Re: [PATCHv3 1/4] ldbl-128: Rename 'long double' to '_Float128'

[Joseph Myers <joseph at codesourcery dot com>]

On Tue, 30 Aug 2016, Paul E. Murphy wrote:

> I do not plan on working on the matherr/_LIB_VERSION
> deprecation until after we work through _Float128.  I
> don't think there is enough time in this release cycle
> to do it.  There are too many machine specific quirks.

There are four months left in this release cycle before freeze.

matherr/_LIB_VERSION deprecation naturally splits into turning symbols 
into compat symbols / updating documentation, and introducing new wrappers 
that don't use _LIB_VERSION / __kernel_standard*.  It's the latter that 
involves machine-specific issues (where machines have their own versions 
of wrappers), but can also be split to keep the machine-independent parts 
of the change separate from the machine-specific ones.

(Actually, I'd also expect preliminary patches to be needed for cases 
where the __ieee754_* functions being wrapped do not return the correct 
values for special cases but this is currently hidden by the wrappers.)

> Though, as noted in earlier discussions, _Float128
> will not support matherr/_LIB_VERSION.

Well, the patches will need to include appropriate rationale for the 
approach taken being the right design (which I'd have thought meant new 
type-generic wrappers that are also used for other types for static 
linking / absence of symbol version compatibility with old releases, but 
it's open to you to argue the merits of some other design in the patch 
submissions).

-- 
Joseph S. Myers
joseph@codesourcery.com