Re: Planning for 2.18?

[Carlos O'Donell <carlos at systemhalted dot org>]

On 01/04/2013 07:40 PM, Roland McGrath wrote:
> It would be nice if we could finally do the revamp of sysdeps selection.
> I still have basically no plan whatsoever about the details, but the
> status quo (especially after Implies-before/Implies-after) is a hazard
> to maintenance and porting.
> 
> It seems a good time to start whittling away at the ports distinction.
> As has been said before, it should be done only via whatever cleanups
> are necessary to ensure that no generic files get machine-specific bits.
> I suggest that arm migrate to the main tree first.  But I don't
> particularly plan to work on this myself.
> 
> Also, not necessarily related to broader planning per se, my day job
> will include in the coming months working on a new port to Native Client
> (first arm and i686, later x86_64) done from scratch.  (There is an
> existing x86-only port now in production use, which was done as a fork a
> few years ago without my involvement--though I now have a hand in
> maintaining it--and is unspeakable in a variety of ways.)  I'll do this
> initially as a separate add-on port not in the repository, which is the
> only appropriate thing until I get the GCC target support merged
> upstream, and decide later whether it will stay that way or go into the
> repository.  (Of course all my work will be available on public git
> servers along the way, and will all be FSF-copyright.)  This was the
> main motivation for my intermittent work over the last several months in
> clearing the bit-rot for "stubby" configurations.  Since I intend to do
> it only in "right" ways, it will motivate further generic cleanups: to
> continue to excise creeping Linuxism; to clean up NPTL to support
> non-Linux configurations with a futex mechanism; some of the sysdeps
> selection stuff, since e.g. it's an x86 configuration that won't use
> ldbl-96 (the target has long double==double); pure cross-testing; surely
> others.

Roland,

I think the kind of information you've just revealed about a new
port to Native Client is an excellent example of how planning early
can help everyone. Just mentioning this will mean I'll be more likely
to keep an eye out for relevant issues and cleanup where appropriate.

Regarding the revamping of the sysdep selection, that's certainly
something that we'd all like to see fixed, but it's honestly such
drudgery that I dread even looking at it. I guess we have to tackle
it at some point. Have you written any hand-waving ideas about the
sysdep selection rewrite into the wiki?

Cheers,
Carlos.