Re: flag day for 64-bit?

Roland McGrath <[email protected]> writes:

> > I think the Right Thing is actually to have the argument remain off_t,
> > and change off_t to be 64 bits.
> 
> I don't know what definitions you are using these days.  But
> the term What We Are Doing is presently defined as judicious
> source compatibility with the GNU/Linux interface.  That is,
> the split 32/64 flavor with -D_FILE_OFFSET_BITS=32 the default.

This doesn't actually break source compatibility, but it does break
binary compatibility, AFAICT.  

> Since that is what we already have in libc (and have had
> since the first go-round on this when we decided this was
> the right thing to do and Mark Kettenis did all the
> necessary libc changes to give us the current libc ABI),
> your suggestion would break all ABIs and require recompiling
> everything in the system, not just libc and hurd-specific stuff.

Sure, but that's what I thought a flag day was. :)

Seriously, I think we should make the end-state as pretty as possible.
The "correct" end state is one which might not follow the broken Linux
definitions.  Still, compatibility (in that same end state) is also
important, and we'd like one day to have generic programs for the Hurd
and for Linux differ only in the library they link against.

Thomas