Fix LIKE escape issues

[Dandandan]

Due to using regex crate directly, it's sensitive to not escape like patterns. A solution to it can be using the escapefunction inregex`: https://docs.rs/regex/1.5.4/regex/fn.escape.html

Next, there should be a possibility to escape the % and _ characters.

Arrow-rs issues/PRs:

apache/arrow-rs#1085
apache/arrow-rs#1042
apache/arrow-rs#415

[jorgecarleitao]

I struggle a bit to implement this because the rules are quite specific to the particular dialect. E.g. databend uses Mysql, Datafusion uses postgres, Polars uses python's regex.

Is there an escape that covers them all? Any ideas how we accommodate these differences, @Dandandan @houqp @ritchie46 @sundy-li ?

[Dandandan]

I think it's best to split this into two:

• (Bug) not escaping regex patterns like .* ., parentheses, etc. That should be easy as the regex crate offers the escape method.
• (Feature) supporting escaping % and _, e.g. with something like \% and \_, or making it configurable.

[sundy-li]

databend built regexp pattern by like pattern using escape.
https://github.com/datafuselabs/databend/blob/44bab57d297c0f05440014a5270318bdb4485d8e/common/datavalues/src/arrays/ops/like.rs

Maybe we can introduce LIKE_OPTIONS to cover possible dialects in arrow2.

[daniel-martinez-maqueda-sap]

I have a proposal I've been working on that could work:

• Escape regex patterns calling escape from the crate regex as it is done at the moment
• If a like wildcard is escaped, remove the escape characters (so that it is matched when running the regex)

[jorgecarleitao]

Really sorry for the late reply! :(

Those makes a lot of sense.