Re: Tile changes

[Chris Metcalf <cmetcalf at tilera dot com>]

On 1/29/2012 10:32 AM, Carlos O'Donell wrote:
> On Sun, Jan 29, 2012 at 8:48 AM, Joseph S. Myers
> <joseph@codesourcery.com> wrote:
>> The port is there (in ports) - it was checked into ports well before 2.15
>> was tagged and branched.  But some of the libc changes it depends on
>> weren't fully reviewed and revised as needed before the release and branch
>> - so actually using the port in 2.15 needs backports of those libc
>> changes.  The question is what risk they pose for other architectures -
>> whether it's safe to backport them so the Tile architectures work fully in
>> 2.15 or whether it would be better for Chris to create his own branch,
>> based on 2.15, for Tile users of 2.15, that has the few extra changes
>> needed.
> Agreed.
>
> I haven't reviewed the patches yet, but having a 2.15 TILE branch would
> make it easier to review what needs merging and provide TILE with a
> temporary upstream branch to use for building.

The five proposed changes for TILE are all in private branches to review
and merge.  I believe they're all trivial in general; the only exception is
the firstversions.awk change, which I checked for other architectures and
confirmed was a no-op on them.  The branches are:

cmetcalf/tile_firstversions - fix version range bug
cmetcalf/tile_elf_h - add Tile elf.h constants
cmetcalf/tile_not_cancel_h - change include brackets from "" to <> in grantpt.c
cmetcalf/tile_epoll_pwait - add internal linkage name for epoll_pwait
cmetcalf/tile_faccessat - use fxstatat, not fstatat

There's the update to config.sub and config.guess, which should be fairly safe;
I went ahead and grabbed the latest from the "config" git repository and
put them in a private glibc branch for merging:

cmetcalf/tile_config - config.guess, config.sub updates

I'd be happy to make a single tile branch holding all the changes if that
would be helpful.

I also have two changes for ports that would need to be backported
to bring everything together, based on the changes above:

- Add the openat() version of linux/generic/not-cancel.h
- Remove now-unnecessary epoll_pwait() internal def in linux/generic/epoll_wait.c

These are also both pretty trivial, and in any case only affect the tile
architecture at this point.

-- 
Chris Metcalf, Tilera Corp.
http://www.tilera.com