While writing code for handling fixed size structs in flatbuffers I had to make a choice between defereed access (reading field on demand) or reading all fields at once.

I’ve noticed that prepending to bytearray is ~50 times slower than preallocating space and overwriting data directly.