Hi, Il 28/11/18 00:10, Dimitri John Ledkov ha scritto: > Mostly, because In terms of what boost builds here, it is irrelevant > what the default version is. And people need to know which versions > they are building for and use the matching boost bits.
I agree on this point: it does not make any sense to ship "default"
links, since those who link with Boost.Python must link with the correct
version anyway. It is not up to the boost package to tell them what is
the correct version.
So let's sink the default links.
> And hopefully, the sonames will converge to the newest upstream abi
> names across all distros such that this is no longer an issue.
>
> Thus I'd want to simply compile and install just the:
>
> libboost_${PACKAGE}${ALL_SUPPORTED_XY_PYVERSIONS}.so.${BOOST_VERSION}
>
> e.g. currently libboost_python{27,36,37}.so.1.67.0
Ok, we agree on this point. And fortunately this is what the current
package does. Good!
> and nothing else, no symlinks to -py things, nor to the "default"
> versions. As that ultimately is the current vision upstream and makes
> a lot of sense downstream too.
>
> similarly, there is upstream support for tags to support dbg builds to
> e.g. finally ship libboost_python27d.so.1.67.0 for the python-dbg
> build.
Ah, good, I didn't know about this. As a matter of fact, I have never
used python-dbg, but I guess it would be nice to support it.
> Dropping those old symlinks will need full-archive rebuild, and I
> thought it would be too disruptive to include that with the 1.67 by
> default transition. Maybe once the dust settles more, we can make a
> move to drop all the additional symlinks.
Ok, I think I am pretty convinced here too.
So let's avoid doing anything for the moment; reverse dependencies that
already fail to compile are considered to be buggy and suggested to link
against shared objects in the format you wrote above (some have actually
asked for help; I can help them directly into getting stuff right).
After buster is released, when a new upstream version of Boost is
uploaded in Debian, we will remove the last symlinks and consider buggy
all reverse dependencies that rely on them. I am leaving this bug open
as a memo for that.
Do you agree with this? (I hope so, because it is basically your
proposal...)
Thanks, Giovanni.
--
Giovanni Mascellani <g.mascell...@gmail.com>
Postdoc researcher - Université Libre de Bruxelles
signature.asc
Description: OpenPGP digital signature
