Control: tags -1 + wontfix
Control: close -1

Hi!  [CC += #953879]

On Wed, Feb 26, 2025 at 02:21:17AM +0100, Cyril Brulebois wrote:
> I don't think that's a practical problem, but that would be a change
> compared to the libnl-genl-3-200-udeb package found in unstable, which
> ships the library under /usr/lib/<triplet>.
> 
> This package builds two udebs, and the way both libraries are managed
> (debs and udebs) is a little bit weird:
> 
>     (unstable-amd64-devel)kibi@tokyo:~/hack/bsp/libnl3 (master =)$ cat 
> debian/libnl-3-200-udeb.install
>     usr/lib/*/libnl-3.so.* usr/lib
> 
>     (unstable-amd64-devel)kibi@tokyo:~/hack/bsp/libnl3 (master =)$ cat 
> debian/libnl-genl-3-200-udeb.install 
>     usr/lib/${DEB_HOST_MULTIARCH}/libnl-genl-3.so.* lib/
> 
>     (unstable-amd64-devel)kibi@tokyo:~/hack/bsp/libnl3 (master =)$ cat 
> debian/libnl-3-200.install 
>     debian/tmp/usr/lib/${DEB_HOST_MULTIARCH}/libnl-3*.so.* 
> usr/lib/${DEB_HOST_MULTIARCH}/
>     debian/tmp/etc/libnl-3/* etc/libnl-3
> 
>     (unstable-amd64-devel)kibi@tokyo:~/hack/bsp/libnl3 (master =)$ cat 
> debian/libnl-genl-3-200.install 
>     debian/tmp/usr/lib/${DEB_HOST_MULTIARCH}/libnl-genl-3*.so.*
> 
> Contrary to the deb world, we don't have a strong urge to move udebs to
> the unaliased location (that was discussed a little while back with
> Helmut), and I've suggested we wait for d-i maintained packages (instead
> of introducing more changes that aren't strictly needed yet).
> 
> For other packages, especially those going through some major rework
> like this one, it's probably best to align debs and udebs. At the very
> least, unless there are some weird hardcoded paths being part of some
> ABI (that's been known to happen…), I'd recommend having both
> libnl-3-200-udeb and libnl-genl-3-200-udeb ship their files under the
> same location, be it /lib, /usr/lib, or /usr/lib/<triplet>.

There's #953879 from 2020 that asks for movement to /lib/
to facilitate mklibs removal (unclear to me what this means)
but I'm taking this to take precedent.

With cdbs, the package was double-built and used the -O2 default for debs
and -Os for udebs, so I'm taking this to also mean that we shouldn't be
even trying to replicate this
(and the .sos end up being the-same-or-smaller than before anyway).
 
> Seeing how both debs are shipping under /usr/lib/<triplet>, that would
> mean doing the same for both udebs, leading up to:
>  - no divergences between debs and udebs;
>  - no divergences between both udebs;
>  - no more lintian error (which wouldn't be critical anyway) for one
>    udeb.

Best,

Attachment: signature.asc
Description: PGP signature

Reply via email to