ng0 <n...@we.make.ritual.n0.is> writes:
>> * The Debian package is built with HAVE_ICONV=1, should we set that too?
> I can send an updated patch with libiconv in the inputs. This is when
> you use a libc which does not provide a (usable) iconv, which is why
> Gentoo provides the option to build it with this too. As we are
> _currently_ only providing options to do one libc globally this is not
> so inmportant. I will even build uclibc-ng (I am working on that) with
> an outside iconv because reportedly their iconv is in bad shape.
> So with this information, should I send a new patch which adds libiconv
> and the build option?
Since we only support glibc at the moment, I don't think adding a
package for libiconv first is necessary. On the other hand, if it can
cause problems on other libc's, it's nice to use an external one.
IMO adding libiconv as input is something that can be done later, when
there is an actual need for it. I guess there are more packages that
will require it. But I don't really mind either way :)
>> Additionally it installs locales, but looks for /usr/share/locale at
>> runtime. Fixing this would probably require upstream help, I don't want
>> to hardcode either "/run/current-system/locale" or ~/.guix-profile. The
>> current version probably works on foreign distros, if nothing else.
> If you know how to fix it, try to bring it to upstream or include a
> patch / substitute phase at our end?
I tried patching whois.c to use the glibc default in the call to
bindtextdomain(3), instead of the LOCALEDIR build option. That almost
worked: it searches the system locales, but also expect to find the
translation files there. I'll see if I can cook up a fix for upstream,
but don't think this should hold back the package.