Cases where a back-reference is part of a larger subexpression that
is quantified have never worked in Spencer's regex engine, because
he used a compile-time transformation that neglected the need to
check the back-reference match in iterations before the last one.
(That was okay for capturing parens, and we still do it if the
regex has *only* capturing parens ... but it's not okay for backrefs.)
To make this work properly, we have to add an "iteration" node type
to the regex engine's vocabulary of sub-regex nodes. Since this is a
moderately large change with a fair risk of introducing new bugs of its
own, apply to HEAD only, even though it's a fix for a longstanding bug.
Create src/backend/regex/README to hold an implementation overview of
the regex package, and fill it in with some preliminary notes about
the code's DFA/NFA processing and colormap management. Much more to
do there of course.
Also, improve some code comments around the colormap and cvec code.
No functional changes except to add one missing assert.
Tom Lane