This language was also discussed in https://github.com/apache/arrow/pull/1959
We should try to be consistent about terminology in all implementations: * A buffer is a contiguous block of memory without any defined semantics (it's just bytes) * An array or vector (we've been saying "array" in C++ because "vector" has a size-mutability connotation there) is a logical 1D sequence of values, represented by an arbitrary number of buffers - Wes On Thu, May 3, 2018 at 8:35 AM, Antoine Pitrou <anto...@python.org> wrote: > > Le 03/05/2018 à 14:28, Andy Grove a écrit : >> Now I have even more questions. >> >> List<List<T>> uses one values buffer for the instances of T (according to >> Layout.md). >> >> List<Struct> couldn't possibly do that. Is that a valid type in Arrow? > > I think you misread the layout document: it says a list array has > """A values array, a child array of type T. T may also be a nested > type""". Note: child array, not buffer. > > So, yes, List<Struct> is a valid type. A List<Struct> array has: > * one null bitmap buffer (optional) to designate null lists > * one offsets buffer for list offsets > * one child struct array (which has an arbitrary number of buffers > depending on the struct type) > > Regards > > Antoine.
