Control: retitle -1 ncurses: regressions from the -dev packages merge
Control: severity -1 normal

On 2018-05-02 19:13 +0200, Sven Joachim wrote:

> On 2018-05-02 13:54 +0200, Matthias Klose wrote:
>
>> Package: src:ncurses
>> Version: 6.1+20180210-2
>> Severity: important
>> Tags: sid buster
>>
>> ncurses normal and wide variants should not share the same -dev package. Up 
>> to
>> now, packages had to be explicit about the ncurses variant.
>
> There are two reasons why I unified the -dev packages: because there is
> now only one set of header files (eliminating possible bugs where
> packages would use the wrong ones), and to get rid of libtinfo-dev which
> had confused both users and package maintainers.
>
>> I fear that packages using the wide variant just might be rebuilt
>> using the normal variant, regressing due to the -dev package merge.
>
> That seems rather unlikely to me, and I have not encountered any example
> in my test rebuilds.  Surely such a package would be have been broken
> already, since libncurses5-dev could be pulled in by another dependency
> or already been present on the build system.

All of the packages which used to depend on libncursesw5 and have been
rebuilt for the transition now depend on libncursesw6, at least on
amd64/main.  A few of them depend also on libncurses6, but there was no
case of a regression.

> The other case where packages used to link against the non-wide library
> but now depend on the wide one is much more likely, but I consider this
> a feature.

What I had not anticipated is that a buildsystem could be so buggy as to
completely disable ncurses support if it finds both ncursesw.so and
ncurses.so.  This happened in #898131, but apparently it is the only
case so far.

> Will keep an eye on the packages which now depend on libncursesw5 to see
> if they regress during the transition, but I'd much rather not undo the
> -dev package merge.

Approximately ⅔ of all packages involved in the ncurses transition have
been rebuilt, and the results did not show a compelling reason to undo
the merge.  But let's keep this bug open as a regression tracker, I had
already added #898131 as a blocker.

Cheers,
       Sven

Reply via email to