Re: prefix btree implementation

Hello all,

I also was examining a similar compression method just.

Qingqing Zhou wrote:
> We can find a way to handle the above case, but it is better to find a 
> general way to handle any data types(include UDT). "Each type optionally 
> provide the required routines" could be a way, more details?

How about the use of difference information with "High key"?
Because "High key" information exists also in the route page, I think
that it seems to be able to use it well. (e.g. new tuple insertion)
# There is a problem that in the rightmost page, "High key" is not...

My consideration was just started, whether it goes well really has not
been understood yet.

---
Junji Teramoto