb_selector_t startsubmsg) {
return &getfh_mutable(h, startsubmsg)->subhandlers;
}
static const upb_handlers **subhandlersptr(upb_handlers *h,
const upb_fielddef *f) {
assert(upb_fielddef_issubmsg(f));
upb_selector_t selector;
bool ok = upb_getselector(f, UPB_HANDLER_STARTSUBMSG, &selector);
UPB_ASSERT_VAR(ok, ok);
return subhandlersptr_sel(h, selector);
}
bool upb_handlers_setsubhandlers(upb_handlers *h, const upb_fielddef *f,
const upb_handlers *sub) {
assert(!upb_handlers_isfrozen(h));
if (!upb_fielddef_issubmsg(f)) return false;
if (sub != NULL &&
upb_upcast(upb_handlers_msgdef(sub)) != upb_fielddef_subdef(f)) {
return false;
}
const upb_handlers **stored = subhandlersptr(h, f);
const upb_handlers *old = *stored;
if (old) upb_unref2(old, h);
*stored = sub;
if (sub) upb_ref2(sub, h);
return true;
}
const upb_handlers *upb_handlers_getsubhandlers(const upb_handlers *h,
const upb_fielddef *f) {
const upb_handlers **stored = subhandlersptr((upb_handlers*)h, f);
return *stored;
}
const upb_handlers *upb_handlers_getsubhandlers_sel(const upb_handlers *h,
upb_selector_t sel) {
const upb_handlers **stored = subhandlersptr_sel((upb_handlers*)h, sel);
return *stored;
}
#define SETTER(name, handlerctype, handlertype) \
bool upb_handlers_set ## name(upb_handlers *h, const upb_fielddef *f, \
handlerctype val, void *data, \
upb_handlerfree *cleanup) { \
assert(!upb_handlers_isfrozen(h)); \
if (upb_handlers_msgdef(h) != upb_fielddef_msgdef(f)) return false; \
upb_selector_t selector; \
bool ok = upb_getselector(f, handlertype, &selector); \
if (!ok) return false; \
do_cleanup(h, f, handlertype); \
upb_fieldhandler *fh = getfh_mutable(h, selector); \
fh->handler = (upb_func*)val; \
fh->data = (upb_func*)data; \
fh->cleanup = (upb_func*)cleanup; \
return true; \
} \
SETTER(int32, upb_int32_handler*, UPB_HANDLER_INT32);
SETTER(int64, upb_int64_handler*, UPB_HANDLER_INT64);
SETTER(uint32, upb_uint32_handler*, UPB_HANDLER_UINT32);
SETTER(uint64, upb_uint64_handler*, UPB_HANDLER_UINT64);
SETTER(float, upb_float_handler*, UPB_HANDLER_FLOAT);
SETTER(double, upb_double_handler*, UPB_HANDLER_DOUBLE);
SETTER(bool, upb_bool_handler*, UPB_HANDLER_BOOL);
SETTER(startstr, upb_startstr_handler*, UPB_HANDLER_STARTSTR);
SETTER(string, upb_string_handler*, UPB_HANDLER_STRING);
SETTER(endstr, upb_endfield_handler*, UPB_HANDLER_ENDSTR);
SETTER(startseq, upb_startfield_handler*, UPB_HANDLER_STARTSEQ);
SETTER(startsubmsg, upb_startfield_handler*, UPB_HANDLER_STARTSUBMSG);
SETTER(endsubmsg, upb_endfield_handler*, UPB_HANDLER_ENDSUBMSG);
SETTER(endseq, upb_endfield_handler*, UPB_HANDLER_ENDSEQ);
// Our current implementation of these "alt" functions is, according to the
// letter of the standard, undefined behavior, because we store the
// upb_int32_handler2* to memory and then read it back (and call it) as a
// upb_int32_handler*. Even though both function pointer types take 32-bit
// integer arguments, they are still technically different types (because one
// takes an "int" argument and one takes a "long" argument), and calling a
// function through a pointer to an incompatible type is undefined behavior.
//
// I think it is exceedingly unlikely that "int" and "long" would ever have
// incompatible calling conventions when both are known to be 32 bit signed
// two's complement integers. But if absolute standards-compliance is ever
// required, either due to a practical problem with the undefined behavior or a
// tool that notices the incongruity, we have an available option for being
// perfectly standard-compliant; we can store a bool for every function pointer
// indicating whether it is an "alt" pointer or not. Then at the call site
// (inside upb_sink) we can do:
//
// if (is_alt) {
// upb_int32_handler2 *func = fp;
// func(...);
// } else {
// upb_int32_handler *func = fp;
// func(...);
// }
//
// We could do this now, but it adds complexity and wastes the memory to store
// these useless bools. The bools are useless because the compiler will almost
// certainly optimize away this branch and elide the two calls into a single
// call with the 32-bit parameter calling convention.
#ifdef UPB_TWO_32BIT_TYPES
SETTER(int32alt, upb_int32_handler2*, UPB_HANDLER_INT32);
SETTER(uint32alt, upb_uint32_handler2*, UPB_HANDLER_UINT32);
#endif
#ifdef UPB_TWO_64BIT_TYPES
SETTER(int64alt, upb_int64_handler2*, UPB_HANDLER_INT64);
SETTER(uint64alt, upb_uint64_handler2*, UPB_HANDLER_UINT64);
#endif
#undef SETTER
upb_func *upb_handlers_gethandler(const upb_handlers *h, upb_selector_t s) {
return getfh(h, s)->handler;
}
void *upb_handlers_gethandlerdata(const upb_handlers *h, upb_selector_t s) {
return getfh(h, s)->data;
}
typedef struct {
upb_inttable tab; // maps upb_msgdef* -> upb_handlers*.
upb_handlers_callback *callback;
void *closure;
} dfs_state;
static upb_handlers *newformsg(const upb_msgdef *m, const upb_frametype *ft,
const void *owner,
dfs_state *s) {
upb_handlers *h = upb_handlers_new(m, ft, owner);
if (!h) return NULL;
if (!upb_inttable_insertptr(&s->tab, m, upb_value_ptr(h))) goto oom;
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
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, ft, &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;
}
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
const upb_handlers *upb_handlers_newfrozen(const upb_msgdef *m,
const upb_frametype *ft,
const void *owner,
upb_handlers_callback *callback,
void *closure) {
dfs_state state;
state.callback = callback;
state.closure = closure;
if (!upb_inttable_init(&state.tab, UPB_CTYPE_PTR)) return NULL;
upb_handlers *ret = newformsg(m, ft, owner, &state);
if (!ret) return NULL;
upb_refcounted *r = upb_upcast(ret);
upb_status status = UPB_STATUS_INIT;
bool ok = upb_refcounted_freeze(&r, 1, &status);
UPB_ASSERT_VAR(ok, ok);
upb_status_uninit(&status);
upb_inttable_uninit(&state.tab);
return ret;
upb_stream: all callbacks registered ahead-of-time. This is a significant change to the upb_stream protocol, and should hopefully be the last significant change. All callbacks are now registered ahead-of-time instead of having delegated callbacks registered at runtime, which makes it much easier to aggressively optimize ahead-of-time (like with a JIT). Other impacts of this change: - You no longer need to have loaded descriptor.proto as a upb_def to load other descriptors! This means the special-case code we used for bootstrapping is no longer necessary, and we no longer need to link the descriptor for descriptor.proto into upb. - A client can now register any upb_value as what will be delivered to their value callback, not just a upb_fielddef*. This should allow for other clients to get more bang out of the streaming decoder. This change unfortunately causes a bit of a performance regression -- I think largely due to highly suboptimal code that GCC generates when structs are returned by value. See: http://blog.reverberate.org/2011/03/19/when-a-compilers-slow-code-actually-bites-you/ On the other hand, once we have a JIT this should no longer matter. Performance numbers: plain.parsestream_googlemessage1.upb_table: 374 -> 396 (5.88) plain.parsestream_googlemessage2.upb_table: 616 -> 449 (-27.11) plain.parsetostruct_googlemessage1.upb_table_byref: 268 -> 269 (0.37) plain.parsetostruct_googlemessage1.upb_table_byval: 215 -> 204 (-5.12) plain.parsetostruct_googlemessage2.upb_table_byref: 307 -> 281 (-8.47) plain.parsetostruct_googlemessage2.upb_table_byval: 297 -> 272 (-8.42) omitfp.parsestream_googlemessage1.upb_table: 423 -> 410 (-3.07) omitfp.parsestream_googlemessage2.upb_table: 679 -> 483 (-28.87) omitfp.parsetostruct_googlemessage1.upb_table_byref: 287 -> 282 (-1.74) omitfp.parsetostruct_googlemessage1.upb_table_byval: 226 -> 219 (-3.10) omitfp.parsetostruct_googlemessage2.upb_table_byref: 315 -> 298 (-5.40) omitfp.parsetostruct_googlemessage2.upb_table_byval: 297 -> 287 (-3.37)
14 years ago
}
#define STDMSG_WRITER(type, ctype) \
bool upb_stdmsg_set ## type (const upb_sinkframe *frame, ctype val) { \
const upb_stdmsg_fval *f = upb_sinkframe_handlerdata(frame); \
uint8_t *m = upb_sinkframe_userdata(frame); \
if (f->hasbit > 0) \
*(uint8_t*)&m[f->hasbit / 8] |= 1 << (f->hasbit % 8); \
*(ctype*)&m[f->offset] = val; \
return true; \
} \
STDMSG_WRITER(double, double)
STDMSG_WRITER(float, float)
STDMSG_WRITER(int32, int32_t)
STDMSG_WRITER(int64, int64_t)
STDMSG_WRITER(uint32, uint32_t)
STDMSG_WRITER(uint64, uint64_t)
STDMSG_WRITER(bool, bool)
#undef STDMSG_WRITER