[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: flex is no longer M-A:foreign



Hi Manoj,

On Tue, Feb 09, 2016 at 08:02:16PM -0800, Manoj Srivastava wrote:
> On Tue, Feb 09 2016, Helmut Grohne wrote:
> > Manoj, could you maybe turn libfl-dev back M-A:same in unstable? That
> > marking still is correct. The current markings make it impossible to
> > cross build pam even after changing B-D to "flex:native, libfl-dev",
> > because flex:arch1 and libfl-dev:arch2 are never coinstallable without
> > M-A:same there.
> 
>         Done.

Thank you.

> 	So, the options are:
>  1. Wait for the archive-rebuild transition on AWS to complete
>  2. Go with Helmut???s patch
>  3. push the experimental flex to unstable, potentially breaking {0,
>     500} packages. Hopefully skewing towards the lower end.
> 
>         I favour 1 or 2; and which one to choose depends on the urgency
>  of the cross-build goal.

My urgency here mostly came from the circumstance that as long as the
continuous integration setup fails quickly, I cannot see issues in
packages that are built at a later point, which in turn makes them
harder to track down. As long as things work, I can often just look at
recent uploads and find the culprit.

To that end, I came up with a workaround[1] wherein I am repacking the
native flex binary package to remove the libfl-dev dependency and to add
M-A:foreign. After the repacking (and your -6 upload), we are exactly in
situation 1. So I hope that this solves my blinding issue and thus
removes my part of the urgency.

So if you think that 1 is more correct than 2, then by all means, we
should go for it.  The problem I see here is that we cannot start fixing
reverse dependencies before we know which route to take.

I have one additional data point already. pam is not only a package that
needs the host arch libfl-dev, but also needs the build arch libfl-dev.
So my earlier statement that we don't have to fix packages when using my
patch simply is wrong: We don't have to fix them for native builds, but
for cross builds they still can be broken. Even under my proposal, pam
would need its Build-Depends changed to "flex, flex:native".

Furthermore, dpkg, guile-2.0 and libsepol don't need to change their B-D
under either proposal. Some reverse dependencies of flex, like hurd,
might be hard to test on AWS though. So almost 1% checked. ;)

Helmut

[1] http://anonscm.debian.org/cgit/users/helmutg/rebootstrap.git/commit/?id=17610cb0e4f5a0f3bd446246f1b2e8d3ce75244d
    I cannot tell yet whether this works.


Reply to: