Comments regarding tensorflow_2.0.0-1_amd64.changes

M. Zhou cdluminate at gmail.com
Mon Mar 2 01:27:08 GMT 2020


Hi Sean,

There are two embedded source tarballs:
1. abseil-cpp: the upstream (google) development model makes
   it unsuitable for debian packaging. This is a mandatory
   dependency of tensorflow. hence the embedded tarball.
2. eigen3: debian ships another version of this package. However,
   tensorflow is tightly coupled with a specific snapshot version
   by upstream. building against debian's eigen3 will result in
   nothing but FTBFS. Patching tensorflow to fix the insane amount
   of API breaks could be impractical as currently there is only
   one person working on this stuff.



On Sun, 1 Mar 2020 at 15:59, Sean Whitton <ftpmaster at ftp-master.debian.org>
wrote:

> Hello Mo,
>
> I'm not comfortable with the debian/embedded/ tarballs especially
> since one of them is already in the archive as a shlib.  I don't see
> why tensorflow should get a special exception to the idea that we have
> only one copy of each shlib in the archive except where needed to
> handle transitions.
>
> I'm not REJECTing this right now because I'd like to get input from
> other ftpteam members, but I wanted to write to you to give you a
> chance to explain why tensorflow might be a special case.
>
> --
> Sean Whitton
>
>
>

-- 
Best,
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://alioth-lists.debian.net/pipermail/debian-science-maintainers/attachments/20200302/2c82271c/attachment.html>


More information about the debian-science-maintainers mailing list