Re: on the 30 FTBFS bugs - let's slow down a bit

[Sylvain Le Gall <gildor@debian.org>]

On 06-10-2009, Stefano Zacchiroli <zack@debian.org> wrote:
>
> Also, I was wondering whether the FTBFS where dh_ocaml exits with an
> error ("interface name blabla re-exported" or something such) should
> really be errors. I agree that at this stage they are very helpful, but
> I was wondering what should be the appropriate behavior in the future.
>

This error is a really important error. Taking the example of
ocaml-fileutils 0.3.0 (I am upstream and DD on this). This package
re-export the module Unix. This means that the library contains the
module Unix. To my mind, this is a real error. Just like linking libz.a
in libssl.a (just example). 

I am very happy that dh_ocaml catch this one. I have removed the
"-linkall" that cause this error. 

For package cothreads, this is the intended behavior. Cothreads mimic
certain module to replace them. In this case there is a flag that
prevent this export. (--nodefined-map)

I think only 30 packages that FTFBS after this kind of change is not a
big deal. I would like to keep calling dh_ocaml by default. This is the
better way to check everything is working (we already kept it too long
inside our repository without using it). 

I will start fixing packages tonight. I will probably keep old
dependencies along computed one until we are 100% sure dh_ocaml is
missing nothing...

Regards,
Sylvain Le Gall