]> granicus.if.org Git - postgresql/commit
Fix regexport.c to behave sanely with lookaround constraints.
authorTom Lane <tgl@sss.pgh.pa.us>
Thu, 13 Apr 2017 21:18:35 +0000 (17:18 -0400)
committerTom Lane <tgl@sss.pgh.pa.us>
Thu, 13 Apr 2017 21:18:35 +0000 (17:18 -0400)
commit67665a71c003d0bc4d261f134cbc0989effc6927
tree7fe205b7a93cf1f31114c2b031f182c0106deaa2
parentbcb1a27dc039f175dc64a231a742e74a8728fb26
Fix regexport.c to behave sanely with lookaround constraints.

regexport.c thought it could just ignore LACON arcs, but the correct
behavior is to treat them as satisfiable while consuming zero input
(rather reminiscently of commit 9f1e642d5).  Otherwise, the emitted
simplified-NFA representation may contain no paths leading from initial
to final state, which unsurprisingly confuses pg_trgm, as seen in
bug #14623 from Jeff Janes.

Since regexport's output representation has no concept of an arc that
consumes zero input, recurse internally to find the next normal arc(s)
after any LACON transitions.  We'd be forced into changing that
representation if a LACON could be the last arc reaching the final
state, but fortunately the regex library never builds NFAs with such
a configuration, so there always is a next normal arc.

Back-patch to 9.3 where this logic was introduced.

Discussion: https://postgr.es/m/20170413180503.25948.94871@wrigleys.postgresql.org
contrib/pg_trgm/expected/pg_trgm.out
contrib/pg_trgm/sql/pg_trgm.sql
src/backend/regex/regexport.c