[openib-general] Re: [PATCH] sockaddr_ll change for IPoIB interface

[Tom Duffy]

On Aug 12, 2005, at 9:07 AM, Roland Dreier wrote:

>     Tom> But, Fedora will rebuild their binary once this change is in.
>     Tom> If the Linux developers cared about this sort of thing, it
>     Tom> would version all its kernel structs and put padding at the
>     Tom> end to ensure new fields could be added.  It has opted for
>     Tom> the cleaner (technical) solution of having all the apps
>     Tom> recompile.  Sure there will be a little bit of growing pain,
>     Tom> but in the end, it won't have all kinds of backwards
>     Tom> compatibility cruft lying around.
>
> No, this is absolutely not true.  The kernel-user ABI is very stable,
> and with very few exceptions, you should be able to take binaries that
> worked on kernel 1.0 and run them on a modern kernel.  For example,
> <http://lkml.org/lkml/2004/10/13/196>
>
> The in-kernel ABI and API can and do change all the time, but that's a
> different story.

I don't want to get into a big debate about this.  If a good solution  
can be had that will both maintain compatibility and allow for IB, I  
would welcome that.  On the other hand, most of the interesting apps  
have broken on Linux in the past few years.  Some examples:

- Loki games
- Word Perfect 8
- Crossover office/plugin
- java

I know that lots of that has to do with gcc, threading, or glibc  
instability, but clearly most interesting binaries that were around  
in the 1.0 days will not run on todays stuff.

Can we do an audit of what stuff will break with this change?  If it  
is a handful of applications that we all have the source to, maybe it  
won't be that big of a deal.

Maybe the better approach is to simply submit the struct change.  And  
let the maintainers object if they want ABI stability.   If they do,  
ask them for an elegant solution ;)

-tduffy