X-Git-Url: https://git.librecmc.org/?a=blobdiff_plain;f=src%2Fregex%2Fregex.c;h=0a2671fa0f24e2c8fe492ab69d80c222fa948ba3;hb=3cd74a0fcf964bee1b4bfd185fb8aa755a3fd089;hp=a5d84ce95612cdd555be13632ed4cfe8f06b0b8b;hpb=4eb238e5115333e3a5724e4ba787e8e9c60cdf2e;p=oweals%2Fgnunet.git diff --git a/src/regex/regex.c b/src/regex/regex.c index a5d84ce95..0a2671fa0 100644 --- a/src/regex/regex.c +++ b/src/regex/regex.c @@ -19,7 +19,9 @@ */ /** * @file src/regex/regex.c - * @brief library to create automatons from regular expressions + * @brief library to create Deterministic Finite Automatons (DFAs) from regular + * expressions (regexes). Used by mesh for announcing regexes in the network and + * matching strings against published regexes. * @author Maximilian Szengel */ #include "platform.h" @@ -28,30 +30,50 @@ #include "gnunet_regex_lib.h" #include "regex_internal.h" - /** - * Constant for how many bits the initial string regex should have. + * Set this to GNUNET_YES to enable state naming. Used to debug NFA->DFA + * creation. Disabled by default for better performance. */ -#define INITIAL_BITS 8 - +#define REGEX_DEBUG_DFA GNUNET_NO /** - * Set of states. + * Set of states using MDLL API. */ -struct GNUNET_REGEX_StateSet +struct GNUNET_REGEX_StateSet_MDLL { /** - * Array of states. + * MDLL of states. + */ + struct GNUNET_REGEX_State *head; + + /** + * MDLL of states. */ - struct GNUNET_REGEX_State **states; + struct GNUNET_REGEX_State *tail; /** - * Length of the 'states' array. + * Length of the MDLL. */ unsigned int len; }; +/** + * Append state to the given StateSet ' + * + * @param set set to be modified + * @param state state to be appended + */ +static void +state_set_append (struct GNUNET_REGEX_StateSet *set, + struct GNUNET_REGEX_State *state) +{ + if (set->off == set->size) + GNUNET_array_grow (set->states, set->size, set->size * 2 + 4); + set->states[set->off++] = state; +} + + /** * Compare two strings for equality. If either is NULL they are not equal. * @@ -86,7 +108,6 @@ state_add_transition (struct GNUNET_REGEX_Context *ctx, struct GNUNET_REGEX_State *from_state, const char *label, struct GNUNET_REGEX_State *to_state) { - int is_dup; struct GNUNET_REGEX_Transition *t; struct GNUNET_REGEX_Transition *oth; @@ -96,22 +117,15 @@ state_add_transition (struct GNUNET_REGEX_Context *ctx, return; } - // Do not add duplicate state transitions - is_dup = GNUNET_NO; + /* Do not add duplicate state transitions */ for (t = from_state->transitions_head; NULL != t; t = t->next) { if (t->to_state == to_state && 0 == nullstrcmp (t->label, label) && t->from_state == from_state) - { - is_dup = GNUNET_YES; - break; - } + return; } - if (GNUNET_YES == is_dup) - return; - - // sort transitions by label + /* sort transitions by label */ for (oth = from_state->transitions_head; NULL != oth; oth = oth->next) { if (0 < nullstrcmp (oth->label, label)) @@ -128,7 +142,7 @@ state_add_transition (struct GNUNET_REGEX_Context *ctx, t->to_state = to_state; t->from_state = from_state; - // Add outgoing transition to 'from_state' + /* Add outgoing transition to 'from_state' */ from_state->transition_count++; GNUNET_CONTAINER_DLL_insert_before (from_state->transitions_head, from_state->transitions_tail, oth, t); @@ -151,10 +165,12 @@ state_remove_transition (struct GNUNET_REGEX_State *state, if (transition->from_state != state) return; + GNUNET_free_non_null (transition->label); + state->transition_count--; GNUNET_CONTAINER_DLL_remove (state->transitions_head, state->transitions_tail, transition); - GNUNET_free_non_null (transition->label); + GNUNET_free (transition); } @@ -172,11 +188,8 @@ state_remove_transition (struct GNUNET_REGEX_State *state, static int state_compare (const void *a, const void *b) { - struct GNUNET_REGEX_State **s1; - struct GNUNET_REGEX_State **s2; - - s1 = (struct GNUNET_REGEX_State **) a; - s2 = (struct GNUNET_REGEX_State **) b; + struct GNUNET_REGEX_State **s1 = (struct GNUNET_REGEX_State **) a; + struct GNUNET_REGEX_State **s2 = (struct GNUNET_REGEX_State **) b; return (*s1)->id - (*s2)->id; } @@ -221,10 +234,7 @@ state_get_edges (struct GNUNET_REGEX_State *s, struct GNUNET_REGEX_Edge *edges) * * @param sset1 first state set * @param sset2 second state set - * - * @return an integer less than, equal to, or greater than zero - * if the first argument is considered to be respectively - * less than, equal to, or greater than the second. + * @return 0 if the sets are equal, otherwise non-zero */ static int state_set_compare (struct GNUNET_REGEX_StateSet *sset1, @@ -236,15 +246,14 @@ state_set_compare (struct GNUNET_REGEX_StateSet *sset1, if (NULL == sset1 || NULL == sset2) return 1; - result = sset1->len - sset2->len; - - for (i = 0; i < sset1->len; i++) - { - if (0 != result) + result = sset1->off - sset2->off; + if (result < 0) + return -1; + if (result > 0) + return 1; + for (i = 0; i < sset1->off; i++) + if (0 != (result = state_compare (&sset1->states[i], &sset2->states[i]))) break; - - result = state_compare (&sset1->states[i], &sset2->states[i]); - } return result; } @@ -257,11 +266,8 @@ state_set_compare (struct GNUNET_REGEX_StateSet *sset1, static void state_set_clear (struct GNUNET_REGEX_StateSet *set) { - if (NULL != set) - { - GNUNET_free_non_null (set->states); - GNUNET_free (set); - } + GNUNET_array_grow (set->states, set->size, 0); + set->off = 0; } @@ -302,17 +308,13 @@ automaton_destroy_state (struct GNUNET_REGEX_State *s) GNUNET_free_non_null (s->name); GNUNET_free_non_null (s->proof); - + state_set_clear (&s->nfa_set); for (t = s->transitions_head; NULL != t; t = next_t) { next_t = t->next; - GNUNET_CONTAINER_DLL_remove (s->transitions_head, s->transitions_tail, t); - GNUNET_free_non_null (t->label); - GNUNET_free (t); + state_remove_transition (s, t); } - state_set_clear (s->nfa_set); - GNUNET_free (s); } @@ -329,40 +331,36 @@ static void automaton_remove_state (struct GNUNET_REGEX_Automaton *a, struct GNUNET_REGEX_State *s) { - struct GNUNET_REGEX_State *ss; struct GNUNET_REGEX_State *s_check; struct GNUNET_REGEX_Transition *t_check; + struct GNUNET_REGEX_Transition *t_check_next; if (NULL == a || NULL == s) return; - // remove state - ss = s; - GNUNET_CONTAINER_DLL_remove (a->states_head, a->states_tail, s); - a->state_count--; - - // remove all transitions leading to this state + /* remove all transitions leading to this state */ for (s_check = a->states_head; NULL != s_check; s_check = s_check->next) { for (t_check = s_check->transitions_head; NULL != t_check; - t_check = t_check->next) + t_check = t_check_next) { - if (t_check->to_state == ss) - { - GNUNET_CONTAINER_DLL_remove (s_check->transitions_head, - s_check->transitions_tail, t_check); - s_check->transition_count--; - } + t_check_next = t_check->next; + if (t_check->to_state == s) + state_remove_transition (s_check, t_check); } } - automaton_destroy_state (ss); + /* remove state */ + GNUNET_CONTAINER_DLL_remove (a->states_head, a->states_tail, s); + a->state_count--; + + automaton_destroy_state (s); } /** * Merge two states into one. Will merge 's1' and 's2' into 's1' and destroy - * 's2'. + * 's2'. 's1' will contain all (non-duplicate) outgoing transitions of 's2'. * * @param ctx context * @param a automaton @@ -379,16 +377,13 @@ automaton_merge_states (struct GNUNET_REGEX_Context *ctx, struct GNUNET_REGEX_Transition *t_check; struct GNUNET_REGEX_Transition *t; struct GNUNET_REGEX_Transition *t_next; - char *new_name; int is_dup; - GNUNET_assert (NULL != ctx && NULL != a && NULL != s1 && NULL != s2); - if (s1 == s2) return; - // 1. Make all transitions pointing to s2 point to s1, unless this transition - // does not already exists, if it already exists remove transition. + /* 1. Make all transitions pointing to s2 point to s1, unless this transition + * does not already exists, if it already exists remove transition. */ for (s_check = a->states_head; NULL != s_check; s_check = s_check->next) { for (t_check = s_check->transitions_head; NULL != t_check; t_check = t_next) @@ -411,19 +406,23 @@ automaton_merge_states (struct GNUNET_REGEX_Context *ctx, } } - // 2. Add all transitions from s2 to sX to s1 + /* 2. Add all transitions from s2 to sX to s1 */ for (t_check = s2->transitions_head; NULL != t_check; t_check = t_check->next) { if (t_check->to_state != s1) state_add_transition (ctx, s1, t_check->label, t_check->to_state); } - // 3. Rename s1 to {s1,s2} + /* 3. Rename s1 to {s1,s2} */ +#if REGEX_DEBUG_DFA + char *new_name; + new_name = s1->name; GNUNET_asprintf (&s1->name, "{%s,%s}", new_name, s2->name); GNUNET_free (new_name); +#endif - // remove state + /* remove state */ GNUNET_CONTAINER_DLL_remove (a->states_head, a->states_tail, s2); a->state_count--; automaton_destroy_state (s2); @@ -539,149 +538,367 @@ GNUNET_REGEX_automaton_traverse (const struct GNUNET_REGEX_Automaton *a, /** - * Context for adding strided transitions to a DFA. + * String container for faster string operations. */ -struct GNUNET_REGEX_Strided_Context +struct StringBuffer { /** - * Length of the strides. + * Buffer holding the string (may start in the middle!); + * NOT 0-terminated! */ - const unsigned int stride; + char *sbuf; /** - * Strided transitions DLL. New strided transitions will be stored in this DLL - * and afterwards added to the DFA. + * Allocated buffer. */ - struct GNUNET_REGEX_Transition *transitions_head; + char *abuf; + + /** + * Length of the string in the buffer. + */ + size_t slen; /** - * Strided transitions DLL. + * Number of bytes allocated for 'sbuf' */ - struct GNUNET_REGEX_Transition *transitions_tail; + unsigned int blen; + + /** + * Buffer currently represents "NULL" (not the empty string!) + */ + int16_t null_flag; + + /** + * If this entry is part of the last/current generation array, + * this flag is GNUNET_YES if the last and current generation are + * identical (and thus copying is unnecessary if the value didn't + * change). This is used in an optimization that improves + * performance by about 1% --- if we use int16_t here. With just + * "int" for both flags, performance drops (on my system) significantly, + * most likely due to increased cache misses. + */ + int16_t synced; + }; /** - * Recursive helper function to add strides to a DFA. + * Compare two strings for equality. If either is NULL they are not equal. * - * @param cls context, contains stride length and strided transitions DLL. - * @param depth current depth of the depth-first traversal of the graph. - * @param label current label, string that contains all labels on the path from - * 'start' to 's'. - * @param start start state for the depth-first traversal of the graph. - * @param s current state in the depth-first traversal + * @param s1 first string for comparison. + * @param s2 second string for comparison. + * + * @return 0 if the strings are the same or both NULL, 1 or -1 if not. */ -void -add_multi_strides_to_dfa_helper (void *cls, const unsigned int depth, - char *label, struct GNUNET_REGEX_State *start, - struct GNUNET_REGEX_State *s) +static int +sb_nullstrcmp (const struct StringBuffer *s1, + const struct StringBuffer *s2) { - struct GNUNET_REGEX_Strided_Context *ctx = cls; - struct GNUNET_REGEX_Transition *t; - char *new_label; + if ( (GNUNET_YES == s1->null_flag) && + (GNUNET_YES == s2->null_flag) ) + return 0; + if ( (GNUNET_YES == s1->null_flag) || + (GNUNET_YES == s2->null_flag) ) + return -1; + if (s1->slen != s2->slen) + return -1; + return memcmp (s1->sbuf, s2->sbuf, s1->slen); +} + - if (depth == ctx->stride) - { - t = GNUNET_malloc (sizeof (struct GNUNET_REGEX_Transition)); - t->label = GNUNET_strdup (label); - t->to_state = s; - t->from_state = start; - GNUNET_CONTAINER_DLL_insert (ctx->transitions_head, ctx->transitions_tail, - t); - } - else - { - for (t = s->transitions_head; NULL != t; t = t->next) - { - /* Do not consider self-loops, because it end's up in too many - * transitions */ - if (t->to_state == t->from_state) - continue; +/** + * Compare two strings for equality. + * + * @param s1 first string for comparison. + * @param s2 second string for comparison. + * + * @return 0 if the strings are the same, 1 or -1 if not. + */ +static int +sb_strcmp (const struct StringBuffer *s1, + const struct StringBuffer *s2) +{ + if (s1->slen != s2->slen) + return -1; + return memcmp (s1->sbuf, s2->sbuf, s1->slen); +} + - if (NULL != label) - { - GNUNET_asprintf (&new_label, "%s%s", label, t->label); - } - else - new_label = GNUNET_strdup (t->label); +/** + * Reallocate the buffer of 'ret' to fit 'nlen' characters; + * move the existing string to the beginning of the new buffer. + * + * @param ret current buffer, to be updated + * @param nlen target length for the buffer, must be at least ret->slen + */ +static void +sb_realloc (struct StringBuffer *ret, + size_t nlen) +{ + char *old; + + GNUNET_assert (nlen >= ret->slen); + old = ret->abuf; + ret->abuf = GNUNET_malloc (nlen); + ret->blen = nlen; + memcpy (ret->abuf, + ret->sbuf, + ret->slen); + ret->sbuf = ret->abuf; + GNUNET_free_non_null (old); +} + - add_multi_strides_to_dfa_helper (cls, (depth + 1), new_label, start, - t->to_state); - } - } - GNUNET_free_non_null (label); +/** + * Append a string. + * + * @param ret where to write the result + * @param sarg string to append + */ +static void +sb_append (struct StringBuffer *ret, + const struct StringBuffer *sarg) +{ + if (GNUNET_YES == ret->null_flag) + ret->slen = 0; + ret->null_flag = GNUNET_NO; + if (ret->blen < sarg->slen + ret->slen) + sb_realloc (ret, ret->blen + sarg->slen + 128); + memcpy (&ret->sbuf[ret->slen], + sarg->sbuf, + sarg->slen); + ret->slen += sarg->slen; } + +/** + * Append a C string. + * + * @param ret where to write the result + * @param cstr string to append + */ +static void +sb_append_cstr (struct StringBuffer *ret, + const char *cstr) +{ + size_t cstr_len = strlen (cstr); + + if (GNUNET_YES == ret->null_flag) + ret->slen = 0; + ret->null_flag = GNUNET_NO; + if (ret->blen < cstr_len + ret->slen) + sb_realloc (ret, ret->blen + cstr_len + 128); + memcpy (&ret->sbuf[ret->slen], + cstr, + cstr_len); + ret->slen += cstr_len; +} + /** - * Function called for each state in the DFA. Starts a traversal of depth set in - * context starting from state 's'. + * Wrap a string buffer, that is, set ret to the format string + * which contains an "%s" which is to be replaced with the original + * content of 'ret'. Note that optimizing this function is not + * really worth it, it is rarely called. + * + * @param ret where to write the result and take the input for %.*s from + * @param format format string, fprintf-style, with exactly one "%.*s" + * @param extra_chars how long will the result be, in addition to 'sarg' length + */ +static void +sb_wrap (struct StringBuffer *ret, + const char *format, + size_t extra_chars) +{ + char *temp; + + if (GNUNET_YES == ret->null_flag) + ret->slen = 0; + ret->null_flag = GNUNET_NO; + temp = GNUNET_malloc (ret->slen + extra_chars + 1); + GNUNET_snprintf (temp, + ret->slen + extra_chars + 1, + format, + (int) ret->slen, + ret->sbuf); + GNUNET_free_non_null (ret->abuf); + ret->abuf = temp; + ret->sbuf = temp; + ret->blen = ret->slen + extra_chars + 1; + ret->slen = ret->slen + extra_chars; +} + + +/** + * Format a string buffer. Note that optimizing this function is not + * really worth it, it is rarely called. * - * @param cls context. - * @param count not used. - * @param s current state. + * @param ret where to write the result + * @param format format string, fprintf-style, with exactly one "%.*s" + * @param extra_chars how long will the result be, in addition to 'sarg' length + * @param sarg string to print into the format */ -void -add_multi_strides_to_dfa (void *cls, const unsigned int count, - struct GNUNET_REGEX_State *s) +static void +sb_printf1 (struct StringBuffer *ret, + const char *format, + size_t extra_chars, + const struct StringBuffer *sarg) { - add_multi_strides_to_dfa_helper (cls, 0, NULL, s, s); + if (ret->blen < sarg->slen + extra_chars + 1) + sb_realloc (ret, + sarg->slen + extra_chars + 1); + ret->null_flag = GNUNET_NO; + ret->sbuf = ret->abuf; + ret->slen = sarg->slen + extra_chars; + GNUNET_snprintf (ret->sbuf, + ret->blen, + format, + (int) sarg->slen, + sarg->sbuf); } /** - * Adds multi-strided transitions to the given 'dfa'. + * Format a string buffer. * - * @param regex_ctx regex context needed to add transitions to the automaton. - * @param dfa DFA to which the multi strided transitions should be added. - * @param stride_len length of the strides. + * @param ret where to write the result + * @param format format string, fprintf-style, with exactly two "%.*s" + * @param extra_chars how long will the result be, in addition to 'sarg1/2' length + * @param sarg1 first string to print into the format + * @param sarg2 second string to print into the format */ -void -GNUNET_REGEX_add_multi_strides_to_dfa (struct GNUNET_REGEX_Context *regex_ctx, - struct GNUNET_REGEX_Automaton *dfa, - const unsigned int stride_len) +static void +sb_printf2 (struct StringBuffer *ret, + const char *format, + size_t extra_chars, + const struct StringBuffer *sarg1, + const struct StringBuffer *sarg2) { - struct GNUNET_REGEX_Strided_Context ctx = { stride_len, NULL, NULL }; - struct GNUNET_REGEX_State *s; - struct GNUNET_REGEX_State *s_next; - struct GNUNET_REGEX_Transition *t; - struct GNUNET_REGEX_Transition *t_next; + if (ret->blen < sarg1->slen + sarg2->slen + extra_chars + 1) + sb_realloc (ret, + sarg1->slen + sarg2->slen + extra_chars + 1); + ret->null_flag = GNUNET_NO; + ret->slen = sarg1->slen + sarg2->slen + extra_chars; + ret->sbuf = ret->abuf; + GNUNET_snprintf (ret->sbuf, + ret->blen, + format, + (int) sarg1->slen, + sarg1->sbuf, + (int) sarg2->slen, + sarg2->sbuf); +} - if (1 > stride_len || GNUNET_YES == dfa->is_multistrided) - return; - // Unmark all states - for (s = dfa->states_head; NULL != s; s = s->next) - s->marked = GNUNET_NO; +/** + * Format a string buffer. Note that optimizing this function is not + * really worth it, it is rarely called. + * + * @param ret where to write the result + * @param format format string, fprintf-style, with exactly three "%.*s" + * @param extra_chars how long will the result be, in addition to 'sarg1/2/3' length + * @param sarg1 first string to print into the format + * @param sarg2 second string to print into the format + * @param sarg3 third string to print into the format + */ +static void +sb_printf3 (struct StringBuffer *ret, + const char *format, + size_t extra_chars, + const struct StringBuffer *sarg1, + const struct StringBuffer *sarg2, + const struct StringBuffer *sarg3) +{ + if (ret->blen < sarg1->slen + sarg2->slen + sarg3->slen + extra_chars + 1) + sb_realloc (ret, + sarg1->slen + sarg2->slen + sarg3->slen + extra_chars + 1); + ret->null_flag = GNUNET_NO; + ret->slen = sarg1->slen + sarg2->slen + sarg3->slen + extra_chars; + ret->sbuf = ret->abuf; + GNUNET_snprintf (ret->sbuf, + ret->blen, + format, + (int) sarg1->slen, + sarg1->sbuf, + (int) sarg2->slen, + sarg2->sbuf, + (int) sarg3->slen, + sarg3->sbuf); +} - // Compute the new transitions of given stride_len - GNUNET_REGEX_automaton_traverse (dfa, dfa->start, NULL, NULL, - &add_multi_strides_to_dfa, &ctx); - // Add all the new transitions to the automaton. - for (t = ctx.transitions_head; NULL != t; t = t_next) +/** + * Free resources of the given string buffer. + * + * @param sb buffer to free (actual pointer is not freed, as they + * should not be individually allocated) + */ +static void +sb_free (struct StringBuffer *sb) +{ + GNUNET_array_grow (sb->abuf, + sb->blen, + 0); + sb->slen = 0; + sb->sbuf = NULL; + sb->null_flag= GNUNET_YES; +} + + +/** + * Copy the given string buffer from 'in' to 'out'. + * + * @param in input string + * @param out output string + */ +static void +sb_strdup (struct StringBuffer *out, + const struct StringBuffer *in) + +{ + out->null_flag = in->null_flag; + if (GNUNET_YES == out->null_flag) + return; + if (out->blen < in->slen) { - t_next = t->next; - state_add_transition (regex_ctx, t->from_state, t->label, t->to_state); - GNUNET_CONTAINER_DLL_remove (ctx.transitions_head, ctx.transitions_tail, t); - GNUNET_free_non_null (t->label); - GNUNET_free (t); + GNUNET_array_grow (out->abuf, + out->blen, + in->slen); } + out->sbuf = out->abuf; + out->slen = in->slen; + memcpy (out->sbuf, in->sbuf, out->slen); +} - // Remove marked states (including their incoming and outgoing transitions) - for (s = dfa->states_head; NULL != s; s = s_next) + +/** + * Copy the given string buffer from 'in' to 'out'. + * + * @param cstr input string + * @param out output string + */ +static void +sb_strdup_cstr (struct StringBuffer *out, + const char *cstr) +{ + if (NULL == cstr) { - s_next = s->next; - if (GNUNET_YES == s->marked) - automaton_remove_state (dfa, s); + out->null_flag = GNUNET_YES; + return; } - - // Mark this automaton as multistrided - dfa->is_multistrided = GNUNET_YES; + out->null_flag = GNUNET_NO; + out->slen = strlen (cstr); + if (out->blen < out->slen) + { + GNUNET_array_grow (out->abuf, + out->blen, + out->slen); + } + out->sbuf = out->abuf; + memcpy (out->sbuf, cstr, out->slen); } - /** * Check if the given string 'str' needs parentheses around it when * using it to generate a regex. @@ -691,35 +908,37 @@ GNUNET_REGEX_add_multi_strides_to_dfa (struct GNUNET_REGEX_Context *regex_ctx, * @return GNUNET_YES if parentheses are needed, GNUNET_NO otherwise */ static int -needs_parentheses (const char *str) +needs_parentheses (const struct StringBuffer *str) { size_t slen; const char *op; const char *cl; const char *pos; + const char *end; unsigned int cnt; - if ((NULL == str) || ((slen = strlen (str)) < 2)) + if ((GNUNET_YES == str->null_flag) || ((slen = str->slen) < 2)) return GNUNET_NO; - - if ('(' != str[0]) + pos = str->sbuf; + if ('(' != pos[0]) return GNUNET_YES; + end = str->sbuf + slen; cnt = 1; - pos = &str[1]; + pos++; while (cnt > 0) { - cl = strchr (pos, ')'); + cl = memchr (pos, ')', end - pos); if (NULL == cl) { GNUNET_break (0); return GNUNET_YES; } - op = strchr (pos, '('); - if ((NULL != op) && (op < cl)) + /* while '(' before ')', count opening parens */ + while ( (NULL != (op = memchr (pos, '(', end - pos))) && + (op < cl) ) { cnt++; pos = op + 1; - continue; } /* got ')' first */ cnt--; @@ -734,28 +953,61 @@ needs_parentheses (const char *str) * Example: "(a)" becomes "a", "(a|b)|(a|c)" stays the same. * You need to GNUNET_free the returned string. * - * @param str string, free'd or re-used by this function, can be NULL - * + * @param str string, modified to contain a * @return string without surrounding parentheses, string 'str' if no preceding * epsilon could be found, NULL if 'str' was NULL */ -static char * -remove_parentheses (char *str) +static void +remove_parentheses (struct StringBuffer *str) { size_t slen; const char *pos; + const char *end; + const char *sbuf; + const char *op; + const char *cp; + unsigned int cnt; - if ((NULL == str) || ('(' != str[0]) || - (str[(slen = strlen (str)) - 1] != ')')) - return str; - - pos = strchr (&str[1], ')'); - if (pos == &str[slen - 1]) + if (0) + return; + sbuf = str->sbuf; + if ( (GNUNET_YES == str->null_flag) || + (1 >= (slen = str->slen)) || + ('(' != str->sbuf[0]) || + (')' != str->sbuf[slen - 1]) ) + return; + cnt = 0; + pos = &sbuf[1]; + end = &sbuf[slen - 1]; + op = memchr (pos, '(', end - pos); + cp = memchr (pos, ')', end - pos); + while (NULL != cp) + { + while ( (NULL != op) && + (op < cp) ) + { + cnt++; + pos = op + 1; + op = memchr (pos, '(', end - pos); + } + while ( (NULL != cp) && + ( (NULL == op) || + (cp < op) ) ) + { + if (0 == cnt) + return; /* can't strip parens */ + cnt--; + pos = cp + 1; + cp = memchr (pos, ')', end - pos); + } + } + if (0 != cnt) { - memmove (str, &str[1], slen - 2); - str[slen - 2] = '\0'; + GNUNET_break (0); + return; } - return str; + str->sbuf++; + str->slen -= 2; } @@ -768,10 +1020,14 @@ remove_parentheses (char *str) * @return 0 if str has no epsilon, 1 if str starts with '(|' and ends with ')' */ static int -has_epsilon (const char *str) +has_epsilon (const struct StringBuffer *str) { - return (NULL != str) && ('(' == str[0]) && ('|' == str[1]) && - (')' == str[strlen (str) - 1]); + return + (GNUNET_YES != str->null_flag) && + (0 < str->slen) && + ('(' == str->sbuf[0]) && + ('|' == str->sbuf[1]) && + (')' == str->sbuf[str->slen - 1]); } @@ -780,57 +1036,136 @@ has_epsilon (const char *str) * Example: str = "(|a|b|c)", result: "a|b|c" * The returned string needs to be freed. * - * @param str string - * - * @return string without preceding epsilon, string 'str' if no preceding + * @param str original string + * @param ret where to return string without preceding epsilon, string 'str' if no preceding * epsilon could be found, NULL if 'str' was NULL */ -static char * -remove_epsilon (const char *str) +static void +remove_epsilon (const struct StringBuffer *str, + struct StringBuffer *ret) { - size_t len; - - if (NULL == str) - return NULL; - if (('(' == str[0]) && ('|' == str[1])) + if (GNUNET_YES == str->null_flag) { - len = strlen (str); - if (')' == str[len - 1]) - return GNUNET_strndup (&str[2], len - 3); + ret->null_flag = GNUNET_YES; + return; + } + if ( (str->slen > 1) && + ('(' == str->sbuf[0]) && + ('|' == str->sbuf[1]) && + (')' == str->sbuf[str->slen - 1]) ) + { + /* remove epsilon */ + if (ret->blen < str->slen - 3) + { + GNUNET_array_grow (ret->abuf, + ret->blen, + str->slen - 3); + } + ret->sbuf = ret->abuf; + ret->slen = str->slen - 3; + memcpy (ret->sbuf, &str->sbuf[2], ret->slen); + return; } - return GNUNET_strdup (str); + sb_strdup (ret, str); } /** - * Compare 'str1', starting from position 'k', with whole 'str2' + * Compare n bytes of 'str1' and 'str2' * - * @param str1 first string to compare, starting from position 'k' + * @param str1 first string to compare * @param str2 second string for comparison - * @param k starting position in 'str1' + * @param n number of bytes to compare * * @return -1 if any of the strings is NULL, 0 if equal, non 0 otherwise */ static int -strkcmp (const char *str1, const char *str2, size_t k) +sb_strncmp (const struct StringBuffer *str1, + const struct StringBuffer *str2, size_t n) { - if ((NULL == str1) || (NULL == str2) || (strlen (str1) < k)) + size_t max; + + if ( (str1->slen != str2->slen) && + ( (str1->slen < n) || + (str2->slen < n) ) ) return -1; - return strcmp (&str1[k], str2); + max = GNUNET_MAX (str1->slen, str2->slen); + if (max > n) + max = n; + return memcmp (str1->sbuf, str2->sbuf, max); } /** - * Helper function used as 'action' in 'GNUNET_REGEX_automaton_traverse' - * function to create the depth-first numbering of the states. + * Compare n bytes of 'str1' and 'str2' * - * @param cls states array. - * @param count current state counter. - * @param s current state. + * @param str1 first string to compare + * @param str2 second C string for comparison + * @param n number of bytes to compare (and length of str2) + * + * @return -1 if any of the strings is NULL, 0 if equal, non 0 otherwise */ -void -number_states (void *cls, const unsigned int count, - struct GNUNET_REGEX_State *s) +static int +sb_strncmp_cstr (const struct StringBuffer *str1, + const char *str2, size_t n) +{ + if (str1->slen < n) + return -1; + return memcmp (str1->sbuf, str2, n); +} + + +/** + * Initialize string buffer for storing strings of up to n + * characters. + * + * @param sb buffer to initialize + * @param n desired target length + */ +static void +sb_init (struct StringBuffer *sb, + size_t n) +{ + sb->null_flag = GNUNET_NO; + sb->abuf = sb->sbuf = (0 == n) ? NULL : GNUNET_malloc (n); + sb->blen = n; + sb->slen = 0; +} + + +/** + * Compare 'str1', starting from position 'k', with whole 'str2' + * + * @param str1 first string to compare, starting from position 'k' + * @param str2 second string for comparison + * @param k starting position in 'str1' + * + * @return -1 if any of the strings is NULL, 0 if equal, non 0 otherwise + */ +static int +sb_strkcmp (const struct StringBuffer *str1, + const struct StringBuffer *str2, size_t k) +{ + if ( (GNUNET_YES == str1->null_flag) || + (GNUNET_YES == str2->null_flag) || + (k > str1->slen) || + (str1->slen - k != str2->slen) ) + return -1; + return memcmp (&str1->sbuf[k], str2->sbuf, str2->slen); +} + + +/** + * Helper function used as 'action' in 'GNUNET_REGEX_automaton_traverse' + * function to create the depth-first numbering of the states. + * + * @param cls states array. + * @param count current state counter. + * @param s current state. + */ +static void +number_states (void *cls, const unsigned int count, + struct GNUNET_REGEX_State *s) { struct GNUNET_REGEX_State **states = cls; @@ -840,6 +1175,12 @@ number_states (void *cls, const unsigned int count, } + +#define PRIS(a) \ + ((GNUNET_YES == a.null_flag) ? 6 : (int) a.slen), \ + ((GNUNET_YES == a.null_flag) ? "(null)" : a.sbuf) + + /** * Construct the regular expression given the inductive step, * $R^{(k)}_{ij} = R^{(k-1)}_{ij} | R^{(k-1)}_{ik} ( R^{(k-1)}_{kk} )^* @@ -851,511 +1192,565 @@ number_states (void *cls, const unsigned int count, * @param R_last_kj value of $R^{(k-1)_{kj}. * @param R_cur_ij result for this inductive step is saved in R_cur_ij, R_cur_ij * is expected to be NULL when called! + * @param R_cur_l optimization -- kept between iterations to avoid realloc + * @param R_cur_r optimization -- kept between iterations to avoid realloc */ static void -automaton_create_proofs_simplify (char *R_last_ij, char *R_last_ik, - char *R_last_kk, char *R_last_kj, - char **R_cur_ij) -{ - char *R_cur_l; - char *R_cur_r; - char *temp_a; - char *temp_b; - char *R_temp_ij; - char *R_temp_ik; - char *R_temp_kj; - char *R_temp_kk; - +automaton_create_proofs_simplify (const struct StringBuffer *R_last_ij, + const struct StringBuffer *R_last_ik, + const struct StringBuffer *R_last_kk, + const struct StringBuffer *R_last_kj, + struct StringBuffer *R_cur_ij, + struct StringBuffer *R_cur_l, + struct StringBuffer *R_cur_r) +{ + struct StringBuffer R_temp_ij; + struct StringBuffer R_temp_ik; + struct StringBuffer R_temp_kj; + struct StringBuffer R_temp_kk; int eps_check; int ij_ik_cmp; int ij_kj_cmp; - int ik_kk_cmp; int kk_kj_cmp; int clean_ik_kk_cmp; int clean_kk_kj_cmp; - unsigned int cnt; - size_t length; size_t length_l; size_t length_r; - GNUNET_assert (NULL == *R_cur_ij && NULL != R_cur_ij); - - // $R^{(k)}_{ij} = R^{(k-1)}_{ij} | R^{(k-1)}_{ik} ( R^{(k-1)}_{kk} )^* R^{(k-1)}_{kj} - // R_last == R^{(k-1)}, R_cur == R^{(k)} - // R_cur_ij = R_cur_l | R_cur_r - // R_cur_l == R^{(k-1)}_{ij} - // R_cur_r == R^{(k-1)}_{ik} ( R^{(k-1)}_{kk} )^* R^{(k-1)}_{kj} + /* + * $R^{(k)}_{ij} = R^{(k-1)}_{ij} | R^{(k-1)}_{ik} ( R^{(k-1)}_{kk} )^* R^{(k-1)}_{kj} + * R_last == R^{(k-1)}, R_cur == R^{(k)} + * R_cur_ij = R_cur_l | R_cur_r + * R_cur_l == R^{(k-1)}_{ij} + * R_cur_r == R^{(k-1)}_{ik} ( R^{(k-1)}_{kk} )^* R^{(k-1)}_{kj} + */ - if ((NULL == R_last_ij) && ((NULL == R_last_ik) || (NULL == R_last_kk) || /* technically cannot happen, but looks saner */ - (NULL == R_last_kj))) + if ( (GNUNET_YES == R_last_ij->null_flag) && + ( (GNUNET_YES == R_last_ik->null_flag) || + (GNUNET_YES == R_last_kj->null_flag))) { /* R^{(k)}_{ij} = N | N */ - *R_cur_ij = NULL; + R_cur_ij->null_flag = GNUNET_YES; + R_cur_ij->synced = GNUNET_NO; return; } - if ((NULL == R_last_ik) || (NULL == R_last_kk) || /* technically cannot happen, but looks saner */ - (NULL == R_last_kj)) + if ( (GNUNET_YES == R_last_ik->null_flag) || + (GNUNET_YES == R_last_kj->null_flag) ) { /* R^{(k)}_{ij} = R^{(k-1)}_{ij} | N */ - *R_cur_ij = GNUNET_strdup (R_last_ij); + if (GNUNET_YES == R_last_ij->synced) + { + R_cur_ij->synced = GNUNET_YES; + R_cur_ij->null_flag = GNUNET_NO; + return; + } + R_cur_ij->synced = GNUNET_YES; + sb_strdup (R_cur_ij, R_last_ij); return; } - - // $R^{(k)}_{ij} = N | R^{(k-1)}_{ik} ( R^{(k-1)}_{kk} )^* R^{(k-1)}_{kj} OR - // $R^{(k)}_{ij} = R^{(k-1)}_{ij} | R^{(k-1)}_{ik} ( R^{(k-1)}_{kk} )^* R^{(k-1)}_{kj} - - R_cur_r = NULL; - R_cur_l = NULL; - - // cache results from strcmp, we might need these many times - ij_kj_cmp = nullstrcmp (R_last_ij, R_last_kj); - ij_ik_cmp = nullstrcmp (R_last_ij, R_last_ik); - ik_kk_cmp = nullstrcmp (R_last_ik, R_last_kk); - kk_kj_cmp = nullstrcmp (R_last_kk, R_last_kj); - - // Assign R_temp_(ik|kk|kj) to R_last[][] and remove epsilon as well - // as parentheses, so we can better compare the contents - R_temp_ik = remove_parentheses (remove_epsilon (R_last_ik)); - R_temp_kk = remove_parentheses (remove_epsilon (R_last_kk)); - R_temp_kj = remove_parentheses (remove_epsilon (R_last_kj)); - - clean_ik_kk_cmp = nullstrcmp (R_last_ik, R_temp_kk); - clean_kk_kj_cmp = nullstrcmp (R_temp_kk, R_last_kj); - - // construct R_cur_l (and, if necessary R_cur_r) - if (NULL != R_last_ij) - { - // Assign R_temp_ij to R_last_ij and remove epsilon as well - // as parentheses, so we can better compare the contents - R_temp_ij = remove_parentheses (remove_epsilon (R_last_ij)); - - if (0 == strcmp (R_temp_ij, R_temp_ik) && 0 == strcmp (R_temp_ik, R_temp_kk) - && 0 == strcmp (R_temp_kk, R_temp_kj)) + R_cur_ij->synced = GNUNET_NO; + + /* $R^{(k)}_{ij} = N | R^{(k-1)}_{ik} ( R^{(k-1)}_{kk} )^* R^{(k-1)}_{kj} OR + * $R^{(k)}_{ij} = R^{(k-1)}_{ij} | R^{(k-1)}_{ik} ( R^{(k-1)}_{kk} )^* R^{(k-1)}_{kj} */ + + R_cur_r->null_flag = GNUNET_YES; + R_cur_r->slen = 0; + R_cur_l->null_flag = GNUNET_YES; + R_cur_l->slen = 0; + + /* cache results from strcmp, we might need these many times */ + ij_kj_cmp = sb_nullstrcmp (R_last_ij, R_last_kj); + ij_ik_cmp = sb_nullstrcmp (R_last_ij, R_last_ik); + ik_kk_cmp = sb_nullstrcmp (R_last_ik, R_last_kk); + kk_kj_cmp = sb_nullstrcmp (R_last_kk, R_last_kj); + + /* Assign R_temp_(ik|kk|kj) to R_last[][] and remove epsilon as well + * as parentheses, so we can better compare the contents */ + + memset (&R_temp_ij, 0, sizeof (struct StringBuffer)); + memset (&R_temp_ik, 0, sizeof (struct StringBuffer)); + memset (&R_temp_kk, 0, sizeof (struct StringBuffer)); + memset (&R_temp_kj, 0, sizeof (struct StringBuffer)); + remove_epsilon (R_last_ik, &R_temp_ik); + remove_epsilon (R_last_kk, &R_temp_kk); + remove_epsilon (R_last_kj, &R_temp_kj); + remove_parentheses (&R_temp_ik); + remove_parentheses (&R_temp_kk); + remove_parentheses (&R_temp_kj); + clean_ik_kk_cmp = sb_nullstrcmp (R_last_ik, &R_temp_kk); + clean_kk_kj_cmp = sb_nullstrcmp (&R_temp_kk, R_last_kj); + + /* construct R_cur_l (and, if necessary R_cur_r) */ + if (GNUNET_YES != R_last_ij->null_flag) + { + /* Assign R_temp_ij to R_last_ij and remove epsilon as well + * as parentheses, so we can better compare the contents */ + remove_epsilon (R_last_ij, &R_temp_ij); + remove_parentheses (&R_temp_ij); + + if ( (0 == sb_strcmp (&R_temp_ij, &R_temp_ik)) && + (0 == sb_strcmp (&R_temp_ik, &R_temp_kk)) && + (0 == sb_strcmp (&R_temp_kk, &R_temp_kj)) ) { - if (0 == strlen (R_temp_ij)) + if (0 == R_temp_ij.slen) { - R_cur_r = GNUNET_strdup (""); + R_cur_r->null_flag = GNUNET_NO; } - else if ((0 == strncmp (R_last_ij, "(|", 2)) || - (0 == strncmp (R_last_ik, "(|", 2) && - 0 == strncmp (R_last_kj, "(|", 2))) + else if ((0 == sb_strncmp_cstr (R_last_ij, "(|", 2)) || + (0 == sb_strncmp_cstr (R_last_ik, "(|", 2) && + 0 == sb_strncmp_cstr (R_last_kj, "(|", 2))) { - // a|(e|a)a*(e|a) = a* - // a|(e|a)(e|a)*(e|a) = a* - // (e|a)|aa*a = a* - // (e|a)|aa*(e|a) = a* - // (e|a)|(e|a)a*a = a* - // (e|a)|(e|a)a*(e|a) = a* - // (e|a)|(e|a)(e|a)*(e|a) = a* - if (GNUNET_YES == needs_parentheses (R_temp_ij)) - GNUNET_asprintf (&R_cur_r, "(%s)*", R_temp_ij); + /* + * a|(e|a)a*(e|a) = a* + * a|(e|a)(e|a)*(e|a) = a* + * (e|a)|aa*a = a* + * (e|a)|aa*(e|a) = a* + * (e|a)|(e|a)a*a = a* + * (e|a)|(e|a)a*(e|a) = a* + * (e|a)|(e|a)(e|a)*(e|a) = a* + */ + if (GNUNET_YES == needs_parentheses (&R_temp_ij)) + sb_printf1 (R_cur_r, "(%.*s)*", 3, &R_temp_ij); else - GNUNET_asprintf (&R_cur_r, "%s*", R_temp_ij); + sb_printf1 (R_cur_r, "%.*s*", 1, &R_temp_ij); } else { - // a|aa*a = a+ - // a|(e|a)a*a = a+ - // a|aa*(e|a) = a+ - // a|(e|a)(e|a)*a = a+ - // a|a(e|a)*(e|a) = a+ - if (GNUNET_YES == needs_parentheses (R_temp_ij)) - GNUNET_asprintf (&R_cur_r, "(%s)+", R_temp_ij); + /* + * a|aa*a = a+ + * a|(e|a)a*a = a+ + * a|aa*(e|a) = a+ + * a|(e|a)(e|a)*a = a+ + * a|a(e|a)*(e|a) = a+ + */ + if (GNUNET_YES == needs_parentheses (&R_temp_ij)) + sb_printf1 (R_cur_r, "(%.*s)+", 3, &R_temp_ij); else - GNUNET_asprintf (&R_cur_r, "%s+", R_temp_ij); + sb_printf1 (R_cur_r, "%.*s+", 1, &R_temp_ij); } } - else if (0 == ij_ik_cmp && 0 == clean_kk_kj_cmp && 0 != clean_ik_kk_cmp) + else if ( (0 == ij_ik_cmp) && (0 == clean_kk_kj_cmp) && (0 != clean_ik_kk_cmp) ) { - // a|ab*b = ab* - if (strlen (R_last_kk) < 1) - R_cur_r = GNUNET_strdup (R_last_ij); - else if (GNUNET_YES == needs_parentheses (R_temp_kk)) - GNUNET_asprintf (&R_cur_r, "%s(%s)*", R_last_ij, R_temp_kk); + /* a|ab*b = ab* */ + if (0 == R_last_kk->slen) + sb_strdup (R_cur_r, R_last_ij); + else if (GNUNET_YES == needs_parentheses (&R_temp_kk)) + sb_printf2 (R_cur_r, "%.*s(%.*s)*", 3, R_last_ij, &R_temp_kk); else - GNUNET_asprintf (&R_cur_r, "%s%s*", R_last_ij, R_last_kk); - - R_cur_l = NULL; + sb_printf2 (R_cur_r, "%.*s%.*s*", 1, R_last_ij, R_last_kk); + R_cur_l->null_flag = GNUNET_YES; } - else if (0 == ij_kj_cmp && 0 == clean_ik_kk_cmp && 0 != clean_kk_kj_cmp) + else if ( (0 == ij_kj_cmp) && (0 == clean_ik_kk_cmp) && (0 != clean_kk_kj_cmp)) { - // a|bb*a = b*a - if (strlen (R_last_kk) < 1) - R_cur_r = GNUNET_strdup (R_last_kj); - else if (GNUNET_YES == needs_parentheses (R_temp_kk)) - GNUNET_asprintf (&R_cur_r, "(%s)*%s", R_temp_kk, R_last_kj); + /* a|bb*a = b*a */ + if (R_last_kk->slen < 1) + { + sb_strdup (R_cur_r, R_last_kj); + } + else if (GNUNET_YES == needs_parentheses (&R_temp_kk)) + sb_printf2 (R_cur_r, "(%.*s)*%.*s", 3, &R_temp_kk, R_last_kj); else - GNUNET_asprintf (&R_cur_r, "%s*%s", R_temp_kk, R_last_kj); + sb_printf2 (R_cur_r, "%.*s*%.*s", 1, &R_temp_kk, R_last_kj); - R_cur_l = NULL; + R_cur_l->null_flag = GNUNET_YES; } - else if (0 == ij_ik_cmp && 0 == kk_kj_cmp && !has_epsilon (R_last_ij) && - has_epsilon (R_last_kk)) + else if ( (0 == ij_ik_cmp) && (0 == kk_kj_cmp) && (! has_epsilon (R_last_ij)) && + has_epsilon (R_last_kk)) { - // a|a(e|b)*(e|b) = a|ab* = a|a|ab|abb|abbb|... = ab* - if (needs_parentheses (R_temp_kk)) - GNUNET_asprintf (&R_cur_r, "%s(%s)*", R_last_ij, R_temp_kk); + /* a|a(e|b)*(e|b) = a|ab* = a|a|ab|abb|abbb|... = ab* */ + if (needs_parentheses (&R_temp_kk)) + sb_printf2 (R_cur_r, "%.*s(%.*s)*", 3, R_last_ij, &R_temp_kk); else - GNUNET_asprintf (&R_cur_r, "%s%s*", R_last_ij, R_temp_kk); - - R_cur_l = NULL; + sb_printf2 (R_cur_r, "%.*s%.*s*", 1, R_last_ij, &R_temp_kk); + R_cur_l->null_flag = GNUNET_YES; } - else if (0 == ij_kj_cmp && 0 == ik_kk_cmp && !has_epsilon (R_last_ij) && + else if ( (0 == ij_kj_cmp) && (0 == ik_kk_cmp) && (! has_epsilon (R_last_ij)) && has_epsilon (R_last_kk)) { - // a|(e|b)(e|b)*a = a|b*a = a|a|ba|bba|bbba|... = b*a - if (needs_parentheses (R_temp_kk)) - GNUNET_asprintf (&R_cur_r, "(%s)*%s", R_temp_kk, R_last_ij); + /* a|(e|b)(e|b)*a = a|b*a = a|a|ba|bba|bbba|... = b*a */ + if (needs_parentheses (&R_temp_kk)) + sb_printf2 (R_cur_r, "(%.*s)*%.*s", 3, &R_temp_kk, R_last_ij); else - GNUNET_asprintf (&R_cur_r, "%s*%s", R_temp_kk, R_last_ij); - - R_cur_l = NULL; + sb_printf2 (R_cur_r, "%.*s*%.*s", 1, &R_temp_kk, R_last_ij); + R_cur_l->null_flag = GNUNET_YES; } else { - temp_a = (NULL == R_last_ij) ? NULL : GNUNET_strdup (R_last_ij); - temp_a = remove_parentheses (temp_a); - R_cur_l = temp_a; + sb_strdup (R_cur_l, R_last_ij); + remove_parentheses (R_cur_l); } - - GNUNET_free_non_null (R_temp_ij); } else { - // we have no left side - R_cur_l = NULL; - } - - // construct R_cur_r, if not already constructed - if (NULL == R_cur_r) - { - length = strlen (R_temp_kk) - strlen (R_last_ik); - - // a(ba)*bx = (ab)+x - if (length > 0 && NULL != R_last_kk && 0 < strlen (R_last_kk) && - NULL != R_last_kj && 0 < strlen (R_last_kj) && NULL != R_last_ik && - 0 < strlen (R_last_ik) && 0 == strkcmp (R_temp_kk, R_last_ik, length) && - 0 == strncmp (R_temp_kk, R_last_kj, length)) - { - temp_a = GNUNET_malloc (length + 1); - temp_b = GNUNET_malloc ((strlen (R_last_kj) - length) + 1); - - length_l = 0; - length_r = 0; - - for (cnt = 0; cnt < strlen (R_last_kj); cnt++) - { - if (cnt < length) - { - temp_a[length_l] = R_last_kj[cnt]; - length_l++; - } - else - { - temp_b[length_r] = R_last_kj[cnt]; - length_r++; - } - } - temp_a[length_l] = '\0'; - temp_b[length_r] = '\0'; - - // e|(ab)+ = (ab)* - if (NULL != R_cur_l && 0 == strlen (R_cur_l) && 0 == strlen (temp_b)) + /* we have no left side */ + R_cur_l->null_flag = GNUNET_YES; + } + + /* construct R_cur_r, if not already constructed */ + if (GNUNET_YES == R_cur_r->null_flag) + { + length = R_temp_kk.slen - R_last_ik->slen; + + /* a(ba)*bx = (ab)+x */ + if ( (length > 0) && + (GNUNET_YES != R_last_kk->null_flag) && + (0 < R_last_kk->slen) && + (GNUNET_YES != R_last_kj->null_flag) && + (0 < R_last_kj->slen) && + (GNUNET_YES != R_last_ik->null_flag) && + (0 < R_last_ik->slen) && + (0 == sb_strkcmp (&R_temp_kk, R_last_ik, length)) && + (0 == sb_strncmp (&R_temp_kk, R_last_kj, length)) ) + { + struct StringBuffer temp_a; + struct StringBuffer temp_b; + + sb_init (&temp_a, length); + sb_init (&temp_b, R_last_kj->slen - length); + + length_l = length; + temp_a.sbuf = temp_a.abuf; + memcpy (temp_a.sbuf, R_last_kj->sbuf, length_l); + temp_a.slen = length_l; + + length_r = R_last_kj->slen - length; + temp_b.sbuf = temp_b.abuf; + memcpy (temp_b.sbuf, &R_last_kj->sbuf[length], length_r); + temp_b.slen = length_r; + + /* e|(ab)+ = (ab)* */ + if ( (GNUNET_YES != R_cur_l->null_flag) && + (0 == R_cur_l->slen) && + (0 == temp_b.slen) ) { - GNUNET_asprintf (&R_cur_r, "(%s%s)*", R_last_ik, temp_a); - GNUNET_free (R_cur_l); - R_cur_l = NULL; + sb_printf2 (R_cur_r, "(%.*s%.*s)*", 3, R_last_ik, &temp_a); + sb_free (R_cur_l); + R_cur_l->null_flag = GNUNET_YES; } else { - GNUNET_asprintf (&R_cur_r, "(%s%s)+%s", R_last_ik, temp_a, temp_b); + sb_printf3 (R_cur_r, "(%.*s%.*s)+%.*s", 3, R_last_ik, &temp_a, &temp_b); } - GNUNET_free (temp_a); - GNUNET_free (temp_b); + sb_free (&temp_a); + sb_free (&temp_b); } - else if (0 == strcmp (R_temp_ik, R_temp_kk) && - 0 == strcmp (R_temp_kk, R_temp_kj)) + else if (0 == sb_strcmp (&R_temp_ik, &R_temp_kk) && + 0 == sb_strcmp (&R_temp_kk, &R_temp_kj)) { - // (e|a)a*(e|a) = a* - // (e|a)(e|a)*(e|a) = a* + /* + * (e|a)a*(e|a) = a* + * (e|a)(e|a)*(e|a) = a* + */ if (has_epsilon (R_last_ik) && has_epsilon (R_last_kj)) { - if (needs_parentheses (R_temp_kk)) - GNUNET_asprintf (&R_cur_r, "(%s)*", R_temp_kk); + if (needs_parentheses (&R_temp_kk)) + sb_printf1 (R_cur_r, "(%.*s)*", 3, &R_temp_kk); else - GNUNET_asprintf (&R_cur_r, "%s*", R_temp_kk); + sb_printf1 (R_cur_r, "%.*s*", 1, &R_temp_kk); } - // aa*a = a+a - else if (0 == clean_ik_kk_cmp && 0 == clean_kk_kj_cmp && - !has_epsilon (R_last_ik)) + /* aa*a = a+a */ + else if ( (0 == clean_ik_kk_cmp) && + (0 == clean_kk_kj_cmp) && + (! has_epsilon (R_last_ik)) ) { - if (needs_parentheses (R_temp_kk)) - GNUNET_asprintf (&R_cur_r, "(%s)+%s", R_temp_kk, R_temp_kk); + if (needs_parentheses (&R_temp_kk)) + sb_printf2 (R_cur_r, "(%.*s)+%.*s", 3, &R_temp_kk, &R_temp_kk); else - GNUNET_asprintf (&R_cur_r, "(%s)+%s", R_temp_kk, R_temp_kk); + sb_printf2 (R_cur_r, "%.*s+%.*s", 1, &R_temp_kk, &R_temp_kk); } - // (e|a)a*a = a+ - // aa*(e|a) = a+ - // a(e|a)*(e|a) = a+ - // (e|a)a*a = a+ + /* + * (e|a)a*a = a+ + * aa*(e|a) = a+ + * a(e|a)*(e|a) = a+ + * (e|a)a*a = a+ + */ else { eps_check = - (has_epsilon (R_last_ik) + has_epsilon (R_last_kk) + - has_epsilon (R_last_kj)); + (has_epsilon (R_last_ik) + has_epsilon (R_last_kk) + + has_epsilon (R_last_kj)); - if (eps_check == 1) + if (1 == eps_check) { - if (needs_parentheses (R_temp_kk)) - GNUNET_asprintf (&R_cur_r, "(%s)+", R_temp_kk); + if (needs_parentheses (&R_temp_kk)) + sb_printf1 (R_cur_r, "(%.*s)+", 3, &R_temp_kk); else - GNUNET_asprintf (&R_cur_r, "%s+", R_temp_kk); + sb_printf1 (R_cur_r, "%.*s+", 1, &R_temp_kk); } } } - // aa*b = a+b - // (e|a)(e|a)*b = a*b - else if (0 == strcmp (R_temp_ik, R_temp_kk)) + /* + * aa*b = a+b + * (e|a)(e|a)*b = a*b + */ + else if (0 == sb_strcmp (&R_temp_ik, &R_temp_kk)) { if (has_epsilon (R_last_ik)) { - if (needs_parentheses (R_temp_kk)) - GNUNET_asprintf (&R_cur_r, "(%s)*%s", R_temp_kk, R_last_kj); + if (needs_parentheses (&R_temp_kk)) + sb_printf2 (R_cur_r, "(%.*s)*%.*s", 3, &R_temp_kk, R_last_kj); else - GNUNET_asprintf (&R_cur_r, "%s*%s", R_temp_kk, R_last_kj); + sb_printf2 (R_cur_r, "%.*s*%.*s", 1, &R_temp_kk, R_last_kj); } else { - if (needs_parentheses (R_temp_kk)) - GNUNET_asprintf (&R_cur_r, "(%s)+%s", R_temp_kk, R_last_kj); + if (needs_parentheses (&R_temp_kk)) + sb_printf2 (R_cur_r, "(%.*s)+%.*s", 3, &R_temp_kk, R_last_kj); else - GNUNET_asprintf (&R_cur_r, "%s+%s", R_temp_kk, R_last_kj); + sb_printf2 (R_cur_r, "%.*s+%.*s", 1, &R_temp_kk, R_last_kj); } } - // ba*a = ba+ - // b(e|a)*(e|a) = ba* - else if (0 == strcmp (R_temp_kk, R_temp_kj)) + /* + * ba*a = ba+ + * b(e|a)*(e|a) = ba* + */ + else if (0 == sb_strcmp (&R_temp_kk, &R_temp_kj)) { if (has_epsilon (R_last_kj)) { - if (needs_parentheses (R_temp_kk)) - GNUNET_asprintf (&R_cur_r, "%s(%s)*", R_last_ik, R_temp_kk); + if (needs_parentheses (&R_temp_kk)) + sb_printf2 (R_cur_r, "%.*s(%.*s)*", 3, R_last_ik, &R_temp_kk); else - GNUNET_asprintf (&R_cur_r, "%s%s*", R_last_ik, R_temp_kk); + sb_printf2 (R_cur_r, "%.*s%.*s*", 1, R_last_ik, &R_temp_kk); } else { - if (needs_parentheses (R_temp_kk)) - GNUNET_asprintf (&R_cur_r, "(%s)+%s", R_last_ik, R_temp_kk); + if (needs_parentheses (&R_temp_kk)) + sb_printf2 (R_cur_r, "(%.*s)+%.*s", 3, R_last_ik, &R_temp_kk); else - GNUNET_asprintf (&R_cur_r, "%s+%s", R_last_ik, R_temp_kk); + sb_printf2 (R_cur_r, "%.*s+%.*s", 1, R_last_ik, &R_temp_kk); } } else { - if (strlen (R_temp_kk) > 0) + if (0 < R_temp_kk.slen) { - if (needs_parentheses (R_temp_kk)) + if (needs_parentheses (&R_temp_kk)) { - GNUNET_asprintf (&R_cur_r, "%s(%s)*%s", R_last_ik, R_temp_kk, - R_last_kj); + sb_printf3 (R_cur_r, "%.*s(%.*s)*%.*s", 3, R_last_ik, &R_temp_kk, + R_last_kj); } else { - GNUNET_asprintf (&R_cur_r, "%s%s*%s", R_last_ik, R_temp_kk, - R_last_kj); + sb_printf3 (R_cur_r, "%.*s%.*s*%.*s", 1, R_last_ik, &R_temp_kk, + R_last_kj); } } else { - GNUNET_asprintf (&R_cur_r, "%s%s", R_last_ik, R_last_kj); + sb_printf2 (R_cur_r, "%.*s%.*s", 0, R_last_ik, R_last_kj); } } } + sb_free (&R_temp_ij); + sb_free (&R_temp_ik); + sb_free (&R_temp_kk); + sb_free (&R_temp_kj); - GNUNET_free_non_null (R_temp_ik); - GNUNET_free_non_null (R_temp_kk); - GNUNET_free_non_null (R_temp_kj); - - if (NULL == R_cur_l && NULL == R_cur_r) + if ( (GNUNET_YES == R_cur_l->null_flag) && + (GNUNET_YES == R_cur_r->null_flag) ) { - *R_cur_ij = NULL; + R_cur_ij->null_flag = GNUNET_YES; return; } - if (NULL != R_cur_l && NULL == R_cur_r) + if ( (GNUNET_YES != R_cur_l->null_flag) && + (GNUNET_YES == R_cur_r->null_flag) ) { - *R_cur_ij = R_cur_l; + struct StringBuffer tmp; + + tmp = *R_cur_ij; + *R_cur_ij = *R_cur_l; + *R_cur_l = tmp; return; } - if (NULL == R_cur_l && NULL != R_cur_r) + if ( (GNUNET_YES == R_cur_l->null_flag) && + (GNUNET_YES != R_cur_r->null_flag) ) { - *R_cur_ij = R_cur_r; + struct StringBuffer tmp; + + tmp = *R_cur_ij; + *R_cur_ij = *R_cur_r; + *R_cur_r = tmp; return; } - if (0 == nullstrcmp (R_cur_l, R_cur_r)) + if (0 == sb_nullstrcmp (R_cur_l, R_cur_r)) { - *R_cur_ij = R_cur_l; - GNUNET_free (R_cur_r); + struct StringBuffer tmp; + + tmp = *R_cur_ij; + *R_cur_ij = *R_cur_l; + *R_cur_l = tmp; return; } - - GNUNET_asprintf (R_cur_ij, "(%s|%s)", R_cur_l, R_cur_r); - - GNUNET_free (R_cur_l); - GNUNET_free (R_cur_r); + sb_printf2 (R_cur_ij, "(%.*s|%.*s)", 3, R_cur_l, R_cur_r); } /** - * create proofs for all states in the given automaton. Implementation of the + * Create proofs for all states in the given automaton. Implementation of the * algorithm descriped in chapter 3.2.1 of "Automata Theory, Languages, and * Computation 3rd Edition" by Hopcroft, Motwani and Ullman. * - * @param a automaton. + * Each state in the automaton gets assigned 'proof' and 'hash' (hash of the + * proof) fields. The starting state will only have a valid proof/hash if it has + * any incoming transitions. + * + * @param a automaton for which to assign proofs and hashes, must not be NULL */ -static void +static int automaton_create_proofs (struct GNUNET_REGEX_Automaton *a) { - if (NULL == a) - { - GNUNET_log (GNUNET_ERROR_TYPE_ERROR, - "Could not create proofs, automaton was NULL\n"); - return; - } - unsigned int n = a->state_count; struct GNUNET_REGEX_State *states[n]; - char *R_last[n][n]; - char *R_cur[n][n]; - char *temp; + struct StringBuffer *R_last; + struct StringBuffer *R_cur; + struct StringBuffer R_cur_r; + struct StringBuffer R_cur_l; + struct StringBuffer *R_swap; struct GNUNET_REGEX_Transition *t; - char *complete_regex; + struct StringBuffer complete_regex; unsigned int i; unsigned int j; unsigned int k; + R_last = GNUNET_malloc_large (sizeof (struct StringBuffer) * n * n); + R_cur = GNUNET_malloc_large (sizeof (struct StringBuffer) * n * n); + if ( (NULL == R_last) || + (NULL == R_cur) ) + { + GNUNET_log_strerror (GNUNET_ERROR_TYPE_ERROR, "malloc"); + GNUNET_free_non_null (R_cur); + GNUNET_free_non_null (R_last); + return GNUNET_SYSERR; + } + /* create depth-first numbering of the states, initializes 'state' */ GNUNET_REGEX_automaton_traverse (a, a->start, NULL, NULL, &number_states, states); for (i = 0; i < n; i++) GNUNET_assert (NULL != states[i]); + for (i = 0; i < n; i++) + for (j = 0; j < n; j++) + R_last[i *n + j].null_flag = GNUNET_YES; /* Compute regular expressions of length "1" between each pair of states */ for (i = 0; i < n; i++) { - for (j = 0; j < n; j++) - { - R_cur[i][j] = NULL; - R_last[i][j] = NULL; - } for (t = states[i]->transitions_head; NULL != t; t = t->next) { j = t->to_state->dfs_id; - if (NULL == R_last[i][j]) - GNUNET_asprintf (&R_last[i][j], "%s", t->label); + if (GNUNET_YES == R_last[i * n + j].null_flag) + { + sb_strdup_cstr (&R_last[i * n + j], t->label); + } else { - temp = R_last[i][j]; - GNUNET_asprintf (&R_last[i][j], "%s|%s", R_last[i][j], t->label); - GNUNET_free (temp); + sb_append_cstr (&R_last[i * n + j], "|"); + sb_append_cstr (&R_last[i * n + j], t->label); } } - if (NULL == R_last[i][i]) - GNUNET_asprintf (&R_last[i][i], ""); + /* add self-loop: i is reachable from i via epsilon-transition */ + if (GNUNET_YES == R_last[i * n + i].null_flag) + { + R_last[i * n + i].slen = 0; + R_last[i * n + i].null_flag = GNUNET_NO; + } else { - temp = R_last[i][i]; - GNUNET_asprintf (&R_last[i][i], "(|%s)", R_last[i][i]); - GNUNET_free (temp); + sb_wrap (&R_last[i * n + i], "(|%.*s)", 3); } } for (i = 0; i < n; i++) for (j = 0; j < n; j++) - if (needs_parentheses (R_last[i][j])) - { - temp = R_last[i][j]; - GNUNET_asprintf (&R_last[i][j], "(%s)", R_last[i][j]); - GNUNET_free (temp); - } - + if (needs_parentheses (&R_last[i * n + j])) + sb_wrap (&R_last[i * n + j], "(%.*s)", 2); /* Compute regular expressions of length "k" between each pair of states per * induction */ + memset (&R_cur_l, 0, sizeof (struct StringBuffer)); + memset (&R_cur_r, 0, sizeof (struct StringBuffer)); for (k = 0; k < n; k++) { for (i = 0; i < n; i++) { for (j = 0; j < n; j++) { - // Basis for the recursion: - // $R^{(k)}_{ij} = R^{(k-1)}_{ij} | R^{(k-1)}_{ik} ( R^{(k-1)}_{kk} )^* R^{(k-1)}_{kj} - // R_last == R^{(k-1)}, R_cur == R^{(k)} - - // Create R_cur[i][j] and simplify the expression - automaton_create_proofs_simplify (R_last[i][j], R_last[i][k], - R_last[k][k], R_last[k][j], - &R_cur[i][j]); + /* Basis for the recursion: + * $R^{(k)}_{ij} = R^{(k-1)}_{ij} | R^{(k-1)}_{ik} ( R^{(k-1)}_{kk} )^* R^{(k-1)}_{kj} + * R_last == R^{(k-1)}, R_cur == R^{(k)} + */ + + /* Create R_cur[i][j] and simplify the expression */ + automaton_create_proofs_simplify (&R_last[i * n + j], &R_last[i * n + k], + &R_last[k * n + k], &R_last[k * n + j], + &R_cur[i * n + j], + &R_cur_l, &R_cur_r); } } - - // set R_last = R_cur + /* set R_last = R_cur */ + R_swap = R_last; + R_last = R_cur; + R_cur = R_swap; + /* clear 'R_cur' for next iteration */ for (i = 0; i < n; i++) - { for (j = 0; j < n; j++) - { - GNUNET_free_non_null (R_last[i][j]); - R_last[i][j] = R_cur[i][j]; - R_cur[i][j] = NULL; - } - } + R_cur[i * n + j].null_flag = GNUNET_YES; } - - // assign proofs and hashes + sb_free (&R_cur_l); + sb_free (&R_cur_r); + /* assign proofs and hashes */ for (i = 0; i < n; i++) { - if (NULL != R_last[a->start->dfs_id][i]) + if (GNUNET_YES != R_last[a->start->dfs_id * n + i].null_flag) { - states[i]->proof = GNUNET_strdup (R_last[a->start->dfs_id][i]); + states[i]->proof = GNUNET_strndup (R_last[a->start->dfs_id * n + i].sbuf, + R_last[a->start->dfs_id * n + i].slen); GNUNET_CRYPTO_hash (states[i]->proof, strlen (states[i]->proof), &states[i]->hash); } } - // complete regex for whole DFA: union of all pairs (start state/accepting - // state(s)). - complete_regex = NULL; + /* complete regex for whole DFA: union of all pairs (start state/accepting + * state(s)). */ + sb_init (&complete_regex, 16 * n); for (i = 0; i < n; i++) { if (states[i]->accepting) { - if (NULL == complete_regex && 0 < strlen (R_last[a->start->dfs_id][i])) + if ( (0 == complete_regex.slen) && + (0 < R_last[a->start->dfs_id * n + i].slen) ) { - GNUNET_asprintf (&complete_regex, "%s", R_last[a->start->dfs_id][i]); + sb_append (&complete_regex, + &R_last[a->start->dfs_id * n + i]); } - else if (NULL != R_last[a->start->dfs_id][i] && - 0 < strlen (R_last[a->start->dfs_id][i])) + else if ( (GNUNET_YES != R_last[a->start->dfs_id * n + i].null_flag) && + (0 < R_last[a->start->dfs_id * n + i].slen) ) { - temp = complete_regex; - GNUNET_asprintf (&complete_regex, "%s|%s", complete_regex, - R_last[a->start->dfs_id][i]); - GNUNET_free (temp); + sb_append_cstr (&complete_regex, "|"); + sb_append (&complete_regex, + &R_last[a->start->dfs_id * n + i]); } } } - a->canonical_regex = complete_regex; + a->canonical_regex = GNUNET_strndup (complete_regex.sbuf, complete_regex.slen); - // cleanup - for (i = 0; i < n; i++) - { + /* cleanup */ + sb_free (&complete_regex); + for (i = 0; i < n; i++) for (j = 0; j < n; j++) - GNUNET_free_non_null (R_last[i][j]); - } + { + sb_free (&R_cur[i * n + j]); + sb_free (&R_last[i * n + j]); + } + GNUNET_free (R_cur); + GNUNET_free (R_last); + return GNUNET_OK; } @@ -1373,22 +1768,16 @@ dfa_state_create (struct GNUNET_REGEX_Context *ctx, struct GNUNET_REGEX_StateSet *nfa_states) { struct GNUNET_REGEX_State *s; - char *name; - int len = 0; + char *pos; + size_t len; struct GNUNET_REGEX_State *cstate; struct GNUNET_REGEX_Transition *ctran; unsigned int i; s = GNUNET_malloc (sizeof (struct GNUNET_REGEX_State)); s->id = ctx->state_id++; - s->accepting = 0; - s->marked = GNUNET_NO; - s->name = NULL; - s->scc_id = 0; s->index = -1; s->lowlink = -1; - s->contained = 0; - s->proof = NULL; if (NULL == nfa_states) { @@ -1396,45 +1785,38 @@ dfa_state_create (struct GNUNET_REGEX_Context *ctx, return s; } - s->nfa_set = nfa_states; + s->nfa_set = *nfa_states; - if (nfa_states->len < 1) + if (nfa_states->off < 1) return s; - // Create a name based on 'nfa_states' - s->name = GNUNET_malloc (sizeof (char) * 2); + /* Create a name based on 'nfa_states' */ + len = nfa_states->off * 14 + 4; + s->name = GNUNET_malloc (len); strcat (s->name, "{"); - name = NULL; + pos = s->name + 1; - for (i = 0; i < nfa_states->len; i++) + for (i = 0; i < nfa_states->off; i++) { cstate = nfa_states->states[i]; - GNUNET_asprintf (&name, "%i,", cstate->id); + GNUNET_snprintf (pos, pos - s->name + len, + "%i,", cstate->id); + pos += strlen (pos); - if (NULL != name) - { - len = strlen (s->name) + strlen (name) + 1; - s->name = GNUNET_realloc (s->name, len); - strcat (s->name, name); - GNUNET_free (name); - name = NULL; - } - - // Add a transition for each distinct label to NULL state - for (ctran = cstate->transitions_head; NULL != ctran; ctran = ctran->next) - { + /* Add a transition for each distinct label to NULL state */ + for (ctran = cstate->transitions_head; NULL != ctran; ctran = ctran->next) if (NULL != ctran->label) - state_add_transition (ctx, s, ctran->label, NULL); - } + state_add_transition (ctx, s, ctran->label, NULL); - // If the nfa_states contain an accepting state, the new dfa state is also - // accepting + /* If the nfa_states contain an accepting state, the new dfa state is also + * accepting. */ if (cstate->accepting) s->accepting = 1; - } - - s->name[strlen (s->name) - 1] = '}'; + } + pos[-1] = '}'; + s->name = GNUNET_realloc (s->name, strlen (s->name) + 1); + memset (nfa_states, 0, sizeof (struct GNUNET_REGEX_StateSet)); return s; } @@ -1483,6 +1865,7 @@ dfa_move (struct GNUNET_REGEX_State **s, const char *str) return max_len; } + /** * Set the given state 'marked' to GNUNET_YES. Used by the * 'dfa_remove_unreachable_states' function to detect unreachable states in the @@ -1492,12 +1875,13 @@ dfa_move (struct GNUNET_REGEX_State **s, const char *str) * @param count count, not used. * @param s state where the marked attribute will be set to GNUNET_YES. */ -void +static void mark_states (void *cls, const unsigned int count, struct GNUNET_REGEX_State *s) { s->marked = GNUNET_YES; } + /** * Remove all unreachable states from DFA 'a'. Unreachable states are those * states that are not reachable from the starting state. @@ -1510,14 +1894,14 @@ dfa_remove_unreachable_states (struct GNUNET_REGEX_Automaton *a) struct GNUNET_REGEX_State *s; struct GNUNET_REGEX_State *s_next; - // 1. unmark all states + /* 1. unmark all states */ for (s = a->states_head; NULL != s; s = s->next) s->marked = GNUNET_NO; - // 2. traverse dfa from start state and mark all visited states + /* 2. traverse dfa from start state and mark all visited states */ GNUNET_REGEX_automaton_traverse (a, a->start, NULL, NULL, &mark_states, NULL); - // 3. delete all states that were not visited + /* 3. delete all states that were not visited */ for (s = a->states_head; NULL != s; s = s_next) { s_next = s->next; @@ -1537,13 +1921,16 @@ static void dfa_remove_dead_states (struct GNUNET_REGEX_Automaton *a) { struct GNUNET_REGEX_State *s; + struct GNUNET_REGEX_State *s_next; struct GNUNET_REGEX_Transition *t; int dead; GNUNET_assert (DFA == a->type); - for (s = a->states_head; NULL != s; s = s->next) + for (s = a->states_head; NULL != s; s = s_next) { + s_next = s->next; + if (s->accepting) continue; @@ -1560,7 +1947,7 @@ dfa_remove_dead_states (struct GNUNET_REGEX_Automaton *a) if (0 == dead) continue; - // state s is dead, remove it + /* state s is dead, remove it */ automaton_remove_state (a, s); } } @@ -1571,12 +1958,13 @@ dfa_remove_dead_states (struct GNUNET_REGEX_Automaton *a) * * @param ctx context * @param a DFA automaton + * @return GNUNET_OK on success */ -static void +static int dfa_merge_nondistinguishable_states (struct GNUNET_REGEX_Context *ctx, struct GNUNET_REGEX_Automaton *a) { - int table[a->state_count][a->state_count]; + uint32_t *table; struct GNUNET_REGEX_State *s1; struct GNUNET_REGEX_State *s2; struct GNUNET_REGEX_Transition *t1; @@ -1586,29 +1974,39 @@ dfa_merge_nondistinguishable_states (struct GNUNET_REGEX_Context *ctx, int change; unsigned int num_equal_edges; unsigned int i; + unsigned int state_cnt; + unsigned long long idx; + unsigned long long idx1; + + if ( (NULL == a) || (0 == a->state_count) ) + { + GNUNET_log (GNUNET_ERROR_TYPE_ERROR, + "Could not merge nondistinguishable states, automaton was NULL.\n"); + return GNUNET_SYSERR; + } - for (i = 0, s1 = a->states_head; i < a->state_count && NULL != s1; - i++, s1 = s1->next) + state_cnt = a->state_count; + table = GNUNET_malloc_large ((sizeof (uint32_t) * state_cnt * state_cnt / 32) + sizeof (uint32_t)); + if (NULL == table) { - s1->marked = i; + GNUNET_log_strerror (GNUNET_ERROR_TYPE_ERROR, "malloc"); + return GNUNET_SYSERR; } - // Mark all pairs of accepting/!accepting states + for (i = 0, s1 = a->states_head; NULL != s1; s1 = s1->next) + s1->marked = i++; + + /* Mark all pairs of accepting/!accepting states */ for (s1 = a->states_head; NULL != s1; s1 = s1->next) - { for (s2 = a->states_head; NULL != s2; s2 = s2->next) - { - table[s1->marked][s2->marked] = 0; - - if ((s1->accepting && !s2->accepting) || - (!s1->accepting && s2->accepting)) + if ( (s1->accepting && !s2->accepting) || + (!s1->accepting && s2->accepting) ) { - table[s1->marked][s2->marked] = 1; + idx = s1->marked * state_cnt + s2->marked; + table[idx / 32] |= (1 << (idx % 32)); } - } - } - // Find all equal states + /* Find all equal states */ change = 1; while (0 != change) { @@ -1617,47 +2015,57 @@ dfa_merge_nondistinguishable_states (struct GNUNET_REGEX_Context *ctx, { for (s2 = a->states_head; NULL != s2 && s1 != s2; s2 = s2->next) { - if (0 != table[s1->marked][s2->marked]) + idx = s1->marked * state_cnt + s2->marked; + if (0 != (table[idx / 32] & (1 << (idx % 32)))) continue; - num_equal_edges = 0; for (t1 = s1->transitions_head; NULL != t1; t1 = t1->next) { for (t2 = s2->transitions_head; NULL != t2; t2 = t2->next) { if (0 == strcmp (t1->label, t2->label)) - { - num_equal_edges++; - if (0 != table[t1->to_state->marked][t2->to_state->marked] || - 0 != table[t2->to_state->marked][t1->to_state->marked]) - { - table[s1->marked][s2->marked] = 1; - change = 1; - } - } - } + { + num_equal_edges++; + /* same edge, but targets definitively different, so we're different + as well */ + if (t1->to_state->marked > t2->to_state->marked) + idx1 = t1->to_state->marked * state_cnt + t2->to_state->marked; + else + idx1 = t2->to_state->marked * state_cnt + t1->to_state->marked; + if (0 != (table[idx1 / 32] & (1 << (idx1 % 32)))) + { + table[idx / 32] |= (1 << (idx % 32)); + change = 1; /* changed a marker, need to run again */ + } + } + } } - if (num_equal_edges != s1->transition_count || - num_equal_edges != s2->transition_count) + if ( (num_equal_edges != s1->transition_count) || + (num_equal_edges != s2->transition_count) ) { - // Make sure ALL edges of possible equal states are the same - table[s1->marked][s2->marked] = -2; + /* Make sure ALL edges of possible equal states are the same */ + table[idx / 32] |= (1 << (idx % 32)); + change = 1; /* changed a marker, need to run again */ } } } } - // Merge states that are equal + /* Merge states that are equal */ for (s1 = a->states_head; NULL != s1; s1 = s1_next) { s1_next = s1->next; for (s2 = a->states_head; NULL != s2 && s1 != s2; s2 = s2_next) { s2_next = s2->next; - if (table[s1->marked][s2->marked] == 0) + idx = s1->marked * state_cnt + s2->marked; + if (0 == (table[idx / 32] & (1 << (idx % 32)))) automaton_merge_states (ctx, a, s1, s2); } } + + GNUNET_free (table); + return GNUNET_OK; } @@ -1667,24 +2075,288 @@ dfa_merge_nondistinguishable_states (struct GNUNET_REGEX_Context *ctx, * * @param ctx context * @param a DFA automaton + * @return GNUNET_OK on success */ -static void +static int dfa_minimize (struct GNUNET_REGEX_Context *ctx, struct GNUNET_REGEX_Automaton *a) { - if (NULL == a) + if (NULL == a) + return GNUNET_SYSERR; + + GNUNET_assert (DFA == a->type); + + /* 1. remove unreachable states */ + dfa_remove_unreachable_states (a); + + /* 2. remove dead states */ + dfa_remove_dead_states (a); + + /* 3. Merge nondistinguishable states */ + if (GNUNET_OK != dfa_merge_nondistinguishable_states (ctx, a)) + return GNUNET_SYSERR; + return GNUNET_OK; +} + + +/** + * Context for adding strided transitions to a DFA. + */ +struct GNUNET_REGEX_Strided_Context +{ + /** + * Length of the strides. + */ + const unsigned int stride; + + /** + * Strided transitions DLL. New strided transitions will be stored in this DLL + * and afterwards added to the DFA. + */ + struct GNUNET_REGEX_Transition *transitions_head; + + /** + * Strided transitions DLL. + */ + struct GNUNET_REGEX_Transition *transitions_tail; +}; + + +/** + * Recursive helper function to add strides to a DFA. + * + * @param cls context, contains stride length and strided transitions DLL. + * @param depth current depth of the depth-first traversal of the graph. + * @param label current label, string that contains all labels on the path from + * 'start' to 's'. + * @param start start state for the depth-first traversal of the graph. + * @param s current state in the depth-first traversal + */ +void +dfa_add_multi_strides_helper (void *cls, const unsigned int depth, char *label, + struct GNUNET_REGEX_State *start, + struct GNUNET_REGEX_State *s) +{ + struct GNUNET_REGEX_Strided_Context *ctx = cls; + struct GNUNET_REGEX_Transition *t; + char *new_label; + + if (depth == ctx->stride) + { + t = GNUNET_malloc (sizeof (struct GNUNET_REGEX_Transition)); + t->label = GNUNET_strdup (label); + t->to_state = s; + t->from_state = start; + GNUNET_CONTAINER_DLL_insert (ctx->transitions_head, ctx->transitions_tail, + t); + } + else + { + for (t = s->transitions_head; NULL != t; t = t->next) + { + /* Do not consider self-loops, because it end's up in too many + * transitions */ + if (t->to_state == t->from_state) + continue; + + if (NULL != label) + { + GNUNET_asprintf (&new_label, "%s%s", label, t->label); + } + else + new_label = GNUNET_strdup (t->label); + + dfa_add_multi_strides_helper (cls, (depth + 1), new_label, start, + t->to_state); + } + } + GNUNET_free_non_null (label); +} + + +/** + * Function called for each state in the DFA. Starts a traversal of depth set in + * context starting from state 's'. + * + * @param cls context. + * @param count not used. + * @param s current state. + */ +void +dfa_add_multi_strides (void *cls, const unsigned int count, + struct GNUNET_REGEX_State *s) +{ + dfa_add_multi_strides_helper (cls, 0, NULL, s, s); +} + + +/** + * Adds multi-strided transitions to the given 'dfa'. + * + * @param regex_ctx regex context needed to add transitions to the automaton. + * @param dfa DFA to which the multi strided transitions should be added. + * @param stride_len length of the strides. + */ +void +GNUNET_REGEX_dfa_add_multi_strides (struct GNUNET_REGEX_Context *regex_ctx, + struct GNUNET_REGEX_Automaton *dfa, + const unsigned int stride_len) +{ + struct GNUNET_REGEX_Strided_Context ctx = { stride_len, NULL, NULL }; + struct GNUNET_REGEX_Transition *t; + struct GNUNET_REGEX_Transition *t_next; + + if (1 > stride_len || GNUNET_YES == dfa->is_multistrided) + return; + + /* Compute the new transitions of given stride_len */ + GNUNET_REGEX_automaton_traverse (dfa, dfa->start, NULL, NULL, + &dfa_add_multi_strides, &ctx); + + /* Add all the new transitions to the automaton. */ + for (t = ctx.transitions_head; NULL != t; t = t_next) + { + t_next = t->next; + state_add_transition (regex_ctx, t->from_state, t->label, t->to_state); + GNUNET_CONTAINER_DLL_remove (ctx.transitions_head, ctx.transitions_tail, t); + GNUNET_free_non_null (t->label); + GNUNET_free (t); + } + + /* Mark this automaton as multistrided */ + dfa->is_multistrided = GNUNET_YES; +} + +/** + * Recursive Helper function for DFA path compression. Does DFS on the DFA graph + * and adds new transitions to the given transitions DLL and marks states that + * should be removed by setting state->contained to GNUNET_YES. + * + * @param dfa DFA for which the paths should be compressed. + * @param start starting state for linear path search. + * @param cur current state in the recursive DFS. + * @param label current label (string of traversed labels). + * @param max_len maximal path compression length. + * @param transitions_head transitions DLL. + * @param transitions_tail transitions DLL. + */ +void +dfa_compress_paths_helper (struct GNUNET_REGEX_Automaton *dfa, + struct GNUNET_REGEX_State *start, + struct GNUNET_REGEX_State *cur, char *label, + unsigned int max_len, + struct GNUNET_REGEX_Transition **transitions_head, + struct GNUNET_REGEX_Transition **transitions_tail) +{ + struct GNUNET_REGEX_Transition *t; + char *new_label; + + + if (NULL != label && + ((cur->incoming_transition_count > 1 || GNUNET_YES == cur->accepting || + GNUNET_YES == cur->marked) || (start != dfa->start && max_len > 0 && + max_len == strlen (label)) || + (start == dfa->start && GNUNET_REGEX_INITIAL_BYTES == strlen (label)))) + { + t = GNUNET_malloc (sizeof (struct GNUNET_REGEX_Transition)); + t->label = GNUNET_strdup (label); + t->to_state = cur; + t->from_state = start; + GNUNET_CONTAINER_DLL_insert (*transitions_head, *transitions_tail, t); + + if (GNUNET_NO == cur->marked) + { + dfa_compress_paths_helper (dfa, cur, cur, NULL, max_len, transitions_head, + transitions_tail); + } + return; + } + else if (cur != start) + cur->contained = GNUNET_YES; + + if (GNUNET_YES == cur->marked && cur != start) + return; + + cur->marked = GNUNET_YES; + + + for (t = cur->transitions_head; NULL != t; t = t->next) + { + if (NULL != label) + GNUNET_asprintf (&new_label, "%s%s", label, t->label); + else + new_label = GNUNET_strdup (t->label); + + if (t->to_state != cur) + { + dfa_compress_paths_helper (dfa, start, t->to_state, new_label, max_len, + transitions_head, transitions_tail); + } + GNUNET_free (new_label); + } +} + + +/** + * Compress paths in the given 'dfa'. Linear paths like 0->1->2->3 will be + * compressed to 0->3 by combining transitions. + * + * @param regex_ctx context for adding new transitions. + * @param dfa DFA representation, will directly modify the given DFA. + * @param max_len maximal length of the compressed paths. + */ +static void +dfa_compress_paths (struct GNUNET_REGEX_Context *regex_ctx, + struct GNUNET_REGEX_Automaton *dfa, unsigned int max_len) +{ + struct GNUNET_REGEX_State *s; + struct GNUNET_REGEX_State *s_next; + struct GNUNET_REGEX_Transition *t; + struct GNUNET_REGEX_Transition *t_next; + struct GNUNET_REGEX_Transition *transitions_head = NULL; + struct GNUNET_REGEX_Transition *transitions_tail = NULL; + + if (NULL == dfa) return; - GNUNET_assert (DFA == a->type); + /* Count the incoming transitions on each state. */ + for (s = dfa->states_head; NULL != s; s = s->next) + { + for (t = s->transitions_head; NULL != t; t = t->next) + { + if (NULL != t->to_state) + t->to_state->incoming_transition_count++; + } + } - // 1. remove unreachable states - dfa_remove_unreachable_states (a); + /* Unmark all states. */ + for (s = dfa->states_head; NULL != s; s = s->next) + { + s->marked = GNUNET_NO; + s->contained = GNUNET_NO; + } - // 2. remove dead states - dfa_remove_dead_states (a); + /* Add strides and mark states that can be deleted. */ + dfa_compress_paths_helper (dfa, dfa->start, dfa->start, NULL, max_len, + &transitions_head, &transitions_tail); + + /* Add all the new transitions to the automaton. */ + for (t = transitions_head; NULL != t; t = t_next) + { + t_next = t->next; + state_add_transition (regex_ctx, t->from_state, t->label, t->to_state); + GNUNET_CONTAINER_DLL_remove (transitions_head, transitions_tail, t); + GNUNET_free_non_null (t->label); + GNUNET_free (t); + } - // 3. Merge nondistinguishable states - dfa_merge_nondistinguishable_states (ctx, a); + /* Remove marked states (including their incoming and outgoing transitions). */ + for (s = dfa->states_head; NULL != s; s = s_next) + { + s_next = s->next; + if (GNUNET_YES == s->contained) + automaton_remove_state (dfa, s); + } } @@ -1791,127 +2463,73 @@ nfa_state_create (struct GNUNET_REGEX_Context *ctx, int accepting) } -/** - * Calculates the NFA closure set for the given state. - * - * @param nfa the NFA containing 's' - * @param s starting point state - * @param label transitioning label on which to base the closure on, - * pass NULL for epsilon transition - * - * @return sorted nfa closure on 'label' (epsilon closure if 'label' is NULL) - */ -static struct GNUNET_REGEX_StateSet * -nfa_closure_create (struct GNUNET_REGEX_Automaton *nfa, - struct GNUNET_REGEX_State *s, const char *label) -{ - struct GNUNET_REGEX_StateSet *cls; - struct GNUNET_REGEX_StateSet *cls_check; - struct GNUNET_REGEX_State *clsstate; - struct GNUNET_REGEX_State *currentstate; - struct GNUNET_REGEX_Transition *ctran; - - if (NULL == s) - return NULL; - - cls = GNUNET_malloc (sizeof (struct GNUNET_REGEX_StateSet)); - cls_check = GNUNET_malloc (sizeof (struct GNUNET_REGEX_StateSet)); - - for (clsstate = nfa->states_head; NULL != clsstate; clsstate = clsstate->next) - clsstate->contained = 0; - - // Add start state to closure only for epsilon closure - if (NULL == label) - GNUNET_array_append (cls->states, cls->len, s); - - GNUNET_array_append (cls_check->states, cls_check->len, s); - while (cls_check->len > 0) - { - currentstate = cls_check->states[cls_check->len - 1]; - GNUNET_array_grow (cls_check->states, cls_check->len, cls_check->len - 1); - - for (ctran = currentstate->transitions_head; NULL != ctran; - ctran = ctran->next) - { - if (NULL != ctran->to_state && 0 == nullstrcmp (label, ctran->label)) - { - clsstate = ctran->to_state; - - if (NULL != clsstate && 0 == clsstate->contained) - { - GNUNET_array_append (cls->states, cls->len, clsstate); - GNUNET_array_append (cls_check->states, cls_check->len, clsstate); - clsstate->contained = 1; - } - } - } - } - GNUNET_assert (0 == cls_check->len); - GNUNET_free (cls_check); - - // sort the states - if (cls->len > 1) - qsort (cls->states, cls->len, sizeof (struct GNUNET_REGEX_State *), - state_compare); - - return cls; -} - - /** * Calculates the closure set for the given set of states. * + * @param ret set to sorted nfa closure on 'label' (epsilon closure if 'label' is NULL) * @param nfa the NFA containing 's' * @param states list of states on which to base the closure on * @param label transitioning label for which to base the closure on, * pass NULL for epsilon transition - * - * @return sorted nfa closure on 'label' (epsilon closure if 'label' is NULL) */ -static struct GNUNET_REGEX_StateSet * -nfa_closure_set_create (struct GNUNET_REGEX_Automaton *nfa, +static void +nfa_closure_set_create (struct GNUNET_REGEX_StateSet *ret, + struct GNUNET_REGEX_Automaton *nfa, struct GNUNET_REGEX_StateSet *states, const char *label) { struct GNUNET_REGEX_State *s; - struct GNUNET_REGEX_StateSet *sset; - struct GNUNET_REGEX_StateSet *cls; unsigned int i; - unsigned int j; - unsigned int k; - unsigned int contains; + struct GNUNET_REGEX_StateSet_MDLL cls_stack; + struct GNUNET_REGEX_State *clsstate; + struct GNUNET_REGEX_State *currentstate; + struct GNUNET_REGEX_Transition *ctran; + memset (ret, 0, sizeof (struct GNUNET_REGEX_StateSet)); if (NULL == states) - return NULL; - - cls = GNUNET_malloc (sizeof (struct GNUNET_REGEX_StateSet)); + return; - for (i = 0; i < states->len; i++) + for (i = 0; i < states->off; i++) { s = states->states[i]; - sset = nfa_closure_create (nfa, s, label); - for (j = 0; j < sset->len; j++) + /* Add start state to closure only for epsilon closure */ + if (NULL == label) + state_set_append (ret, s); + + /* initialize work stack */ + cls_stack.head = NULL; + cls_stack.tail = NULL; + GNUNET_CONTAINER_MDLL_insert (ST, cls_stack.head, cls_stack.tail, s); + cls_stack.len = 1; + + while (NULL != (currentstate = cls_stack.tail)) { - contains = 0; - for (k = 0; k < cls->len; k++) + GNUNET_CONTAINER_MDLL_remove (ST, cls_stack.head, cls_stack.tail, + currentstate); + cls_stack.len--; + for (ctran = currentstate->transitions_head; NULL != ctran; + ctran = ctran->next) { - if (sset->states[j]->id == cls->states[k]->id) - { - contains = 1; - break; - } - } - if (!contains) - GNUNET_array_append (cls->states, cls->len, sset->states[j]); + if (NULL == (clsstate = ctran->to_state)) + continue; + if (0 != clsstate->contained) + continue; + if (0 != nullstrcmp (label, ctran->label)) + continue; + state_set_append (ret, clsstate); + GNUNET_CONTAINER_MDLL_insert_tail (ST, cls_stack.head, cls_stack.tail, + clsstate); + cls_stack.len++; + clsstate->contained = 1; + } } - state_set_clear (sset); } + for (i = 0; i < ret->off; i++) + ret->states[i]->contained = 0; - if (cls->len > 1) - qsort (cls->states, cls->len, sizeof (struct GNUNET_REGEX_State *), - state_compare); - - return cls; + if (ret->off > 1) + qsort (ret->states, ret->off, sizeof (struct GNUNET_REGEX_State *), + &state_compare); } @@ -2005,6 +2623,14 @@ nfa_add_plus_op (struct GNUNET_REGEX_Context *ctx) struct GNUNET_REGEX_Automaton *a; a = ctx->stack_tail; + + if (NULL == a) + { + GNUNET_log (GNUNET_ERROR_TYPE_ERROR, + "nfa_add_plus_op failed, because there was no element on the stack"); + return; + } + GNUNET_CONTAINER_DLL_remove (ctx->stack_head, ctx->stack_tail, a); state_add_transition (ctx, a->end, NULL, a->start); @@ -2048,9 +2674,8 @@ nfa_add_question_op (struct GNUNET_REGEX_Context *ctx) new_nfa = nfa_fragment_create (start, end); nfa_add_states (new_nfa, a->states_head, a->states_tail); - automaton_fragment_clear (a); - GNUNET_CONTAINER_DLL_insert_tail (ctx->stack_head, ctx->stack_tail, new_nfa); + automaton_fragment_clear (a); } @@ -2161,7 +2786,8 @@ GNUNET_REGEX_construct_nfa (const char *regex, const size_t len) unsigned int count; unsigned int altcount; unsigned int atomcount; - unsigned int pcount; + unsigned int poff; + unsigned int psize; struct { int altcount; @@ -2175,7 +2801,6 @@ GNUNET_REGEX_construct_nfa (const char *regex, const size_t len) return NULL; } - GNUNET_REGEX_context_init (&ctx); regexp = regex; @@ -2184,7 +2809,8 @@ GNUNET_REGEX_construct_nfa (const char *regex, const size_t len) error_msg = NULL; altcount = 0; atomcount = 0; - pcount = 0; + poff = 0; + psize = 0; for (count = 0; count < len && *regexp; count++, regexp++) { @@ -2196,9 +2822,11 @@ GNUNET_REGEX_construct_nfa (const char *regex, const size_t len) --atomcount; nfa_add_concatenation (&ctx); } - GNUNET_array_grow (p, pcount, pcount + 1); - p[pcount - 1].altcount = altcount; - p[pcount - 1].atomcount = atomcount; + if (poff == psize) + GNUNET_array_grow (p, psize, psize * 2 + 4); + p[poff].altcount = altcount; + p[poff].atomcount = atomcount; + poff++; altcount = 0; atomcount = 0; break; @@ -2213,26 +2841,26 @@ GNUNET_REGEX_construct_nfa (const char *regex, const size_t len) altcount++; break; case ')': - if (0 == pcount) + if (0 == poff) { error_msg = "Missing opening '('"; goto error; } if (0 == atomcount) { - // Ignore this: "()" - pcount--; - altcount = p[pcount].altcount; - atomcount = p[pcount].atomcount; + /* Ignore this: "()" */ + poff--; + altcount = p[poff].altcount; + atomcount = p[poff].atomcount; break; } while (--atomcount > 0) nfa_add_concatenation (&ctx); for (; altcount > 0; altcount--) nfa_add_alternation (&ctx); - pcount--; - altcount = p[pcount].altcount; - atomcount = p[pcount].atomcount; + poff--; + altcount = p[poff].altcount; + atomcount = p[poff].atomcount; atomcount++; break; case '*': @@ -2271,7 +2899,7 @@ GNUNET_REGEX_construct_nfa (const char *regex, const size_t len) break; } } - if (0 != pcount) + if (0 != poff) { error_msg = "Unbalanced parenthesis"; goto error; @@ -2281,7 +2909,7 @@ GNUNET_REGEX_construct_nfa (const char *regex, const size_t len) for (; altcount > 0; altcount--) nfa_add_alternation (&ctx); - GNUNET_free_non_null (p); + GNUNET_array_grow (p, psize, 0); nfa = ctx.stack_tail; GNUNET_CONTAINER_DLL_remove (ctx.stack_head, ctx.stack_tail, nfa); @@ -2304,7 +2932,7 @@ GNUNET_REGEX_construct_nfa (const char *regex, const size_t len) return nfa; error: - GNUNET_log (GNUNET_ERROR_TYPE_ERROR, "Could not parse regex: %s\n", regex); + GNUNET_log (GNUNET_ERROR_TYPE_ERROR, "Could not parse regex: `%s'\n", regex); if (NULL != error_msg) GNUNET_log (GNUNET_ERROR_TYPE_ERROR, "%s\n", error_msg); @@ -2336,31 +2964,34 @@ construct_dfa_states (struct GNUNET_REGEX_Context *ctx, struct GNUNET_REGEX_State *dfa_state) { struct GNUNET_REGEX_Transition *ctran; - struct GNUNET_REGEX_State *state_iter; struct GNUNET_REGEX_State *new_dfa_state; struct GNUNET_REGEX_State *state_contains; - struct GNUNET_REGEX_StateSet *tmp; - struct GNUNET_REGEX_StateSet *nfa_set; + struct GNUNET_REGEX_State *state_iter; + struct GNUNET_REGEX_StateSet tmp; + struct GNUNET_REGEX_StateSet nfa_set; for (ctran = dfa_state->transitions_head; NULL != ctran; ctran = ctran->next) { if (NULL == ctran->label || NULL != ctran->to_state) continue; - tmp = nfa_closure_set_create (nfa, dfa_state->nfa_set, ctran->label); - nfa_set = nfa_closure_set_create (nfa, tmp, 0); - state_set_clear (tmp); - new_dfa_state = dfa_state_create (ctx, nfa_set); + nfa_closure_set_create (&tmp, nfa, &dfa_state->nfa_set, ctran->label); + nfa_closure_set_create (&nfa_set, nfa, &tmp, NULL); + state_set_clear (&tmp); + state_contains = NULL; for (state_iter = dfa->states_head; NULL != state_iter; state_iter = state_iter->next) { - if (0 == state_set_compare (state_iter->nfa_set, new_dfa_state->nfa_set)) + if (0 == state_set_compare (&state_iter->nfa_set, &nfa_set)) + { state_contains = state_iter; + break; + } } - if (NULL == state_contains) { + new_dfa_state = dfa_state_create (ctx, &nfa_set); automaton_add_state (dfa, new_dfa_state); ctran->to_state = new_dfa_state; construct_dfa_states (ctx, nfa, dfa, new_dfa_state); @@ -2368,31 +2999,43 @@ construct_dfa_states (struct GNUNET_REGEX_Context *ctx, else { ctran->to_state = state_contains; - automaton_destroy_state (new_dfa_state); + state_set_clear (&nfa_set); } } } /** - * Construct DFA for the given 'regex' of length 'len' - * - * @param regex regular expression string - * @param len length of the regular expression - * - * @return DFA, needs to be freed using GNUNET_REGEX_destroy_automaton + * Construct DFA for the given 'regex' of length 'len'. + * + * Path compression means, that for example a DFA o -> a -> b -> c -> o will be + * compressed to o -> abc -> o. Note that this parameter influences the + * non-determinism of states of the resulting NFA in the DHT (number of outgoing + * edges with the same label). For example for an application that stores IPv4 + * addresses as bitstrings it could make sense to limit the path compression to + * 4 or 8. + * + * @param regex regular expression string. + * @param len length of the regular expression. + * @param max_path_len limit the path compression length to the + * given value. If set to 1, no path compression is applied. Set to 0 for + * maximal possible path compression (generally not desireable). + * @return DFA, needs to be freed using GNUNET_REGEX_automaton_destroy. */ struct GNUNET_REGEX_Automaton * -GNUNET_REGEX_construct_dfa (const char *regex, const size_t len) +GNUNET_REGEX_construct_dfa (const char *regex, const size_t len, + unsigned int max_path_len) { struct GNUNET_REGEX_Context ctx; struct GNUNET_REGEX_Automaton *dfa; struct GNUNET_REGEX_Automaton *nfa; - struct GNUNET_REGEX_StateSet *nfa_start_eps_cls; + struct GNUNET_REGEX_StateSet nfa_start_eps_cls; + struct GNUNET_REGEX_StateSet singleton_set; GNUNET_REGEX_context_init (&ctx); - // Create NFA + /* Create NFA */ + // fprintf (stderr, "N"); nfa = GNUNET_REGEX_construct_nfa (regex, len); if (NULL == nfa) @@ -2404,29 +3047,38 @@ GNUNET_REGEX_construct_dfa (const char *regex, const size_t len) dfa = GNUNET_malloc (sizeof (struct GNUNET_REGEX_Automaton)); dfa->type = DFA; - dfa->state_count = 0; - dfa->states_head = NULL; - dfa->states_tail = NULL; dfa->regex = GNUNET_strdup (regex); - dfa->is_multistrided = GNUNET_NO; - // Create DFA start state from epsilon closure - nfa_start_eps_cls = nfa_closure_create (nfa, nfa->start, 0); - dfa->start = dfa_state_create (&ctx, nfa_start_eps_cls); + /* Create DFA start state from epsilon closure */ + memset (&singleton_set, 0, sizeof (struct GNUNET_REGEX_StateSet)); + state_set_append (&singleton_set, nfa->start); + nfa_closure_set_create (&nfa_start_eps_cls, nfa, &singleton_set, NULL); + state_set_clear (&singleton_set); + dfa->start = dfa_state_create (&ctx, &nfa_start_eps_cls); automaton_add_state (dfa, dfa->start); + // fprintf (stderr, "D"); construct_dfa_states (&ctx, nfa, dfa, dfa->start); - GNUNET_REGEX_automaton_destroy (nfa); - // Minimize DFA - dfa_minimize (&ctx, dfa); + /* Minimize DFA */ + // fprintf (stderr, "M"); + if (GNUNET_OK != dfa_minimize (&ctx, dfa)) + { + GNUNET_REGEX_automaton_destroy (dfa); + return NULL; + } - // Create proofs for all states - automaton_create_proofs (dfa); + /* Create proofs and hashes for all states */ + if (GNUNET_OK != automaton_create_proofs (dfa)) + { + GNUNET_REGEX_automaton_destroy (dfa); + return NULL; + } - // Add strides to DFA - // GNUNET_REGEX_add_multi_strides_to_dfa (&ctx, dfa, 2); + /* Compress linear DFA paths */ + if (1 != max_path_len) + dfa_compress_paths (&ctx, dfa, max_path_len); return dfa; } @@ -2450,11 +3102,11 @@ GNUNET_REGEX_automaton_destroy (struct GNUNET_REGEX_Automaton *a) GNUNET_free_non_null (a->regex); GNUNET_free_non_null (a->canonical_regex); - for (s = a->states_head; NULL != s;) + for (s = a->states_head; NULL != s; s = next_state) { next_state = s->next; + GNUNET_CONTAINER_DLL_remove (a->states_head, a->states_tail, s); automaton_destroy_state (s); - s = next_state; } GNUNET_free (a); @@ -2485,7 +3137,7 @@ evaluate_dfa (struct GNUNET_REGEX_Automaton *a, const char *string) s = a->start; - // If the string is empty but the starting state is accepting, we accept. + /* If the string is empty but the starting state is accepting, we accept. */ if ((NULL == string || 0 == strlen (string)) && s->accepting) return 0; @@ -2518,8 +3170,9 @@ evaluate_nfa (struct GNUNET_REGEX_Automaton *a, const char *string) const char *strp; char str[2]; struct GNUNET_REGEX_State *s; - struct GNUNET_REGEX_StateSet *sset; - struct GNUNET_REGEX_StateSet *new_sset; + struct GNUNET_REGEX_StateSet sset; + struct GNUNET_REGEX_StateSet new_sset; + struct GNUNET_REGEX_StateSet singleton_set; unsigned int i; int result; @@ -2530,34 +3183,37 @@ evaluate_nfa (struct GNUNET_REGEX_Automaton *a, const char *string) return -1; } - // If the string is empty but the starting state is accepting, we accept. + /* If the string is empty but the starting state is accepting, we accept. */ if ((NULL == string || 0 == strlen (string)) && a->start->accepting) return 0; result = 1; - sset = nfa_closure_create (a, a->start, 0); + memset (&singleton_set, 0, sizeof (struct GNUNET_REGEX_StateSet)); + state_set_append (&singleton_set, a->start); + nfa_closure_set_create (&sset, a, &singleton_set, NULL); + state_set_clear (&singleton_set); str[1] = '\0'; for (strp = string; NULL != strp && *strp; strp++) { str[0] = *strp; - new_sset = nfa_closure_set_create (a, sset, str); - state_set_clear (sset); - sset = nfa_closure_set_create (a, new_sset, 0); - state_set_clear (new_sset); + nfa_closure_set_create (&new_sset, a, &sset, str); + state_set_clear (&sset); + nfa_closure_set_create (&sset, a, &new_sset, 0); + state_set_clear (&new_sset); } - for (i = 0; i < sset->len; i++) + for (i = 0; i < sset.off; i++) { - s = sset->states[i]; - if (NULL != s && s->accepting) + s = sset.states[i]; + if ( (NULL != s) && (s->accepting) ) { result = 0; break; } } - state_set_clear (sset); + state_set_clear (&sset); return result; } @@ -2631,10 +3287,9 @@ GNUNET_REGEX_get_transition_count (struct GNUNET_REGEX_Automaton *a) if (NULL == a) return 0; - for (t_count = 0, s = a->states_head; NULL != s; s = s->next) - { + t_count = 0; + for (s = a->states_head; NULL != s; s = s->next) t_count += s->transition_count; - } return t_count; } @@ -2657,7 +3312,9 @@ GNUNET_REGEX_get_first_key (const char *input_string, size_t string_len, { unsigned int size; - size = string_len < INITIAL_BITS ? string_len : INITIAL_BITS; + size = + string_len < + GNUNET_REGEX_INITIAL_BYTES ? string_len : GNUNET_REGEX_INITIAL_BYTES; if (NULL == input_string) { @@ -2697,12 +3354,10 @@ GNUNET_REGEX_check_proof (const char *proof, const struct GNUNET_HashCode *key) /** - * Recursive helper function for iterate_initial_edges. Will call iterator - * function for each initial state. + * Recursive function that calls the iterator for each synthetic start state. * * @param min_len minimum length of the path in the graph. * @param max_len maximum length of the path in the graph. - * @param cur_len current length of the path already traversed. * @param consumed_string string consumed by traversing the graph till this state. * @param state current state of the automaton. * @param iterator iterator function called for each edge. @@ -2710,8 +3365,7 @@ GNUNET_REGEX_check_proof (const char *proof, const struct GNUNET_HashCode *key) */ static void iterate_initial_edge (const unsigned int min_len, const unsigned int max_len, - unsigned int cur_len, char *consumed_string, - struct GNUNET_REGEX_State *state, + char *consumed_string, struct GNUNET_REGEX_State *state, GNUNET_REGEX_KeyIterator iterator, void *iterator_cls) { unsigned int i; @@ -2719,24 +3373,64 @@ iterate_initial_edge (const unsigned int min_len, const unsigned int max_len, struct GNUNET_REGEX_Transition *t; unsigned int num_edges = state->transition_count; struct GNUNET_REGEX_Edge edges[num_edges]; + struct GNUNET_REGEX_Edge edge[1]; struct GNUNET_HashCode hash; + struct GNUNET_HashCode hash_new; + + unsigned int cur_len; + + if (NULL != consumed_string) + cur_len = strlen (consumed_string); + else + cur_len = 0; - if (cur_len > min_len && NULL != consumed_string && cur_len <= max_len) + if ((cur_len >= min_len || GNUNET_YES == state->accepting) && cur_len > 0 && + NULL != consumed_string) { - for (i = 0, t = state->transitions_head; NULL != t; t = t->next, i++) + if (cur_len <= max_len) { - edges[i].label = t->label; - edges[i].destination = t->to_state->hash; - } + if (state->proof != NULL && 0 != strcmp (consumed_string, state->proof)) + { + for (i = 0, t = state->transitions_head; NULL != t && i < num_edges; + t = t->next, i++) + { + edges[i].label = t->label; + edges[i].destination = t->to_state->hash; + } + GNUNET_CRYPTO_hash (consumed_string, strlen (consumed_string), &hash); + iterator (iterator_cls, &hash, consumed_string, state->accepting, + num_edges, edges); + } - GNUNET_CRYPTO_hash (consumed_string, strlen (consumed_string), &hash); - iterator (iterator_cls, &hash, consumed_string, state->accepting, num_edges, - edges); + if (GNUNET_YES == state->accepting && cur_len > 1 && + state->transition_count < 1 && cur_len < max_len) + { + /* Special case for regex consisting of just a string that is shorter than + * max_len */ + edge[0].label = &consumed_string[cur_len - 1]; + edge[0].destination = state->hash; + temp = GNUNET_strdup (consumed_string); + temp[cur_len - 1] = '\0'; + GNUNET_CRYPTO_hash (temp, cur_len - 1, &hash_new); + iterator (iterator_cls, &hash_new, temp, GNUNET_NO, 1, edge); + GNUNET_free (temp); + } + } + else if (max_len < cur_len) + { + /* Case where the concatenated labels are longer than max_len, then split. */ + edge[0].label = &consumed_string[max_len]; + edge[0].destination = state->hash; + temp = GNUNET_strdup (consumed_string); + temp[max_len] = '\0'; + GNUNET_CRYPTO_hash (temp, max_len, &hash); + iterator (iterator_cls, &hash, temp, GNUNET_NO, 1, edge); + GNUNET_free (temp); + } } if (cur_len < max_len) { - cur_len++; for (t = state->transitions_head; NULL != t; t = t->next) { if (NULL != consumed_string) @@ -2744,8 +3438,8 @@ iterate_initial_edge (const unsigned int min_len, const unsigned int max_len, else GNUNET_asprintf (&temp, "%s", t->label); - iterate_initial_edge (min_len, max_len, cur_len, temp, t->to_state, - iterator, iterator_cls); + iterate_initial_edge (min_len, max_len, temp, t->to_state, iterator, + iterator_cls); GNUNET_free (temp); } } @@ -2753,116 +3447,276 @@ iterate_initial_edge (const unsigned int min_len, const unsigned int max_len, /** - * Iterate over all initial edges that aren't actually part of the automaton. - * This is needed to find the initial states returned by - * GNUNET_REGEX_get_first_key. Iteration will start at the first state that has - * more than one outgoing edge, i.e. the state that branches the graph. - * For example consider the following graph: - * a -> b -> c -> d -> ... - * \-> e -> ... - * - * This function will not iterate over the edges leading to "c", because these - * will be covered by the iterate_edges function. + * Iterate over all edges starting from start state of automaton 'a'. Calling + * iterator for each edge. * - * @param a the automaton for which the initial states should be computed. - * @param initial_len length of the initial state string. - * @param iterator iterator function called for each edge. - * @param iterator_cls closure for the iterator function. + * @param a automaton. + * @param iterator iterator called for each edge. + * @param iterator_cls closure. */ void -iterate_initial_edges (struct GNUNET_REGEX_Automaton *a, - const unsigned int initial_len, - GNUNET_REGEX_KeyIterator iterator, void *iterator_cls) +GNUNET_REGEX_iterate_all_edges (struct GNUNET_REGEX_Automaton *a, + GNUNET_REGEX_KeyIterator iterator, + void *iterator_cls) { - char *consumed_string; - char *temp; struct GNUNET_REGEX_State *s; - unsigned int cur_len; - if (1 > initial_len) - return; + for (s = a->states_head; NULL != s; s = s->next) + { + struct GNUNET_REGEX_Edge edges[s->transition_count]; + unsigned int num_edges; - consumed_string = NULL; - s = a->start; - cur_len = 0; + num_edges = state_get_edges (s, edges); - if (1 == s->transition_count) - { - do - { - if (NULL != consumed_string) - { - temp = consumed_string; - GNUNET_asprintf (&consumed_string, "%s%s", consumed_string, - s->transitions_head->label); - GNUNET_free (temp); - } - else - GNUNET_asprintf (&consumed_string, "%s", s->transitions_head->label); + if ((NULL != s->proof && 0 < strlen (s->proof)) || s->accepting) + iterator (iterator_cls, &s->hash, s->proof, s->accepting, num_edges, + edges); - s = s->transitions_head->to_state; - cur_len += strlen (s->transitions_head->label); - } - while (cur_len < initial_len && 1 == s->transition_count); + s->marked = GNUNET_NO; } - iterate_initial_edge (cur_len, initial_len, cur_len, consumed_string, s, - iterator, iterator_cls); - - GNUNET_free_non_null (consumed_string); + iterate_initial_edge (GNUNET_REGEX_INITIAL_BYTES, GNUNET_REGEX_INITIAL_BYTES, + NULL, a->start, iterator, iterator_cls); } /** - * Iterate over all edges helper function starting from state 's', calling - * iterator function for each edge. + * Create a string with binary IP notation for the given 'addr' in 'str'. * - * @param s state. - * @param iterator iterator function called for each edge. - * @param iterator_cls closure. + * @param af address family of the given 'addr'. + * @param addr address that should be converted to a string. + * struct in_addr * for IPv4 and struct in6_addr * for IPv6. + * @param str string that will contain binary notation of 'addr'. Expected + * to be at least 33 bytes long for IPv4 and 129 bytes long for IPv6. */ static void -iterate_edge (struct GNUNET_REGEX_State *s, GNUNET_REGEX_KeyIterator iterator, - void *iterator_cls) +iptobinstr (const int af, const void *addr, char *str) { - struct GNUNET_REGEX_Transition *t; - struct GNUNET_REGEX_Edge edges[s->transition_count]; - unsigned int num_edges; + int i; - if (GNUNET_YES != s->marked) + switch (af) { - s->marked = GNUNET_YES; - - num_edges = state_get_edges (s, edges); + case AF_INET: + { + uint32_t b = htonl (((struct in_addr *) addr)->s_addr); - if ((NULL != s->proof && 0 < strlen (s->proof)) || s->accepting) - iterator (iterator_cls, &s->hash, s->proof, s->accepting, num_edges, - edges); + str[32] = '\0'; + str += 31; + for (i = 31; i >= 0; i--) + { + *str = (b & 1) + '0'; + str--; + b >>= 1; + } + break; + } + case AF_INET6: + { + struct in6_addr b = *(const struct in6_addr *) addr; - for (t = s->transitions_head; NULL != t; t = t->next) - iterate_edge (t->to_state, iterator, iterator_cls); + str[128] = '\0'; + str += 127; + for (i = 127; i >= 0; i--) + { + *str = (b.s6_addr[i / 8] & 1) + '0'; + str--; + b.s6_addr[i / 8] >>= 1; + } + break; + } } } /** - * Iterate over all edges starting from start state of automaton 'a'. Calling - * iterator for each edge. + * Get the ipv4 network prefix from the given 'netmask'. * - * @param a automaton. - * @param iterator iterator called for each edge. - * @param iterator_cls closure. + * @param netmask netmask for which to get the prefix len. + * + * @return length of ipv4 prefix for 'netmask'. + */ +static unsigned int +ipv4netmasktoprefixlen (const char *netmask) +{ + struct in_addr a; + unsigned int len; + uint32_t t; + + if (1 != inet_pton (AF_INET, netmask, &a)) + return 0; + len = 32; + for (t = htonl (~a.s_addr); 0 != t; t >>= 1) + len--; + return len; +} + + +/** + * Create a regex in 'rxstr' from the given 'ip' and 'netmask'. + * + * @param ip IPv4 representation. + * @param netmask netmask for the ip. + * @param rxstr generated regex, must be at least GNUNET_REGEX_IPV4_REGEXLEN + * bytes long. */ void -GNUNET_REGEX_iterate_all_edges (struct GNUNET_REGEX_Automaton *a, - GNUNET_REGEX_KeyIterator iterator, - void *iterator_cls) +GNUNET_REGEX_ipv4toregex (const struct in_addr *ip, const char *netmask, + char *rxstr) { - struct GNUNET_REGEX_State *s; + unsigned int pfxlen; - for (s = a->states_head; NULL != s; s = s->next) - s->marked = GNUNET_NO; + pfxlen = ipv4netmasktoprefixlen (netmask); + iptobinstr (AF_INET, ip, rxstr); + rxstr[pfxlen] = '\0'; + if (pfxlen < 32) + strcat (rxstr, "(0|1)+"); +} + + +/** + * Create a regex in 'rxstr' from the given 'ipv6' and 'prefixlen'. + * + * @param ipv6 IPv6 representation. + * @param prefixlen length of the ipv6 prefix. + * @param rxstr generated regex, must be at least GNUNET_REGEX_IPV6_REGEXLEN + * bytes long. + */ +void +GNUNET_REGEX_ipv6toregex (const struct in6_addr *ipv6, unsigned int prefixlen, + char *rxstr) +{ + iptobinstr (AF_INET6, ipv6, rxstr); + rxstr[prefixlen] = '\0'; + if (prefixlen < 128) + strcat (rxstr, "(0|1)+"); +} + + +struct RegexCombineCtx { + struct RegexCombineCtx *next; + struct RegexCombineCtx *prev; + + struct RegexCombineCtx *head; + struct RegexCombineCtx *tail; + + char *s; +}; + + +static char * +regex_combine (struct RegexCombineCtx *ctx) +{ + struct RegexCombineCtx *p; + size_t len; + char *regex; + char *tmp; + char *s; + + if (NULL != ctx->s) + GNUNET_asprintf (®ex, "%s(", ctx->s); + else + regex = GNUNET_strdup ("("); + GNUNET_log (GNUNET_ERROR_TYPE_DEBUG, "prefix: %s\n", regex); + + for (p = ctx->head; NULL != p; p = p->next) + { + s = regex_combine (p); + GNUNET_asprintf (&tmp, "%s%s|", regex, s); + GNUNET_free_non_null (s); + GNUNET_free_non_null (regex); + regex = tmp; + } + len = strlen (regex); + if (1 == len) + return GNUNET_strdup (""); + + GNUNET_log (GNUNET_ERROR_TYPE_DEBUG, "pre-partial: %s\n", regex); + if ('|' == regex[len - 1]) + regex[len - 1] = ')'; + if ('(' == regex[len - 1]) + regex[len - 1] = '\0'; - iterate_initial_edges (a, INITIAL_BITS, iterator, iterator_cls); - iterate_edge (a->start, iterator, iterator_cls); + GNUNET_log (GNUNET_ERROR_TYPE_DEBUG, "partial: %s\n", regex); + return regex; } + +static void +regex_add (struct RegexCombineCtx *ctx, const char *regex) +{ + struct RegexCombineCtx *p; + const char *rest; + + rest = ®ex[1]; + for (p = ctx->head; NULL != p; p = p->next) + { + if (p->s[0] == regex[0]) + { + if (1 == strlen(p->s)) + { + GNUNET_log (GNUNET_ERROR_TYPE_DEBUG, "common char %s\n", p->s); + GNUNET_log (GNUNET_ERROR_TYPE_DEBUG, "adding %s\n", rest); + regex_add (p, rest); + } + else + { + struct RegexCombineCtx *new; + new = GNUNET_malloc (sizeof (struct RegexCombineCtx)); + new->s = GNUNET_strdup (&p->s[1]); + GNUNET_log (GNUNET_ERROR_TYPE_DEBUG, " p has now %s\n", p->s); + GNUNET_log (GNUNET_ERROR_TYPE_DEBUG, " p will have %.1s\n", p->s); + GNUNET_log (GNUNET_ERROR_TYPE_DEBUG, " regex is %s\n", regex); + GNUNET_log (GNUNET_ERROR_TYPE_DEBUG, " new has now %s\n", new->s); + GNUNET_log (GNUNET_ERROR_TYPE_DEBUG, " rest is now %s\n", rest); + p->s[1] = '\0'; /* dont realloc */ + GNUNET_CONTAINER_DLL_insert (p->head, p->tail, new); + regex_add (p, rest); + } + return; + } + } + GNUNET_log (GNUNET_ERROR_TYPE_DEBUG, " no match\n"); + GNUNET_log (GNUNET_ERROR_TYPE_DEBUG, " new state %s\n", regex); + GNUNET_log (GNUNET_ERROR_TYPE_DEBUG, " under %s\n", ctx->s); + p = GNUNET_malloc (sizeof (struct RegexCombineCtx)); + p->s = GNUNET_strdup (regex); + GNUNET_CONTAINER_DLL_insert (ctx->head, ctx->tail, p); +} +/* +static void +debug (struct RegexCombineCtx *ctx, int lvl) +{ + struct RegexCombineCtx *p; + unsigned int i; + + for (i = 0; i < lvl; i++) fprintf (stderr, " "); + fprintf (stderr, "%s\n", ctx->s); + + for (p = ctx->head; NULL != p; p = p->next) + { + debug (p, lvl + 2); + } +}*/ + +char * +GNUNET_REGEX_combine (char * const regexes[]) +{ + unsigned int i; + char *combined; + const char *current; + struct RegexCombineCtx ctx; + + memset (&ctx, 0, sizeof (struct RegexCombineCtx)); + for (i = 0; regexes[i]; i++) + { + current = regexes[i]; + GNUNET_log (GNUNET_ERROR_TYPE_DEBUG, "Regex %u: %s\n", i, current); + regex_add (&ctx, current); + } + GNUNET_log (GNUNET_ERROR_TYPE_DEBUG, "\nCombining...\n"); + + combined = regex_combine (&ctx); + + return combined; +} + +/* end of regex.c */