On 08/08/2012 22:58, Xueming Shen wrote:
Hi

It appears the optimization (flatten a binary tree structure into a branch/
switch) we put into JDK6 for alternation operation [1] has problem if the
first construct is a group "(...)" followed by a greedy/reluctant "once or not at all" quantifier. For example regex "(a)?bc|d" or "(a)??bc|d" can't be found
for input "d", while "a?bc|d" or "a??bc|d" just works fine.

The root cause is that expr() mistakenly to use the Branch node from the
sub-expression (in above case, the branch node for "(a)?" instead of using its own Branch node (because the incorrect use of "prev instance of Branch",
see the diff), which basically messes up the internal regex node tree [2]

The webrev is at
http://cr.openjdk.java.net/~sherman/7189363/webrev
This looks okay to me too. I'm also surprised it hasn't been noticed before now.

-Alan.

Reply via email to