[Openib-windows] [RFC] changing kernel complib to be static l ibrary

[Fab Tillier]

> From: Leonid Keller [mailto:leonid at mellanox.co.il]
> Sent: Thursday, July 14, 2005 10:46 AM
> 
> -----Original Message-----
> From: Tzachi Dar
> > 
> > I disagree with most of the drawbacks that you have mentioned. Please note
> > that there is no code duplication, but rather using the same compiled code
> > a few times.
> [Leonid Keller] Yes, it's obvious. I was talking about a siuation, that for
> some reason only a part of the drivers get replaced and different
> communicating components are working each with its own version of COMPLIB

In the kernel, we should strive to eliminate any linkage interdependencies
between drivers.  All inter-driver communication should be performed using the
device interface methods, which do provide versioning.  If we want a complib
interface for drivers to query, we can do that.  I would prefer to see the
drivers migrate to using OS-native support where available, and slowly phase out
use of parts of complib that duplicate existing capabilities.

In user-mode, adding versioning would probably be useful, especially with
respect to the access layer.  Something similar to the WSAStartup mechanism but
without the backward compatibility baggage.  The side effects of a user-mode
application crashing are far less drastic than the kernel crashing at boot time,
so this isn't real high on the priority list.

- Fab