Hi,
Arrays of non-nillable types are represented as arrays of pointers to those
types. This is true for arrays of primitives and arrays of complex types. I
discovered this by looking at the AxisBench and PrimitiveAndArray tests.
The problem with this is that it appears more complicated than it needs to
be, but it makes the code the same for nillable and non-nillable.

Are people OK with this as the long-term solution?

If so, then the generated code in the deserializer that extends an array by
new'ing one twice the size, memcpying it, and deleting the old one is
redundant now because array elements are no longer contiguous.

Comments?
I'm away for a week or so, so will not respond immediately to any
discussion,
Mark
Mark Whitlock
IBM

Reply via email to