From 6bdbb45e88e7b88b294dfb6e4cb493cbc3c8cf74 Mon Sep 17 00:00:00 2001 From: Joshua Haberman Date: Sun, 13 Feb 2011 12:59:54 -0800 Subject: Merged core/ and stream/ -> src/. The split wasn't worth it. --- src/upb_msg.c | 253 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 253 insertions(+) create mode 100644 src/upb_msg.c (limited to 'src/upb_msg.c') diff --git a/src/upb_msg.c b/src/upb_msg.c new file mode 100644 index 0000000..9dfbea4 --- /dev/null +++ b/src/upb_msg.c @@ -0,0 +1,253 @@ +/* + * upb - a minimalist implementation of protocol buffers. + * + * Copyright (c) 2010 Joshua Haberman. See LICENSE for details. + * + * Data structure for storing a message of protobuf data. + */ + +#include "upb_msg.h" +#include "upb_decoder.h" +#include "upb_strstream.h" + +static uint32_t upb_round_up_pow2(uint32_t v) { + // http://graphics.stanford.edu/~seander/bithacks.html#RoundUpPowerOf2 + v--; + v |= v >> 1; + v |= v >> 2; + v |= v >> 4; + v |= v >> 8; + v |= v >> 16; + v++; + return v; +} + +static void upb_elem_free(upb_value v, upb_fielddef *f) { + switch(f->type) { + case UPB_TYPE(MESSAGE): + case UPB_TYPE(GROUP): + _upb_msg_free(upb_value_getmsg(v), upb_downcast_msgdef(f->def)); + break; + case UPB_TYPE(STRING): + case UPB_TYPE(BYTES): + _upb_string_free(upb_value_getstr(v)); + break; + default: + abort(); + } +} + +static void upb_elem_unref(upb_value v, upb_fielddef *f) { + assert(upb_elem_ismm(f)); + upb_atomic_refcount_t *refcount = upb_value_getrefcount(v); + if (refcount && upb_atomic_unref(refcount)) + upb_elem_free(v, f); +} + +static void upb_field_free(upb_value v, upb_fielddef *f) { + if (upb_isarray(f)) { + _upb_array_free(upb_value_getarr(v), f); + } else { + upb_elem_free(v, f); + } +} + +static void upb_field_unref(upb_value v, upb_fielddef *f) { + assert(upb_field_ismm(f)); + upb_atomic_refcount_t *refcount = upb_value_getrefcount(v); + if (refcount && upb_atomic_unref(refcount)) + upb_field_free(v, f); +} + + +/* upb_array ******************************************************************/ + +upb_array *upb_array_new(void) { + upb_array *arr = malloc(sizeof(*arr)); + upb_atomic_refcount_init(&arr->refcount, 1); + arr->size = 0; + arr->len = 0; + arr->ptr = NULL; + return arr; +} + +void upb_array_recycle(upb_array **_arr, upb_fielddef *f) { + upb_array *arr = *_arr; + if(arr && upb_atomic_only(&arr->refcount)) { + arr->len = 0; + } else { + upb_array_unref(arr, f); + *_arr = upb_array_new(); + } +} + +void _upb_array_free(upb_array *arr, upb_fielddef *f) { + if (upb_elem_ismm(f)) { + // Need to release refs on sub-objects. + upb_valuetype_t type = upb_elem_valuetype(f); + for (upb_arraylen_t i = 0; i < arr->size; i++) { + upb_valueptr p = _upb_array_getptr(arr, f, i); + upb_elem_unref(upb_value_read(p, type), f); + } + } + free(arr->ptr); + free(arr); +} + +void upb_array_resize(upb_array *arr, upb_fielddef *f, upb_arraylen_t len) { + size_t type_size = upb_types[f->type].size; + upb_arraylen_t old_size = arr->size; + if (old_size < len) { + // Need to resize. + size_t new_size = upb_round_up_pow2(len); + arr->ptr = realloc(arr->ptr, new_size * type_size); + arr->size = new_size; + memset(arr->ptr + (old_size * type_size), 0, + (new_size - old_size) * type_size); + } + arr->len = len; +} + + +/* upb_msg ********************************************************************/ + +upb_msg *upb_msg_new(upb_msgdef *md) { + upb_msg *msg = malloc(md->size); + // Clear all set bits and cached pointers. + memset(msg, 0, md->size); + upb_atomic_refcount_init(&msg->refcount, 1); + return msg; +} + +void _upb_msg_free(upb_msg *msg, upb_msgdef *md) { + // Need to release refs on all sub-objects. + upb_msg_iter i; + for(i = upb_msg_begin(md); !upb_msg_done(i); i = upb_msg_next(md, i)) { + upb_fielddef *f = upb_msg_iter_field(i); + upb_valueptr p = _upb_msg_getptr(msg, f); + upb_valuetype_t type = upb_field_valuetype(f); + if (upb_field_ismm(f)) upb_field_unref(upb_value_read(p, type), f); + } + free(msg); +} + +void upb_msg_recycle(upb_msg **_msg, upb_msgdef *msgdef) { + upb_msg *msg = *_msg; + if(msg && upb_atomic_only(&msg->refcount)) { + upb_msg_clear(msg, msgdef); + } else { + upb_msg_unref(msg, msgdef); + *_msg = upb_msg_new(msgdef); + } +} + +INLINE void upb_msg_sethas(upb_msg *msg, upb_fielddef *f) { + msg->data[f->set_bit_offset] |= f->set_bit_mask; +} + +static upb_valueptr upb_msg_getappendptr(upb_msg *msg, upb_fielddef *f) { + upb_valueptr p = _upb_msg_getptr(msg, f); + if (upb_isarray(f)) { + // Create/recycle/resize the array if necessary, and find a pointer to + // a newly-appended element. + if (!upb_msg_has(msg, f)) { + upb_array_recycle(p.arr, f); + upb_msg_sethas(msg, f); + } + assert(*p.arr != NULL); + upb_arraylen_t oldlen = upb_array_len(*p.arr); + upb_array_resize(*p.arr, f, oldlen + 1); + p = _upb_array_getptr(*p.arr, f, oldlen); + } + return p; +} + +static void upb_msg_appendval(upb_msg *msg, upb_fielddef *f, upb_value val) { + upb_valueptr p = upb_msg_getappendptr(msg, f); + if (upb_isstring(f)) { + // We do: + // - upb_string_recycle(), upb_string_substr() instead of + // - upb_string_unref(), upb_string_getref() + // because we can conveniently cache these upb_string objects in the + // upb_msg, whereas the upb_src who is sending us these strings may not + // have a good way of caching them. This saves the upb_src from allocating + // new upb_strings all the time to give us. + // + // If you were using this to copy one upb_msg to another this would + // allocate string objects whereas a upb_string_getref could have avoided + // those allocations completely; if this is an issue, we could make it an + // option of the upb_msgpopulator which behavior is desired. + upb_string *src = upb_value_getstr(val); + upb_string_recycle(p.str); + upb_string_substr(*p.str, src, 0, upb_string_len(src)); + } else { + upb_value_write(p, val, f->type); + } + upb_msg_sethas(msg, f); +} + +upb_msg *upb_msg_appendmsg(upb_msg *msg, upb_fielddef *f, upb_msgdef *msgdef) { + upb_valueptr p = upb_msg_getappendptr(msg, f); + if (upb_isarray(f) || !upb_msg_has(msg, f)) { + upb_msg_recycle(p.msg, msgdef); + upb_msg_sethas(msg, f); + } + return *p.msg; +} + + +/* upb_msgpopulator ***********************************************************/ + +void upb_msgpopulator_init(upb_msgpopulator *p) { + upb_status_init(&p->status); +} + +void upb_msgpopulator_reset(upb_msgpopulator *p, upb_msg *m, upb_msgdef *md) { + p->top = p->stack; + p->limit = p->stack + sizeof(p->stack); + p->top->msg = m; + p->top->msgdef = md; +} + +void upb_msgpopulator_uninit(upb_msgpopulator *p) { + upb_status_uninit(&p->status); +} + +static upb_flow_t upb_msgpopulator_value(void *_p, upb_fielddef *f, upb_value val) { + upb_msgpopulator *p = _p; + upb_msg_appendval(p->top->msg, f, val); + return UPB_CONTINUE; +} + +static upb_flow_t upb_msgpopulator_startsubmsg(void *_p, upb_fielddef *f, + upb_handlers *delegate_to) { + upb_msgpopulator *p = _p; + (void)delegate_to; + upb_msg *oldmsg = p->top->msg; + if (++p->top == p->limit) { + upb_seterr(&p->status, UPB_ERROR, "Exceeded maximum nesting"); + return UPB_BREAK; + } + upb_msgdef *msgdef = upb_downcast_msgdef(f->def); + p->top->msgdef = msgdef; + p->top->msg = upb_msg_appendmsg(oldmsg, f, msgdef); + return UPB_CONTINUE; +} + +static upb_flow_t upb_msgpopulator_endsubmsg(void *_p) { + upb_msgpopulator *p = _p; + --p->top; + return UPB_CONTINUE; +} + +void upb_msgpopulator_register_handlers(upb_msgpopulator *p, upb_handlers *h) { + static upb_handlerset handlerset = { + NULL, // startmsg + NULL, // endmsg + &upb_msgpopulator_value, + &upb_msgpopulator_startsubmsg, + &upb_msgpopulator_endsubmsg, + }; + upb_register_handlerset(h, &handlerset); + upb_set_handler_closure(h, p, &p->status); +} -- cgit v1.2.3