I think the current compute functions for casting from other String types can be improved in several areas. I elaborate on it here <https://github.com/apache/arrow/issues/46128>.
I give a summary here 1- The current cast compute functions consider offset+length elements for view buffers. However, it is possible to reduce the allocation to length. Although the bitmap buffer is copied, I think it's acceptable overhead to prevent larger allocation 2- Overflowing is checked after buffers are allocated, however, it's possible to check before the allocation Aside from that, it's possible to create a utf8() type string with more than max int32_t in casting from StringView types to utf8(), binary() types. 3- The current algorithms consider large strings with more than max int32_t data(length of each element less than max int32_t) as invalid casting to utf8_view() type. I think it's possible to have a simple implementation(at least for correctness) 4-The current implementation may cause memory bloat. Is it worthwhile to implement a garbage collector to address this? 5- one of the key features of StringView types is to prevent duplication. C asting from FixedSizeBinary to BinaryView types is a good case to avoid duplication.
