Re: primitive data types, was: ICE and TIPC

[Ralph Lange <Ralph.Lange@bessy.de>]

Kay-Uwe Kasemir wrote:

> The gateway, on the other hand, needs to understand the data,
> so that it can send updates to different clients using
> different deadbands.
> This mean we have to nail the basic types down.
>
> Or would the gateway only fan data out to multiple clients
> if they use the same subscription request (filter, events),
> so that the gateway can treat the data as an octet array?
My guess is that if the Gateway would keep a separate copy of the data 
for each different subscription request, it would end up being quite 
useless - or: much less useful.
It would have to keep multiple copies of the same data just because 
clients wanted a slightly different filter (e.g max update rate). Not 
efficient.
It would completely lose its functionality to shield the IOC from 
misbehaving clients: If e.g. a single client would subscribe to the same 
channel many times with a slightly changed max-update-rate, every single 
connection would lead to another subscription down to the IOC. Using 
that knowledge you could write a ten-liner that kills an IOC through a 
Gateway. Not good.
The Gateway needs to understand the data.
We have to nail the basic types down.

I would prefer the type info thing to be a struct (type, bitwidth, ...) 
rather than a large enum.
I would rather not call these types "network" types because they are 
mandatory for DataAccess even if there is no network involved. The 
network must support them, too - sure. Just as every other app or layer 
that uses DataAccess.
Ralph