Storage in a high-level extensible distributed persistent
system is a very tricky problem:
This problem is actually a specific instance of the problems tackled
in the Migration subproject.
But here we document the choices made in our LLL.
A first, original version should be written using widely
available existing tools, and not bother about speed efficiency of
generated code. It can use dirty hacks of all kinds.
The definitive version will be written using the own Tunes
language system; it will be as clean as can be, and rely on all our usual
Tunes tools to produce
"optimized" code (oh I hate this
word).
It should map as well as possible our current targets as well as
future ones.