Unnecessary dependency on regex crate

[abgros]

I noticed that this project has a transitive dependency on the regex crate via arrow-string. Apparently the purpose of this is to allow you to run queries on a string column to see whether they match a regex. I am not using this functionality, yet regex is unconditionally pulled, resulting in bloated compile times and build sizes. Please put this behind a (perhaps default) feature flag so we aren't unnecessarily pulling this heavy dependency.

[abgros]

Not sure what label to add here, since this isn't strictly a bug but rather poor practice...

[jhorstmann]

Are you suggesting to split arrow-string into two separate crates, one with the sql LIKE and regular expression functions, and a separate one containing concat, length and substring functions?

[abgros]

@jhorstmann that seems like a good approach. My use case for this crate is limited to converting between Parquet files and other formats so I don't actually need anything at all besides the basic RecordBatch APIs.

[jhorstmann]

Do you have a dependency on the arrow crate? That crate is basically just a wrapper around several other crates. If you depend on arrow-array and parquet directly and with the default features, then there shouldn't be a dependency on regex.

[abgros]

@jhorstmann yep, it turned out that I can depend on the subcrates directly. Thanks for the help.