1.0b1 on PyPI?

368 views
Skip to first unread message

Jim Fulton

unread,
Mar 19, 2012, 5:52:53 PM3/19/12
to gev...@googlegroups.com
Is there any good reason not to release 1.0b1 to PyPI?

I've started releasing a package to PyPI that requires
1.0b1. It might be confusing to people who try to get it
that they can't get one of its dependencies on PyPI.

Jim

--
Jim Fulton
http://www.linkedin.com/in/jimfulton

Denis Bilenko

unread,
Mar 22, 2012, 7:00:06 AM3/22/12
to gev...@googlegroups.com
On Tue, Mar 20, 2012 at 1:52 AM, Jim Fulton <j...@zope.com> wrote:
Is there any good reason not to release 1.0b1 to PyPI?

I've started releasing a package to PyPI that requires
1.0b1.  It might be confusing to people who try to get it
that they can't get one of its dependencies on PyPI.

I've tried uploading 1.0b1 to PyPI but hiding it, however "pip install gevent" still gets 1.0b1 and not 0.13.6 as it supposed to.

Cheers,
Denis.

Matthias Urlichs

unread,
Mar 22, 2012, 7:19:35 AM3/22/12
to gev...@googlegroups.com
Hi,

Denis Bilenko:


> On Tue, Mar 20, 2012 at 1:52 AM, Jim Fulton <j...@zope.com> wrote:
>
> > Is there any good reason not to release 1.0b1 to PyPI?
> >
>
> Yeah, http://ziade.org/2011/02/15/do-not-upload-dev-releases-at-pypi/
>

>> the best practice for mature projects is to avoid publishing anything
>> that is not a final release at PyPI

gevent is NOT a mature project; we don't even have a 1.0 release yet.

Since it's a beta and people really should use that instead of relying on
0.13, I'm +1 for uploading.

--
-- Matthias Urlichs

Jim Fulton

unread,
Mar 22, 2012, 12:00:04 PM3/22/12
to gev...@googlegroups.com
On Thu, Mar 22, 2012 at 7:00 AM, Denis Bilenko <denis....@gmail.com> wrote:
> On Tue, Mar 20, 2012 at 1:52 AM, Jim Fulton <j...@zope.com> wrote:
>>
>> Is there any good reason not to release 1.0b1 to PyPI?
>
>
> Yeah, http://ziade.org/2011/02/15/do-not-upload-dev-releases-at-pypi/

That statement is unfortunate and a little misleading:

"Until our packaging ecosystem knows how to handle properly
development releases,"

Buildout has properly handled development releases since 2007.

Tarek acknowledged that buildout has a prefer-final option later
in his message.

>> I've started releasing a package to PyPI that requires
>> 1.0b1.  It might be confusing to people who try to get it
>> that they can't get one of its dependencies on PyPI.
>
>
> I've tried uploading 1.0b1 to PyPI but hiding it, however "pip install
> gevent" still gets 1.0b1 and not 0.13.6 as it supposed to.

It's sad that more than a year since Tarek's post, pip still
doesn't address this issue. Maybe you should use a better
tool. :)

But then the state of packaging in Python is pretty sad overall.

I still think you should release 1.0b1 to PyPI. You want people
to use it. It's mostly backward compatible and I'm pretty sure pip
lets you set versions if you need to.

Jim

P.S. Thanks again for gevent. It's an impressive package.

Nikolay Kim

unread,
Mar 22, 2012, 12:45:37 PM3/22/12
to gev...@googlegroups.com
Denis, is there reason to hold on 1.0 release?

Florian von Bock

unread,
Mar 22, 2012, 1:43:58 PM3/22/12
to gev...@googlegroups.com
hi everyone,

i'd like to cite jims comment before any of my own sermon first:

> P.S. Thanks again for gevent. It's an impressive package.

(!)

also i am fairly new to the gevent community and i might miss a point here...

(always wait until somebody says...)

but

;)

i fail to see an issue in not releasing 1.0b(whatever) to wherever.

we can all write requirements.txt files in our own packages pointing to
hg+https://something@someversion#egg=gevent if we have to.

gevent is great software that we like to use and _we_want_it_to_evolve_. and
preferrably at a high pace.

when i went to pycon 2 weeks ago there was some talk about having gevent
included into the std lib and how it might effect its development in a most
likely not so disireable manner (once you're in you have to support every
mistke you made until that point ... and so forth ...)

i am quite confident that gevent 1.0 will be released in an all official
manner some time soon and until then we will have to make some small amends to
freeze a version in our own packages if we really need to.

i think that for a package with as much interest, and attention as gevent it
is best to live with the fact that the maintainers do not yet feel comfortable
enough with it (tip) to fully put it "out there" until they think the time is
right. we can still use every version we want to as it is tagged in the repos
and otherwise sticking to the "debian way" of "it's done when it is done" way
does not seem like a bad idea to me.

the fact that the ability to deal with beta, final, nightly releases of
software should be handled by package-management and not package developers
remains untouched of course! i totally agree here!

but until that gets solved i think that it is fine that "we" - the people
using a package X - can live with the nuissance of doing a little bit of
version management ourselves - and have the people working on those packages
work on those packages... - rather than having them deal with version
management.

just some devs 2 cents... :)

best
_florian

signature.asc

WouterVH

unread,
Mar 27, 2012, 7:44:51 AM3/27/12
to gevent: coroutine-based Python network library
> i fail to see an issue in not releasing 1.0b(whatever) to wherever.

Go the www.gevent.org and click on the huge link in the top-navigation
"Download"
It takes you to http://pypi.python.org/pypi/gevent#downloads point to
0.13.6 as most recent release.
That is an issue.

The other issue is that people using Pypi directly via pip/
easy_install are receiving on older release then what is actually
available,
they get 0.13.6 instead of 1.0b1.

I fail to see why this would benefit the developers. Does anyone like
to receive feedback on already fixed bugs?

I fail to see what the problem is in releasing 1.0b1 on Pypi,
Enough tools exist to make it a one-minute job. (just search
"releaser" on Pypi)



> we can all write requirements.txt files in our own packages pointing to
> hg+https://something@someversion#egg=geventif we have to.

You are assuming people know about the existence of gevent 1.0b1.




> gevent is great software that we like to use and _we_want_it_to_evolve_. and
> preferrably at a high pace.

Yes, that is exactly the reason why everyone would benefit if having
1.0b1 on Pypi.
If 0.13.6 is on Pypi, then certainly 1.0b1

Why make a beta-release and then hide it from real-world usage?
The only reason I can think is 1.0b1 might be a brown-bag release that
nobody should use.


> i am quite confident that gevent 1.0 will be released in an all official
> manner some time soon and until then we will have to make some small amends to
> freeze a version in our own packages if we really need to.

Actually I have to modfiy my buildout and add an extra part to it,
just to remove it again when gevent 1.0 is released.
I certainly hope this will not become a de facto practice considering
the multitude of python-eggs some projects depend on.


> i think that for a package with as much interest, and attention as gevent it
> is best to live with the fact that the maintainers do not yet feel comfortable
> enough with it (tip) to fully put it "out there" until they think the time is
> right.

Begs the questions: is 1.0b1 "out there"? Or is it not "out there"?
To most it's not "out there". Why is it called a "beta" and then not
publish it?

+100 for 1.0b1 on Pypi.

Randall Leeds

unread,
Mar 27, 2012, 11:43:47 AM3/27/12
to gev...@googlegroups.com

On Mar 27, 2012 4:47 AM, "WouterVH" <wouter.va...@gmail.com> wrote:
>
> > i fail to see an issue in not releasing 1.0b(whatever) to wherever.
>
> Go the www.gevent.org and click on the huge link in the top-navigation
> "Download"
> It takes you to http://pypi.python.org/pypi/gevent#downloads point to
> 0.13.6 as most recent release.
> That is an issue.

It's not clear to me. On one hand, I don't like the idea of publishing beta software in PyPI or on a downloads page.  On the other hand, what is a version like 0.13 if not some sort of beta. There's a reason 1.0 bumps the major revision number.

I would back 1.0b in PyPI, but not 1.1b. Ultimately, I don't think it's clear cut either way and I don't think waiting is doing significant harm and may reduce the possibility for frustrating new users if there are issues. More beta testers is great, but only if they know that this is what they are.

-Randall

Reply all
Reply to author
Forward
0 new messages