1/*-------------------------------------------------------------------------
4 * Extract a common prefix, if any, from a compiled regex.
7 * Portions Copyright (c) 2012-2025, PostgreSQL Global Development Group
8 * Portions Copyright (c) 1998, 1999 Henry Spencer
11 * src/backend/regex/regprefix.c
13 *-------------------------------------------------------------------------
20 * forward declarations
23 chr *
string,
size_t *slength);
27 * pg_regprefix - get common prefix for regular expression
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
35 * In the non-failure cases, *string is set to a palloc'd string containing
36 * the common prefix or exact value, of length *slength (measured in chrs
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.
55 if (
string == NULL || slength == NULL)
57 *
string = NULL;
/* initialize for failure cases */
59 if (re == NULL || re->re_magic !=
REMAGIC)
61 if (re->re_csize !=
sizeof(
chr))
64 /* Initialize locale-dependent support */
68 g = (
struct guts *) re->re_guts;
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.
80 /* matchall NFAs never have a fixed prefix */
85 * Since a correct NFA should never contain any exit-free loops, it should
86 * not be possible for our traversal to return to a previously visited NFA
87 * state. Hence we need at most nstates chrs in the output string.
96 assert(*slength <= cnfa->nstates);
110 * findprefix - extract common prefix from cNFA
112 * Results are returned into the preallocated chr array string[], with
113 * *slength (which must be preset to zero) incremented for each chr.
115static int /* regprefix return code */
128 * The "pre" state must have only BOS/BOL outarcs, else pattern isn't
129 * anchored left. If we have both BOS and BOL, they must go to the same
140 else if (nextst != ca->
to)
150 * Scan through successive states, stopping as soon as we find one with
151 * more than one acceptable transition character (either multiple colors
152 * on out-arcs, or a color with more than one member chr).
154 * We could find a state with multiple out-arcs that are all labeled with
155 * the same singleton color; this comes from patterns like "^ab(cde|cxy)".
156 * In that case we add the chr "c" to the output string but then exit the
157 * loop with nextst == -1. This leaves a little bit on the table: if the
158 * pattern is like "^ab(cde|cdy)", we won't notice that "d" could be added
159 * to the prefix. But chasing multiple parallel state chains doesn't seem
169 /* We can ignore BOS/BOL arcs */
174 * ... but EOS/EOL arcs terminate the search, as do RAINBOW arcs
185 /* First plain outarc */
189 else if (thiscolor == ca->
co)
191 /* Another plain outarc for same color */
196 /* More than one plain outarc color terminates the search */
201 /* Done if we didn't find exactly one color on plain outarcs */
204 /* The color must be a singleton */
207 /* Must not have any high-color-map entries */
212 * Identify the color's sole member chr and add it to the prefix
213 * string. In general the colormap data structure doesn't provide a
214 * way to find color member chrs, except by trying GETCOLOR() on each
215 * possible chr value, which won't do at all. However, for the cases
216 * we care about it should be sufficient to test the "firstchr" value,
217 * that is the first chr ever added to the color. There are cases
218 * where this might no longer be a member of the color (so we do need
219 * to test), but none of them are likely to arise for a character that
220 * is a member of a common prefix. If we do hit such a corner case,
221 * we just fall out without adding anything to the prefix string.
227 string[(*slength)++] =
c;
229 /* Advance to next state, but only if we have a unique next state */
230 }
while (nextst != -1);
233 * If we ended at a state that only has EOS/EOL outarcs leading to the
234 * "post" state, then we have an exact-match string. Note this is true
235 * even if the string is of zero length.
244 else if (nextst != ca->
to)
260 * Otherwise, if we were unable to identify any prefix characters, say
261 * NOMATCH --- the pattern is anchored left, but doesn't specify any
262 * particular first character.
if(TABLE==NULL||TABLE_index==NULL)
void pg_set_regex_collation(Oid collation)
int pg_regprefix(regex_t *re, chr **string, size_t *slength)
static int findprefix(struct cnfa *cnfa, struct colormap *cm, chr *string, size_t *slength)