diff --git a/ChangeLog b/ChangeLog index 4cdd9a0..b07ea57 100644 --- a/ChangeLog +++ b/ChangeLog @@ -1,4 +1,6 @@ 0.9 + * optimizations to json_tokener_parse_ex(), printbuf_memappend() + Brent Miller, bdmiller at yahoo dash inc dot com * Don't use this as a variable, so we can compile with a C++ compiler * Add casts from void* to type of assignment when using malloc * Add #ifdef __cplusplus guards to all of the headers diff --git a/json_tokener.c b/json_tokener.c index beaa956..8e8c6d9 100644 --- a/json_tokener.c +++ b/json_tokener.c @@ -7,6 +7,10 @@ * This library is free software; you can redistribute it and/or modify * it under the terms of the MIT license. See COPYING for details. * + * + * Copyright (c) 2008-2009 Yahoo! Inc. All rights reserved. + * The copyrights to the contents of this file are licensed under the MIT License + * (http://www.opensource.org/licenses/mit-license.php) */ #include "config.h" @@ -135,35 +139,68 @@ char* strndup(const char* str, size_t n) #define current tok->stack[tok->depth].current #define obj_field_name tok->stack[tok->depth].obj_field_name +/* Optimization: + * json_tokener_parse_ex() consumed a lot of CPU in its main loop, + * iterating character-by character. A large performance boost is + * achieved by using tighter loops to locally handle units such as + * comments and strings. Loops that handle an entire token within + * their scope also gather entire strings and pass them to + * printbuf_memappend() in a single call, rather than calling + * printbuf_memappend() one char at a time. + * + * POP_CHAR() and ADVANCE_CHAR() macros are used for code that is + * common to both the main loop and the tighter loops. + */ + +/* POP_CHAR(dest, tok) macro: + * Not really a pop()...peeks at the current char and stores it in dest. + * Returns 1 on success, sets tok->err and returns 0 if no more chars. + * Implicit inputs: str, len vars + */ +#define POP_CHAR(dest, tok) \ + (((tok)->char_offset == len) ? \ + (((tok)->depth == 0 && state == json_tokener_state_eatws && saved_state == json_tokener_state_finish) ? \ + (((tok)->err = json_tokener_success), 0) \ + : \ + (((tok)->err = json_tokener_continue), 0) \ + ) : \ + (((dest) = *str), 1) \ + ) + +/* ADVANCE_CHAR() macro: + * Incrementes str & tok->char_offset. + * For convenience of existing conditionals, returns the old value of c (0 on eof) + * Implicit inputs: c var + */ +#define ADVANCE_CHAR(str, tok) \ + ( ++(str), ((tok)->char_offset)++, c) + +/* End optimization macro defs */ + + struct json_object* json_tokener_parse_ex(struct json_tokener *tok, char *str, int len) { struct json_object *obj = NULL; - char c; + char c = '\1'; tok->char_offset = 0; tok->err = json_tokener_success; - do { - if(tok->char_offset == len) { - if(tok->depth == 0 && state == json_tokener_state_eatws && - saved_state == json_tokener_state_finish) - tok->err = json_tokener_success; - else - tok->err = json_tokener_continue; - goto out; - } + while (POP_CHAR(c, tok)) { - c = *str; redo_char: switch(state) { case json_tokener_state_eatws: - if(isspace(c)) { - /* okay */ - } else if(c == '/') { + /* Advance until we change state */ + while (isspace(c)) { + if ((!ADVANCE_CHAR(str, tok)) || (!POP_CHAR(c, tok))) + goto out; + } + if(c == '/') { printbuf_reset(tok->pb); - printbuf_memappend(tok->pb, &c, 1); + printbuf_memappend_fast(tok->pb, &c, 1); state = json_tokener_state_comment_start; } else { state = saved_state; @@ -236,7 +273,7 @@ struct json_object* json_tokener_parse_ex(struct json_tokener *tok, goto redo_char; case json_tokener_state_null: - printbuf_memappend(tok->pb, &c, 1); + printbuf_memappend_fast(tok->pb, &c, 1); if(strncasecmp(json_null_str, tok->pb->buf, min(tok->st_pos+1, strlen(json_null_str))) == 0) { if(tok->st_pos == strlen(json_null_str)) { @@ -261,25 +298,42 @@ struct json_object* json_tokener_parse_ex(struct json_tokener *tok, tok->err = json_tokener_error_parse_comment; goto out; } - printbuf_memappend(tok->pb, &c, 1); + printbuf_memappend_fast(tok->pb, &c, 1); break; case json_tokener_state_comment: - if(c == '*') state = json_tokener_state_comment_end; - printbuf_memappend(tok->pb, &c, 1); - break; + { + /* Advance until we change state */ + char *case_start = str; + while(c != '*') { + if (!ADVANCE_CHAR(str, tok) || !POP_CHAR(c, tok)) { + printbuf_memappend_fast(tok->pb, case_start, str-case_start); + goto out; + } + } + printbuf_memappend_fast(tok->pb, case_start, 1+str-case_start); + state = json_tokener_state_comment_end; + } + break; case json_tokener_state_comment_eol: - if(c == '\n') { + { + /* Advance until we change state */ + char *case_start = str; + while(c != '\n') { + if (!ADVANCE_CHAR(str, tok) || !POP_CHAR(c, tok)) { + printbuf_memappend_fast(tok->pb, case_start, str-case_start); + goto out; + } + } + printbuf_memappend_fast(tok->pb, case_start, str-case_start); MC_DEBUG("json_tokener_comment: %s\n", tok->pb->buf); state = json_tokener_state_eatws; - } else { - printbuf_memappend(tok->pb, &c, 1); } break; case json_tokener_state_comment_end: - printbuf_memappend(tok->pb, &c, 1); + printbuf_memappend_fast(tok->pb, &c, 1); if(c == '/') { MC_DEBUG("json_tokener_comment: %s\n", tok->pb->buf); state = json_tokener_state_eatws; @@ -289,15 +343,27 @@ struct json_object* json_tokener_parse_ex(struct json_tokener *tok, break; case json_tokener_state_string: - if(c == tok->quote_char) { - current = json_object_new_string(tok->pb->buf); - saved_state = json_tokener_state_finish; - state = json_tokener_state_eatws; - } else if(c == '\\') { - saved_state = json_tokener_state_string; - state = json_tokener_state_string_escape; - } else { - printbuf_memappend(tok->pb, &c, 1); + { + /* Advance until we change state */ + char *case_start = str; + while(1) { + if(c == tok->quote_char) { + printbuf_memappend_fast(tok->pb, case_start, str-case_start); + current = json_object_new_string(tok->pb->buf); + saved_state = json_tokener_state_finish; + state = json_tokener_state_eatws; + break; + } else if(c == '\\') { + printbuf_memappend_fast(tok->pb, case_start, str-case_start); + saved_state = json_tokener_state_string; + state = json_tokener_state_string_escape; + break; + } + if (!ADVANCE_CHAR(str, tok) || !POP_CHAR(c, tok)) { + printbuf_memappend_fast(tok->pb, case_start, str-case_start); + goto out; + } + } } break; @@ -306,17 +372,17 @@ struct json_object* json_tokener_parse_ex(struct json_tokener *tok, case '"': case '\\': case '/': - printbuf_memappend(tok->pb, &c, 1); + printbuf_memappend_fast(tok->pb, &c, 1); state = saved_state; break; case 'b': case 'n': case 'r': case 't': - if(c == 'b') printbuf_memappend(tok->pb, "\b", 1); - else if(c == 'n') printbuf_memappend(tok->pb, "\n", 1); - else if(c == 'r') printbuf_memappend(tok->pb, "\r", 1); - else if(c == 't') printbuf_memappend(tok->pb, "\t", 1); + if(c == 'b') printbuf_memappend_fast(tok->pb, "\b", 1); + else if(c == 'n') printbuf_memappend_fast(tok->pb, "\n", 1); + else if(c == 'r') printbuf_memappend_fast(tok->pb, "\r", 1); + else if(c == 't') printbuf_memappend_fast(tok->pb, "\t", 1); state = saved_state; break; case 'u': @@ -331,33 +397,46 @@ struct json_object* json_tokener_parse_ex(struct json_tokener *tok, break; case json_tokener_state_escape_unicode: - if(strchr(json_hex_chars, c)) { - tok->ucs_char += ((unsigned int)hexdigit(c) << ((3-tok->st_pos++)*4)); - if(tok->st_pos == 4) { - unsigned char utf_out[3]; - if (tok->ucs_char < 0x80) { - utf_out[0] = tok->ucs_char; - printbuf_memappend(tok->pb, (char*)utf_out, 1); - } else if (tok->ucs_char < 0x800) { - utf_out[0] = 0xc0 | (tok->ucs_char >> 6); - utf_out[1] = 0x80 | (tok->ucs_char & 0x3f); - printbuf_memappend(tok->pb, (char*)utf_out, 2); - } else { - utf_out[0] = 0xe0 | (tok->ucs_char >> 12); - utf_out[1] = 0x80 | ((tok->ucs_char >> 6) & 0x3f); - utf_out[2] = 0x80 | (tok->ucs_char & 0x3f); - printbuf_memappend(tok->pb, (char*)utf_out, 3); - } - state = saved_state; + /* Note that the following code is inefficient for handling large + * chunks of extended chars, calling printbuf_memappend() once + * for each multi-byte character of input. + * This is a good area for future optimization. + */ + { + /* Advance until we change state */ + while(1) { + if(strchr(json_hex_chars, c)) { + tok->ucs_char += ((unsigned int)hexdigit(c) << ((3-tok->st_pos++)*4)); + if(tok->st_pos == 4) { + unsigned char utf_out[3]; + if (tok->ucs_char < 0x80) { + utf_out[0] = tok->ucs_char; + printbuf_memappend_fast(tok->pb, (char*)utf_out, 1); + } else if (tok->ucs_char < 0x800) { + utf_out[0] = 0xc0 | (tok->ucs_char >> 6); + utf_out[1] = 0x80 | (tok->ucs_char & 0x3f); + printbuf_memappend_fast(tok->pb, (char*)utf_out, 2); + } else { + utf_out[0] = 0xe0 | (tok->ucs_char >> 12); + utf_out[1] = 0x80 | ((tok->ucs_char >> 6) & 0x3f); + utf_out[2] = 0x80 | (tok->ucs_char & 0x3f); + printbuf_memappend_fast(tok->pb, (char*)utf_out, 3); + } + state = saved_state; + break; + } + } else { + tok->err = json_tokener_error_parse_string; + goto out; + } + if (!ADVANCE_CHAR(str, tok) || !POP_CHAR(c, tok)) + goto out; } - } else { - tok->err = json_tokener_error_parse_string; - goto out; } break; case json_tokener_state_boolean: - printbuf_memappend(tok->pb, &c, 1); + printbuf_memappend_fast(tok->pb, &c, 1); if(strncasecmp(json_true_str, tok->pb->buf, min(tok->st_pos+1, strlen(json_true_str))) == 0) { if(tok->st_pos == strlen(json_true_str)) { @@ -382,23 +461,35 @@ struct json_object* json_tokener_parse_ex(struct json_tokener *tok, break; case json_tokener_state_number: - if(c && strchr(json_number_chars, c)) { - printbuf_memappend(tok->pb, &c, 1); - if(c == '.' || c == 'e' || c == 'E') tok->is_double = 1; - } else { - int numi; - double numd; - if(!tok->is_double && sscanf(tok->pb->buf, "%d", &numi) == 1) { - current = json_object_new_int(numi); - } else if(tok->is_double && sscanf(tok->pb->buf, "%lf", &numd) == 1) { - current = json_object_new_double(numd); - } else { - tok->err = json_tokener_error_parse_number; - goto out; + { + /* Advance until we change state */ + char *case_start = str; + int case_len=0; + while(c && strchr(json_number_chars, c)) { + ++case_len; + if(c == '.' || c == 'e') tok->is_double = 1; + if (!ADVANCE_CHAR(str, tok) || !POP_CHAR(c, tok)) { + printbuf_memappend_fast(tok->pb, case_start, case_len); + goto out; + } } - saved_state = json_tokener_state_finish; - state = json_tokener_state_eatws; - goto redo_char; + if (case_len>0) + printbuf_memappend_fast(tok->pb, case_start, case_len); + } + { + int numi; + double numd; + if(!tok->is_double && sscanf(tok->pb->buf, "%d", &numi) == 1) { + current = json_object_new_int(numi); + } else if(tok->is_double && sscanf(tok->pb->buf, "%lf", &numd) == 1) { + current = json_object_new_double(numd); + } else { + tok->err = json_tokener_error_parse_number; + goto out; + } + saved_state = json_tokener_state_finish; + state = json_tokener_state_eatws; + goto redo_char; } break; @@ -452,15 +543,27 @@ struct json_object* json_tokener_parse_ex(struct json_tokener *tok, break; case json_tokener_state_object_field: - if(c == tok->quote_char) { - obj_field_name = strdup(tok->pb->buf); - saved_state = json_tokener_state_object_field_end; - state = json_tokener_state_eatws; - } else if(c == '\\') { - saved_state = json_tokener_state_object_field; - state = json_tokener_state_string_escape; - } else { - printbuf_memappend(tok->pb, &c, 1); + { + /* Advance until we change state */ + char *case_start = str; + while(1) { + if(c == tok->quote_char) { + printbuf_memappend_fast(tok->pb, case_start, str-case_start); + obj_field_name = strdup(tok->pb->buf); + saved_state = json_tokener_state_object_field_end; + state = json_tokener_state_eatws; + break; + } else if(c == '\\') { + printbuf_memappend_fast(tok->pb, case_start, str-case_start); + saved_state = json_tokener_state_object_field; + state = json_tokener_state_string_escape; + break; + } + if (!ADVANCE_CHAR(str, tok) || !POP_CHAR(c, tok)) { + printbuf_memappend_fast(tok->pb, case_start, str-case_start); + goto out; + } + } } break; @@ -506,15 +609,17 @@ struct json_object* json_tokener_parse_ex(struct json_tokener *tok, break; } - str++; - tok->char_offset++; - } while(c); - - if(state != json_tokener_state_finish && - saved_state != json_tokener_state_finish) - tok->err = json_tokener_error_parse_eof; + if (!ADVANCE_CHAR(str, tok)) + goto out; + } /* while(POP_CHAR) */ out: + if (!c) { /* We hit an eof char (0) */ + if(state != json_tokener_state_finish && + saved_state != json_tokener_state_finish) + tok->err = json_tokener_error_parse_eof; + } + if(tok->err == json_tokener_success) return json_object_get(current); MC_DEBUG("json_tokener_parse_ex: error %s at offset %d\n", json_tokener_errors[tok->err], tok->char_offset); diff --git a/printbuf.c b/printbuf.c index a2182f4..a9c711c 100644 --- a/printbuf.c +++ b/printbuf.c @@ -7,6 +7,10 @@ * This library is free software; you can redistribute it and/or modify * it under the terms of the MIT license. See COPYING for details. * + * + * Copyright (c) 2008-2009 Yahoo! Inc. All rights reserved. + * The copyrights to the contents of this file are licensed under the MIT License + * (http://www.opensource.org/licenses/mit-license.php) */ #include "config.h" @@ -118,16 +122,15 @@ int sprintbuf(struct printbuf *p, const char *msg, ...) if output is truncated whereas some return the number of bytes that would have been writen - this code handles both cases. */ if(size == -1 || size > 127) { - int ret; va_start(ap, msg); - size = vasprintf(&t, msg, ap); + if((size = vasprintf(&t, msg, ap)) == -1) return -1; va_end(ap); - if(size == -1) return -1; - ret = printbuf_memappend(p, t, size); + printbuf_memappend(p, t, size); free(t); - return ret; + return size; } else { - return printbuf_memappend(p, buf, size); + printbuf_memappend(p, buf, size); + return size; } } diff --git a/printbuf.h b/printbuf.h index 20c81cb..53fa921 100644 --- a/printbuf.h +++ b/printbuf.h @@ -7,6 +7,10 @@ * This library is free software; you can redistribute it and/or modify * it under the terms of the MIT license. See COPYING for details. * + * + * Copyright (c) 2008-2009 Yahoo! Inc. All rights reserved. + * The copyrights to the contents of this file are licensed under the MIT License + * (http://www.opensource.org/licenses/mit-license.php) */ #ifndef _printbuf_h_ @@ -27,9 +31,23 @@ struct printbuf { extern struct printbuf* printbuf_new(void); +/* As an optimization, printbuf_memappend is defined as a macro that + * handles copying data if the buffer is large enough; otherwise it + * invokes printbuf_memappend_real() which performs the heavy lifting + * of realloc()ing the buffer and copying data. + */ extern int printbuf_memappend(struct printbuf *p, const char *buf, int size); +#define printbuf_memappend_fast(p, bufptr, bufsize) \ +do { \ + if ((p->size - p->bpos) > bufsize) { \ + memcpy(p->buf + p->bpos, (bufptr), bufsize); \ + p->bpos += bufsize; \ + p->buf[p->bpos]= '\0'; \ + } else { printbuf_memappend(p, (bufptr), bufsize); } \ +} while (0) + extern int sprintbuf(struct printbuf *p, const char *msg, ...);