Re: build depends on kernel-headers

[Sam Hartman <hartmans@mit.edu>]

>>>>> "Manoj" == Manoj Srivastava <srivasta@debian.org> writes:


    Manoj> 	I would lean towards correctness rather than
    Manoj> popularity. If I wanted a popular, shoddy system, I would
    Manoj> stay with windows.  If people are going to be making
    Manoj> decisions on shallow reasons like those you menti0on, I am
    Manoj> not sure I want them using Debian; our support structure is
    Manoj> trained enough as it is.


So, the question is how you get to the correct solution.  Basically,
in this case and in many others, we have discovered a broken
interface--the libc kernel headers.  You have three things you need to
do: adopt a new interface, get people to use it and then drop the old
interface.

My original comments were stated in terms of the user experience,
which I find quite compelling as an argument.  I could have easily
stated it in terms of correctness at a different level.

Debian should be committed to smoothe transitions when interfaces
change.  We have a lot of past history in this area--significant
effort is spent making multiple versions of libraries be able to be on
the system at once; I find numerous other examples where we have
supported an incorrect interface longer so that our users could
upgrade smoothly.

The important lesson to learn from this is that phased upgrades need
to happen in the order I listed.  In the case at hand, I don't think
most maintainers (or even a significant fraction) have really adapted
to how they will handle the kernel interfaces.  Thus, they haven't
adopted the new interface yet, so perhaps it is premature to be
pulling the old interface.