[
https://issues.apache.org/jira/browse/FLINK-3318?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15839492#comment-15839492
]
ASF GitHub Bot commented on FLINK-3318:
---------------------------------------
Github user kl0u commented on the issue:
https://github.com/apache/flink/pull/2361
Hi @mushketyk ! Thanks a lot!
The problem is in the `NFACompiler` and more specifically the way the
pattern is compiled. I have not come up with a concrete solution yet but I
would say that:
i) there should be another `StateType` as well, sth like `PotentialFinal`
(or a better name) and in the case that the final pattern is optional, then it
gets assigned this `StateType`. This also propagates backwards in the pattern
graph in the case that the final pattern is optional (0 to sth times).
ii) the `addTransitions()` should also be extended to account for
quantifiers at the end and at the start of the pattern when "unrolling" or
expanding the original pattern, e.g. the `expandRepeatingPattern()` should not
always assign `State.StateType.Normal` to the unrolled patterns as they may be
at the end or the start.
iii) the `NFA` should also be modified to account for the new
`PotentialFinal` state. In this case, when we see such a pattern, a flag should
be set, and for example in the one-to-many case the pattern should be matched
when the next non-matching element comes.
Again these are some initial thoughts. We can discuss more on the design if
you want, as this seems to be a bigger change than expected ;)
> Add support for quantifiers to CEP's pattern API
> ------------------------------------------------
>
> Key: FLINK-3318
> URL: https://issues.apache.org/jira/browse/FLINK-3318
> Project: Flink
> Issue Type: Improvement
> Components: CEP
> Affects Versions: 1.0.0
> Reporter: Till Rohrmann
> Assignee: Ivan Mushketyk
> Priority: Minor
>
> It would be a good addition to extend the pattern API to support quantifiers
> known from regular expressions (e.g. Kleene star, ?, +, or count bounds).
> This would considerably enrich the set of supported patterns.
> Implementing the count bounds could be done by unrolling the pattern state.
> In order to support the Kleene star operator, the {{NFACompiler}} has to be
> extended to insert epsilon-transition between a Kleene start state and the
> succeeding pattern state. In order to support {{?}}, one could insert two
> paths from the preceding state, one which accepts the event and another which
> directly goes into the next pattern state.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
