On Mon, Apr 8, 2024 at 9:15 PM Attila Szalay <s...@ubainba.hu> wrote:
>
> Ok. I re-checked the patch and realized that I checked the only wrong
> module (the only one which is arch all and not any).
>
> So my apologies for the fuzz and will apply the patch with the
> appropriate changes.
>
> But here my original reply too:
>
> I admit that I joined late to this conversation. But my complain is not
> about the time_t change.
>
> My complain is the contradiction between two thing:
> 1. https://wiki.debian.org/binNMU says that I should declar[e]
> dependency between an arch: all to an arch: any package: Depends: foo
> (>= ${source:Version}), foo (<< ${source:Version}.1~)
>
This is for arch: all -> arch: any. However I see most syslog-ng-mod-*
packages are arch: any. So it should just use strict equal on
syslog-ng-core.
What I'm confused about is some syslog-ng-mod-* packages are arch:
all, which don't have .so in it. Then why do they need to depend on
syslog-ng-core?
> 2. The bug report ask to depend on 'syslog-ng-core (=
> $${binary:Version})'
>
> This two is contradict and because syslog-ng complies with the binNMU
> request, I really reluctant to change that. Especially because in that
> case another ticket will be created in no-time to revert it.
>
> This is from the proposed changelog:
> + * Adjust shlibs for syslog-ng-core to use a strict versioned depends;
> + previously, modules used >=, << dependencies which did not account for
> + the possibility of ABI skew in a binNMU, which is exactly what happens
> + with the 64-bit time_t transition.
>
> And my question is again, is the rules really changed or we bend the
> rules just because of one transition?
--
Shengjing Zhu
