On 2016-12-03 17:31 -0500, Thomas Dickey wrote:

> On Wed, Nov 30, 2016 at 08:54:02PM +0100, Sven Joachim wrote:
>> On 2012-01-22 10:58 +0100, Sven Joachim wrote:
>> 
>> Fortunately this has all been done by upstream in the meantime, the
>> ncurses package in Debian uses versioned symbols[1], and all reverse
>> dependencies in testing have been rebuilt[2].  So the long requested
>> soname bump can finally happen after the Stretch release. :-)
>> 
>> Before I start working, there are some things which should be decided.
>> 
>> - Right now we ship both wide and non-wide versions of the libraries and
>>   the -dev packages, with the ncursesw headers shipped in a separate
>>   directory.  This leads to problems where reverse dependencies use the
>>   wrong headers, see [3] for an example, and I want to change it.
>> 
>>   There are two possibilities here:
>> 
>>    1. Install the headers for the wide libraries only, and use them for
>>       the non-wide libraries.  This is expected to work and is what
>>       Fedora has been doing for a long time, but there seem to be a few
>>       compatibility problems in the ncurses 6 API[4].
>
> The problem is that ncurses6 uses extended color information which can't
> fit into chtype's (the basis of non-wide ncurses5).  That makes it use
> cchar_t's (which are the basis of wide ncurses5/ncurses6).
>
> The Red Hat bug report points out that the macros which use details of
> the WINDOW structure are a problem which prevents using the headers as-is
> between the three flavors (ncurses5 normal/wide and ncurses6).  In a quick
> review of curses.h, those are
>
>       wattrset
>       wattr_set
>       wattr_get
>
> If the macros are not available, the library provides functions for
> each of those macros.  The performance impact for using the functions
> rather than the macros may not be that high.

One week later you added a --disable-wattr-macros configure option which
Fedora uses now, I guess we should do the same.

>> - Should we continue to distribute separate debugging libraries? Nobody
>>   else does that, apparently.

Tracked in #849003, I guess we'll have to see if someone complains after
the packages are gone.

>> - The multilib packages have been a constant source of bugs and are
>>   considered unsupportable by the dpkg maintainer[6], I want to get rid
>>   of them.  Reverse dependencies are grub-legacy and readline whose
>>   multilib packages don't have any reverse dependencies themselves.

Filed #848166 and #848168 for those.

>> - Should we continue to ship libncurses5 and libncursesw5 packages?  As
>>   long as upstream supports building them, this might make sense.
>
> I've no intention of breaking the ncurses5 stuff (keeping in mind that
> LSB specifies ABI 5...).

While Debian no longer supports the LSB, I suspect there are enough
popular proprietary applications out there that use ABI 5 ncurses
libraries, so I guess they should be kept for now.

Cheers,
       Sven

Reply via email to