Re: binNMU or reproducible builds (choose only one)
On Thu, 24 Sep 2015, Santiago Vila wrote:
> You don't appreciate the beauty of simplicity.
* I do appreciate not triggering a libreoffice rebuild on the slow arches when
a binNMU is required on amd64.
* I do appreciate multiarch not being broken by binNMU'd packages, as well
as getting rid of the nasty effects binNMUs have on package dependencies
(source version x binary version, not to mention the arch x arch).
In fact, the dependency chain of several packages are already more
fragile than they should be (as in: it was relaxed to allow broken/
undesired version skews) to tolerate binNMUs.
And this is not trivial to get right, either. It is a constant source
of packaging bugs.
I believe those two are the important tradeoffs that need to be considered.
Keeping in mind that binNMUs are, in fact, _very_ expensive because they
limit what we can do in undesireable ways, IMHO, the important question is:
are binNMUs worth the *MAJOR* savings in buildd times on the slow arches,
given the *MAJOR* issues and limitations they cause?
Let's answer that one, and if the answer is "lets drop binNMUs", then we can
work towards source-based auto-NMUs being at least as easy to use/trigger as
binNMUs currently are.
"One disk to rule them all, One disk to find them. One disk to bring
them all and in the darkness grind them. In the Land of Redmond
where the shadows lie." -- The Silicon Valley Tarot