I considered doing endianness swaps as part of the extract calls, instead of at data load time. I've put it off for now because (a) my personal use cases so far almost always use all properties in each element anyway, so I'd still end up converting all the data anyway; and (b) big-endian files are pretty rare in practice. If you have a use case where this might help, please let me know!
- I have a use case for this now. The ply_io_benchmark tests only extract vertex positions, ignoring any additional vertex attributes. Doing the conversion on demand only will help improve miniply's performance in the benchmarks! (And also for that use case too, of course. But... benchmarks!)
Reading variable-size properties (aka lists) is pretty slow. If we know in
advance that all rows for a list property always have the same size, we can
improve performance massively by replacing the list property with a set of
fixed-size properties. That's what the
PLYElement::convert_list_to_fixed_size() method is for.