| 1 | /*------------------------------------------------------------------------- |
| 2 | * |
| 3 | * regprefix.c |
| 4 | * Extract a common prefix, if any, from a compiled regex. |
| 5 | * |
| 6 | * |
| 7 | * Portions Copyright (c) 2012-2019, PostgreSQL Global Development Group |
| 8 | * Portions Copyright (c) 1998, 1999 Henry Spencer |
| 9 | * |
| 10 | * IDENTIFICATION |
| 11 | * src/backend/regex/regprefix.c |
| 12 | * |
| 13 | *------------------------------------------------------------------------- |
| 14 | */ |
| 15 | |
| 16 | #include "regex/regguts.h" |
| 17 | |
| 18 | |
| 19 | /* |
| 20 | * forward declarations |
| 21 | */ |
| 22 | static int findprefix(struct cnfa *cnfa, struct colormap *cm, |
| 23 | chr *string, size_t *slength); |
| 24 | |
| 25 | |
| 26 | /* |
| 27 | * pg_regprefix - get common prefix for regular expression |
| 28 | * |
| 29 | * Returns one of: |
| 30 | * REG_NOMATCH: there is no common prefix of strings matching the regex |
| 31 | * REG_PREFIX: there is a common prefix of strings matching the regex |
| 32 | * REG_EXACT: all strings satisfying the regex must match the same string |
| 33 | * or a REG_XXX error code |
| 34 | * |
| 35 | * In the non-failure cases, *string is set to a malloc'd string containing |
| 36 | * the common prefix or exact value, of length *slength (measured in chrs |
| 37 | * not bytes!). |
| 38 | * |
| 39 | * This function does not analyze all complex cases (such as lookaround |
| 40 | * constraints) exactly. Therefore it is possible that some strings matching |
| 41 | * the reported prefix or exact-match string do not satisfy the regex. But |
| 42 | * it should never be the case that a string satisfying the regex does not |
| 43 | * match the reported prefix or exact-match string. |
| 44 | */ |
| 45 | int |
| 46 | pg_regprefix(regex_t *re, |
| 47 | chr **string, |
| 48 | size_t *slength) |
| 49 | { |
| 50 | struct guts *g; |
| 51 | struct cnfa *cnfa; |
| 52 | int st; |
| 53 | |
| 54 | /* sanity checks */ |
| 55 | if (string == NULL || slength == NULL) |
| 56 | return REG_INVARG; |
| 57 | *string = NULL; /* initialize for failure cases */ |
| 58 | *slength = 0; |
| 59 | if (re == NULL || re->re_magic != REMAGIC) |
| 60 | return REG_INVARG; |
| 61 | if (re->re_csize != sizeof(chr)) |
| 62 | return REG_MIXED; |
| 63 | |
| 64 | /* Initialize locale-dependent support */ |
| 65 | pg_set_regex_collation(re->re_collation); |
| 66 | |
| 67 | /* setup */ |
| 68 | g = (struct guts *) re->re_guts; |
| 69 | if (g->info & REG_UIMPOSSIBLE) |
| 70 | return REG_NOMATCH; |
| 71 | |
| 72 | /* |
| 73 | * This implementation considers only the search NFA for the topmost regex |
| 74 | * tree node. Therefore, constraints such as backrefs are not fully |
| 75 | * applied, which is allowed per the function's API spec. |
| 76 | */ |
| 77 | assert(g->tree != NULL); |
| 78 | cnfa = &g->tree->cnfa; |
| 79 | |
| 80 | /* |
| 81 | * Since a correct NFA should never contain any exit-free loops, it should |
| 82 | * not be possible for our traversal to return to a previously visited NFA |
| 83 | * state. Hence we need at most nstates chrs in the output string. |
| 84 | */ |
| 85 | *string = (chr *) MALLOC(cnfa->nstates * sizeof(chr)); |
| 86 | if (*string == NULL) |
| 87 | return REG_ESPACE; |
| 88 | |
| 89 | /* do it */ |
| 90 | st = findprefix(cnfa, &g->cmap, *string, slength); |
| 91 | |
| 92 | assert(*slength <= cnfa->nstates); |
| 93 | |
| 94 | /* clean up */ |
| 95 | if (st != REG_PREFIX && st != REG_EXACT) |
| 96 | { |
| 97 | FREE(*string); |
| 98 | *string = NULL; |
| 99 | *slength = 0; |
| 100 | } |
| 101 | |
| 102 | return st; |
| 103 | } |
| 104 | |
| 105 | /* |
| 106 | * findprefix - extract common prefix from cNFA |
| 107 | * |
| 108 | * Results are returned into the preallocated chr array string[], with |
| 109 | * *slength (which must be preset to zero) incremented for each chr. |
| 110 | */ |
| 111 | static int /* regprefix return code */ |
| 112 | findprefix(struct cnfa *cnfa, |
| 113 | struct colormap *cm, |
| 114 | chr *string, |
| 115 | size_t *slength) |
| 116 | { |
| 117 | int st; |
| 118 | int nextst; |
| 119 | color thiscolor; |
| 120 | chr c; |
| 121 | struct carc *ca; |
| 122 | |
| 123 | /* |
| 124 | * The "pre" state must have only BOS/BOL outarcs, else pattern isn't |
| 125 | * anchored left. If we have both BOS and BOL, they must go to the same |
| 126 | * next state. |
| 127 | */ |
| 128 | st = cnfa->pre; |
| 129 | nextst = -1; |
| 130 | for (ca = cnfa->states[st]; ca->co != COLORLESS; ca++) |
| 131 | { |
| 132 | if (ca->co == cnfa->bos[0] || ca->co == cnfa->bos[1]) |
| 133 | { |
| 134 | if (nextst == -1) |
| 135 | nextst = ca->to; |
| 136 | else if (nextst != ca->to) |
| 137 | return REG_NOMATCH; |
| 138 | } |
| 139 | else |
| 140 | return REG_NOMATCH; |
| 141 | } |
| 142 | if (nextst == -1) |
| 143 | return REG_NOMATCH; |
| 144 | |
| 145 | /* |
| 146 | * Scan through successive states, stopping as soon as we find one with |
| 147 | * more than one acceptable transition character (either multiple colors |
| 148 | * on out-arcs, or a color with more than one member chr). |
| 149 | * |
| 150 | * We could find a state with multiple out-arcs that are all labeled with |
| 151 | * the same singleton color; this comes from patterns like "^ab(cde|cxy)". |
| 152 | * In that case we add the chr "c" to the output string but then exit the |
| 153 | * loop with nextst == -1. This leaves a little bit on the table: if the |
| 154 | * pattern is like "^ab(cde|cdy)", we won't notice that "d" could be added |
| 155 | * to the prefix. But chasing multiple parallel state chains doesn't seem |
| 156 | * worth the trouble. |
| 157 | */ |
| 158 | do |
| 159 | { |
| 160 | st = nextst; |
| 161 | nextst = -1; |
| 162 | thiscolor = COLORLESS; |
| 163 | for (ca = cnfa->states[st]; ca->co != COLORLESS; ca++) |
| 164 | { |
| 165 | /* We can ignore BOS/BOL arcs */ |
| 166 | if (ca->co == cnfa->bos[0] || ca->co == cnfa->bos[1]) |
| 167 | continue; |
| 168 | /* ... but EOS/EOL arcs terminate the search, as do LACONs */ |
| 169 | if (ca->co == cnfa->eos[0] || ca->co == cnfa->eos[1] || |
| 170 | ca->co >= cnfa->ncolors) |
| 171 | { |
| 172 | thiscolor = COLORLESS; |
| 173 | break; |
| 174 | } |
| 175 | if (thiscolor == COLORLESS) |
| 176 | { |
| 177 | /* First plain outarc */ |
| 178 | thiscolor = ca->co; |
| 179 | nextst = ca->to; |
| 180 | } |
| 181 | else if (thiscolor == ca->co) |
| 182 | { |
| 183 | /* Another plain outarc for same color */ |
| 184 | nextst = -1; |
| 185 | } |
| 186 | else |
| 187 | { |
| 188 | /* More than one plain outarc color terminates the search */ |
| 189 | thiscolor = COLORLESS; |
| 190 | break; |
| 191 | } |
| 192 | } |
| 193 | /* Done if we didn't find exactly one color on plain outarcs */ |
| 194 | if (thiscolor == COLORLESS) |
| 195 | break; |
| 196 | /* The color must be a singleton */ |
| 197 | if (cm->cd[thiscolor].nschrs != 1) |
| 198 | break; |
| 199 | /* Must not have any high-color-map entries */ |
| 200 | if (cm->cd[thiscolor].nuchrs != 0) |
| 201 | break; |
| 202 | |
| 203 | /* |
| 204 | * Identify the color's sole member chr and add it to the prefix |
| 205 | * string. In general the colormap data structure doesn't provide a |
| 206 | * way to find color member chrs, except by trying GETCOLOR() on each |
| 207 | * possible chr value, which won't do at all. However, for the cases |
| 208 | * we care about it should be sufficient to test the "firstchr" value, |
| 209 | * that is the first chr ever added to the color. There are cases |
| 210 | * where this might no longer be a member of the color (so we do need |
| 211 | * to test), but none of them are likely to arise for a character that |
| 212 | * is a member of a common prefix. If we do hit such a corner case, |
| 213 | * we just fall out without adding anything to the prefix string. |
| 214 | */ |
| 215 | c = cm->cd[thiscolor].firstchr; |
| 216 | if (GETCOLOR(cm, c) != thiscolor) |
| 217 | break; |
| 218 | |
| 219 | string[(*slength)++] = c; |
| 220 | |
| 221 | /* Advance to next state, but only if we have a unique next state */ |
| 222 | } while (nextst != -1); |
| 223 | |
| 224 | /* |
| 225 | * If we ended at a state that only has EOS/EOL outarcs leading to the |
| 226 | * "post" state, then we have an exact-match string. Note this is true |
| 227 | * even if the string is of zero length. |
| 228 | */ |
| 229 | nextst = -1; |
| 230 | for (ca = cnfa->states[st]; ca->co != COLORLESS; ca++) |
| 231 | { |
| 232 | if (ca->co == cnfa->eos[0] || ca->co == cnfa->eos[1]) |
| 233 | { |
| 234 | if (nextst == -1) |
| 235 | nextst = ca->to; |
| 236 | else if (nextst != ca->to) |
| 237 | { |
| 238 | nextst = -1; |
| 239 | break; |
| 240 | } |
| 241 | } |
| 242 | else |
| 243 | { |
| 244 | nextst = -1; |
| 245 | break; |
| 246 | } |
| 247 | } |
| 248 | if (nextst == cnfa->post) |
| 249 | return REG_EXACT; |
| 250 | |
| 251 | /* |
| 252 | * Otherwise, if we were unable to identify any prefix characters, say |
| 253 | * NOMATCH --- the pattern is anchored left, but doesn't specify any |
| 254 | * particular first character. |
| 255 | */ |
| 256 | if (*slength > 0) |
| 257 | return REG_PREFIX; |
| 258 | |
| 259 | return REG_NOMATCH; |
| 260 | } |
| 261 | |