Long release cycle (was: RE: uclibc tool chain installation problem)

[Peter S. Mazinger]

On Tue, 21 Mar 2006, Khem Raj wrote:

> Mike Frysinger said the following on 03/21/2006 04:29 PM:
> > On Tuesday 21 March 2006 14:00, Erik Andersen wrote:
> >   
> >> And I would like to see us getting
> >> a release out in the very near future.  Its definately been
> >> overly long since the last one.
> >>     
> >
> > ive kept the TODO list pretty up to date ... the current list in there for the 
> > 0.9.29 release is enough for me ... if there's something people would like to 
> > see, i'd mention it now ...
> >   
> getopt for ARM port was not working after feb 14th snapshot. I did not 
> verify it since then the problem
> was due to opt* symbols being hidden in libc. I discussed it briefly 
> with Peter then. It would be worth
> trying out to boot an arm systerm.

Those failures (and the ones w/ environ) were not reproducable natively on 
x86, but I have removed since then hidden_data_def() around all  
non-constant global data.

Another report said that strong_alias w/ global_data does not work at all 
(error/environ), could you test this w/ current svn in your env?

I am thinking of following additions:
a. full gnu glob (that passes all tests, but is very big) or alternatively 
modify glob-susv3 (already in tree) to support the gnu options as well
b. fnmatch-susv3 (from Rich Felker), passes all fnmatch tests that are not 
related to collation unlike the current ones (old and new)
c. stripped/adapted TRE regex (this will become in the near future LGPL 
licensed and passes more tests of the AT&T testregex then the current 
ones)
d. alternate getopt-susv3 (from Rich Felker) that supports multibyte 
characters

Corrections that should be done:
a. if NPTL is added to this release, then get new linuxthreads working as 
well, else disable new linuxthreads in release
b. get getopt-susv3 + getopt_long-simple work as replacement of gnu getopt 
(so that bb at least is usable w/ it), maybe support for reordering the 
options is needed as well

Some other ideas, maybe not really .29 stuff:
1. Get rid of LFS/non-LFS duplicates, having only one version, without the 
redirect/redefine stuff, this would shrink libc a lot (providing only 
aliases to x64() if needed)
2. wide character enabled per default, providing only some fake 
replacements for the non-wide case
3. UTF-8 default maybe
4. I began "marking" the functions as SVID/GNU, BSD/XOPEN will follow. 
This will allow to limit the functions to a specific subset.

Peter

-- 
Peter S. Mazinger <ps dot m at gmx dot net>           ID: 0xA5F059F2
Key fingerprint = 92A4 31E1 56BC 3D5A 2D08  BB6E C389 975E A5F0 59F2