I'm cancelling this vote -- it doesn't feel right to change what we are
voting on mid-vote.
And doing it now we can do it as a three day vote, so we only loose one
day.
New vote for 2.0.0rc2 to follow shortly
On Sat, 12 Dec, 2020 at 12:25, Jarek Potiuk <jarek.pot...@polidea.com>
wrote:
Ah cool. So I think we can drop sdist indeed :).
I think if it is the only problem found till Monday, I would be OK
in dropping sdist from 2.0.0 release then - if this is going to slow
down the process by a week. I am happy to change my vote if others
are happy with it.
Regardless, I updated our tests:
<https://github.com/apache/airflow/pull/13033>. During the tests we
installed airflow always from .whl only providers were installed from
either wheel or sdist. Now once it works, both airflow and providers
should be installed using the same package format.
J.
On Sat, Dec 12, 2020 at 11:15 AM Ash Berlin-Taylor <a...@apache.org
<mailto:a...@apache.org>> wrote:
The BigQuery dataset PyPi publishes has a file.type column:
https://packagingpython.org/guides/analyzing-pypi-package-downloads/
<https://packaging.python.org/guides/analyzing-pypi-package-downloads/>
Since we started publishing wheels (1.10.3) the wheel almost all of
the downloads. This 2020-12-01 (picked randomly):
1.10.12bdist_wheel108311.10.12sdist121.10.13bdist_wheel31451.10.13sdist32
1.10.4sdist2
1.10.4bdist_wheel2399
1.10.3bdist_wheel1079
1.10.3sdist3
On Sat, 12 Dec, 2020 at 10:50, Jarek Potiuk
<jarek.pot...@polidea.com <mailto:jarek.pot...@polidea.com>> wrote:
Update: not provider packages, but the packages that airflow
depends on. indeed all our provider/airflow package are pure python
and as long as there are no "special cases" like anyone requiring
source packages (for whatever reason) they *should* work.
On Sat, Dec 12, 2020 at 10:48 AM Jarek Potiuk
<jarek.pot...@polidea.com <mailto:jarek.pot...@polidea.com>> wrote:
It might well be, yes, that we have no such problem in the airflow
package. I referred more to the provider packages. I think it
would like to hear from the user why sdist was used in this case.
Also, I believe there are some distros Gentoo, that always use
sources for everything it can (and I am not sure if they would
work with .whl only).
I am not against removing it, just want to be sure there are no
unforeseen consequences. Unfortunately, there are no stats to show
which type of packages are downloaded :(
<https://pypistats.org/packages/apache-airflow>
J,
On Sat, Dec 12, 2020 at 10:38 AM Ash Berlin-Taylor <a...@apache.org
<mailto:a...@apache.org>> wrote:
On Sat, 12 Dec, 2020 at 10:34, Jarek Potiuk
<jarek.pot...@polidea.com <mailto:jarek.pot...@polidea.com>>
wrote:
So I think there will be systems where sdist is used
automatically because the host system is not as close to what we
use when build wheels.
I think that is only a problem for wheels with binary
components,but our wheels are not -- they are universal (i.e.
just contain pure python).
--
Jarek Potiuk
Polidea <https://www.polidea.com/> | Principal Software Engineer
M: +48 660 796 129 <tel:+48660796129>
<https://www.polidea.com/>
--
Jarek Potiuk
Polidea <https://www.polidea.com/> | Principal Software Engineer
M: +48 660 796 129 <tel:+48660796129>
<https://www.polidea.com/>
--
Jarek Potiuk
Polidea <https://www.polidea.com/> | Principal Software Engineer
M: +48 660 796 129 <tel:+48660796129>
<https://www.polidea.com/>
