[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: [rfc] alpha-linux changing to 128-bit long double



At Fri, 12 Mar 2004 11:36:58 -0500,
Daniel Jacobowitz wrote:
> On Fri, Mar 12, 2004 at 09:30:51PM +0900, GOTO Masanori wrote:
> > At Fri, 12 Mar 2004 10:44:05 +0000,
> > James Troup wrote:
> > > >> Not to mention it was for all architectures - I have no idea how
> > > >> this could be sanely done for a single architecture.  Remember you
> > > >> have to rename all the library packages...
> > > >
> > > > This transition is only for alpha.
> > > 
> > > You still can't do a soname transition for libc without changing the
> > > name of every library package which means changing the source which
> > > means it'll affect all architectures.
> > 
> > Uhm, exactly - I forgot that point.  It becomes big transition not
> > only alpha.
> > 
> > So after sarge, we have at least 3 big transition: (1) alpha libc6.2
> > transition (2) m68k/hppa gcc 3.4 (sjlj -> dwarf2) transition (3)
> > biarch support.  Our life will be filled with difficulties ;)
> 
> Actually, I don't think we need to convert the sonames of anything.
> Unlike libdb, for instance, libc.so.6.1 and libc.so.6.2 can in theory
> coexist in the same binary.  I'm sure something somewhere will get
> confused, but the point of symbol versioning is to make this possible.
> 
> Now, of course, there's a whole separate problem for any _other_
> library whose interface uses long doubles.

Yes, it can apply symbol versioning, so it may not become problem
seriously.  BTW, interestingly, it seems there is another way to
handle for this.

	http://sources.redhat.com/ml/libc-alpha/2004-03/msg00241.html

Regards,
-- gotom



Reply to: