Re: [patches] Including IBM add-ons in EGLIBC?

["Joseph S. Myers" <joseph@xxxxxxxxxxxxxxxx>]

On Fri, 17 Nov 2006, Steve Papacharalambous wrote:

> If there is a directory for the e500 in ports I don't see much benefit
> in separating out the SPE PIM functions, on balance it would be better
> to keep them together.

There are two compatibility issues here:

* Compatibility with FSF glibc.  New functions like this should generally 
only be added to the EGLIBC ABI and API when accepted for FSF glibc, and 
then at the symbol version for which they are accepted in FSF glibc.  
Now, this a new port, and new ports can have new symbols, and new ports 
can reasonably be added to EGLIBC first - in which case the question is 
whether there'd be any objection to these symbols when the port is 
submitted for the ports collection of FSF glibc.

* Compatibility with users of the SPE add-on.  They will expect these 
functions to be found in libspe.  Is it OK to move them to libc and cease 
to provide a separate libspe?  I'd think so; note that the SPE add-on 
provides just libspe.so rather than setting a proper SONAME, which clearly 
would need to be fixed if the add-on stays.

On the whole I think putting everything in libc is reasonable as long as 
FSF glibc either excludes the port or includes it complete with the SPE 
PIM functions - but if FSF glibc includes it without the functions, then 
for compatibility they'll need to go elsewhere.

-- 
Joseph S. Myers
joseph@xxxxxxxxxxxxxxxx