[debhelper-devel] Bug#747141: Bug#747141: ping
Niels Thykier
niels at thykier.net
Mon Mar 13 06:54:00 UTC 2017
Dima Kogan:
> Niels Thykier <niels at thykier.net> writes:
>
>> The issue in geda-gaf cannot be solved in debhelper correctly. Please
>> have a look at #766711 to understand the reasoning.
>>
>> Basically:
>> * We can only support two style of binNMUs: arch:all -> arch:all OR
>> arch:any -> arch: any. Anything else will break.
>> * Compat 10 and later will forbid possibly broken setups. We tried to
>> retro fit it into compat 9 and earlier, but it broke in unintended
>> ways (we could not reliably detect the issues).
>
> Hi. Thanks for the reply.
Hi,
You are welcome. :)
> Like you say, "at first glance" (pkg =
> ${source:Version}) would work for this specific case, and that's what
> the second patch here does.
>
FTR, the ${source:Version} is only true for arch:any -> arch:all; for
all "valid" cases, a ${binary:Version} solves it (${source:Version} does
not work for an arch:any -> arch:any). Finally, there is no solution
for arch:all -> arch:any.
> Clearly I haven't thought about this as much as you have, and if it
> really looks unfixable, it'd be great if
>
> - An attempt to make a broken binNMU would hard-fail instead of throwing
> an easily-missed warning
>
This is what we introduced in compat 10 and later.
> - The error message has a link to this and related bugs
>
I suppose this could be done as well.
> Currently dh_installdocs throws an error if compat != 9. Is it
> acceptable to make it fail even for compat == 9?
>
The release team rejected that on the grounds that too many packages
might start to FTBFS. I should mention last time this was considered was
during the jessie freeze, so it would have been rather unhelpful for
them at that time.
I attempted to support detection of the binNMU case and abort on only on
that (so we got a warning on a regular build and a FTBFS on binNMU
rather than a broken package). Unfortunately, it did not work so well.
In the end I settled with fixing it for compat 10.
Thanks,
~Niels
More information about the debhelper-devel
mailing list