I'm -1 on making a new primitive type in the memory layout spec [1]. +1 on clarifying [2], to indicate it is expected that the "Values array" for Utf8 and Binary types should never contain null elements.
[1] https://github.com/apache/arrow/blob/master/format/Layout.md [2] https://github.com/apache/arrow/blob/master/format/Message.fbs On Thu, Apr 28, 2016 at 3:08 PM, Wes McKinney <w...@cloudera.com> wrote: > Bumping this conversation. > > I'm +0 on making VARBINARY and String (identical VARBINARY but with a > UTF8 guarantee) primitive types in the spec. Let me know what others > think. > > Thanks > > On Fri, Apr 22, 2016 at 6:30 PM, Wes McKinney <w...@cloudera.com> wrote: >> On Fri, Apr 22, 2016 at 6:06 PM, Jacques Nadeau <jacq...@apache.org> wrote: >>> On Fri, Apr 22, 2016 at 2:42 PM, Wes McKinney <w...@cloudera.com> wrote: >>> >>>> On Fri, Apr 22, 2016 at 4:56 PM, Micah Kornfield <emkornfi...@gmail.com> >>>> wrote: >>>> > I like the current scheme of making String (UTF8) a primitive type in >>>> > regards to RPC but not modeling it as a special Array type. I think >>>> > the key is formally describing how logical types map to physical types >>>> > either is the Flatbuffer schema or in a separate document. >>>> > >>>> > I think there are two use-cases here: >>>> > 1. Reconstructing Array's off the wire. >>>> > 2. Writing algorithms/builders to deal with specific logical types >>>> > built on Arrays. >>>> > >>>> > For case 1, I think it is simpler to not special case string types as >>>> > primitives. Understanding that a logical String type maps to a >>>> > List<Utf8> should be sufficient and allows us to re-use the >>>> > serialization code for ListArrays for these types. >>>> > >>>> >>>> It is simpler for the IPC serde code-path. I'll let Jacques comment >>>> but one downside of having strings as a nested type is that there are >>>> certain code paths (for example: Parquet-related) which deal with the >>>> flat table case. To make a Parquet analogy, there is the special >>>> BYTE_ARRAY primitive type, even though you could technically represent >>>> variable-length binary data using a repeated field and using >>>> repetition/definition levels (but the encoding/decoding overhead for >>>> this in Parquet is much more significant than Arrow). There may be >>>> other reasons. >>>> >>> >>> I'm a bit confused about what everyone means. I didn't actually realize >>> that this [1] had been merged yet but I'm generally on board with how it is >>> constructed. >>> >>> With regards to the c++ implementation of the items at [1], abstracting >>> shared physical representations out seems fine to me but I don't think we >>> should necessitate effective 3NF for [1]. >>> >>> One of the key points that I'm focused on in the Java space is that I'd >>> like to move to an always nullable pattern. This is vastly simplifying from >>> a code generation, casting and complexity perspective and is a nominal cost >>> when using column execution. If binary and varchar are primitive types as >>> there there is no weird special casing of avoiding the nullability bitmap >>> in the case of variable width items (for the offsets). But that is an >>> implementation detail of the Java library. >>> >>> So in general, I like the scheme at [1] for the concepts that we all are >>> talking about (as opposed to eliminating lines 67 & 68) >>> >>> [1] https://github.com/apache/arrow/blob/master/format/Message.fbs >>> >> >> Well, the issue is that mapping of metadata onto memory layout for IPC >> purposes, at least. You can use the List code path for arbitrary List >> types as well as strings and binary. It sounds like either way on the >> Java side you're going to collapse UTF8 / BINARY into a primitive so >> that you don't have to manage a separate never-used bitmap for the >> string/binary data. It seems useful enough to me to have a primitive >> variable-length binary/UTF8 type but I do not feel strongly about it. >> >>> >>> >>>> > For case 2, it would be nice to utilize the type system of the host >>>> > programming language to express the semantics of a function call (e.g. >>>> > ParseString(StringArray strings) vs ParseString(ListArray strings), >>>> > but I think this can be implemented without requiring a new primitive >>>> > type in the spec. >>>> > >>>> > The more interesting thing to me is if we should have a new primitive >>>> > type for fixed length lists (e.g. the logical type CHAR). The >>>> > offsets array isn't necessary in this case for random access. >>>> > >>>> > Also, the way the VARCHAR types (based on a comment in the C++ >>>> > (https://github.com/apache/arrow/blob/master/cpp/src/arrow/type.h#L63) >>>> > are currently described as a null terminated UTF8 is problematic. I >>>> > believe null bytes are valid UTF8 characters. >>>> >>>> >>>> > >>>> >>>> Good point, sorry about that. We probably would need to length-prefix >>>> the values, then. >>>> >>> >>> >>> Is this an input/output interface? Arrow structures should all be 4 byte >>> offset based and be neither length prefixed nor null terminated. >> >> This was a question around the VARCHAR(k) type (which in many >> databases is distinct from a TEXT type in which any value can be >> arbitrary length). So if you have a VARCHAR(50), you guarantee that no >> value exceeds 50 characters. In Arrow I suppose this is just metadata >> because you have the offsets encoding length (pardon the jet lag). >> Micah -- I think we can nix the `VarcharType` in the C++ code, >> leftovers from my earliest draft implementation. >> >> - Wes
