[PATCH, RFC] Enable IBM long double for PPC32 Linux

[David Edelsohn]

>>>>> Mark Mitchell writes:

Mark> So, the unfortunate thing here is that we (the GCC maintainers and the
Mark> GLIBC maintainers) failed to coordinate well; ideally, the GLIBC
Mark> maintainers would have told us they wanted to have the 128-bit long
Mark> double support in GCC 4.1 some while ago, so that we could have
Mark> incoroprated that into the GCC release plans.  Or, perhaps this
Mark> requirement just popped up today, and it just happens that we're
Mark> unluckily late in the GCC release cycle.  Anyhow, it would certainly be
Mark> good if in future we can try to avoid the schedule issue that we
Mark> have now. 

	From my vantage point, this has been a long-requested feature of
the various GLIBC ports, but the window of opportunity to implement the
change only appeared a little over a week ago.

Mark> I think the key consideration for me is: how confident are we that the
Mark> current patches will not affect people who do not use the option?  If we
Mark> are convinced that the 128-bit long double support only affects people
Mark> who use the flag, and the flag is off by default, then it seems like the
Mark> benefits of incorporation in 4.1 outweigh the risks.  Have their been
Mark> cases reported of the patches causing problems for people not using the
Mark> flag, whether or not those problems have been subsequently fixed?

	The patches mostly are confined to the various architecture ports.
There have not been any reports of problems from either the affected
targets or other targets in the same ports.  Jakub extensively tested PPC
Linux last week and has been testing zSeries Linux this weekend.

	We do not live in an ideal world and need to try to accomodate
dynamic schedules when possible.

David