Drop 2.6 and 3.3 support #1245

Closed
mrocklin opened this Issue Jun 7, 2016 · 10 comments

Comments

Projects
None yet
7 participants
@mrocklin
Member

mrocklin commented Jun 7, 2016

Should we drop support for Python 2.6 or 3.3?

Historically we have targetted these versions because we intend dask to be an infrastructural library and because these versions are fairly easy to support in the core of the library.

We might consider dropping support for these versions in some submodules (like dask.dataframe) where users are typically more up to date than in other modules (like dask.array) or the core.

@eriknw

This comment has been minimized.

Show comment
Hide comment
@eriknw

eriknw Jun 7, 2016

Contributor

+1 for at least dropping Python 2.6.

Contributor

eriknw commented Jun 7, 2016

+1 for at least dropping Python 2.6.

@jcrist

This comment has been minimized.

Show comment
Hide comment
@jcrist

jcrist Jun 7, 2016

Member

Scikit-image just dropped 2.6 support (scikit-image/scikit-image#2033), and xarray only supports 2.7 and 3.3+, so I think we're fine dropping 2.6 for at least dask.array. I don't think we can drop 3.3 yet, as both packages still support 3.3. So I'm also +1 on dropping 2.6.

Member

jcrist commented Jun 7, 2016

Scikit-image just dropped 2.6 support (scikit-image/scikit-image#2033), and xarray only supports 2.7 and 3.3+, so I think we're fine dropping 2.6 for at least dask.array. I don't think we can drop 3.3 yet, as both packages still support 3.3. So I'm also +1 on dropping 2.6.

@seibert

This comment has been minimized.

Show comment
Hide comment
@seibert

seibert Jun 7, 2016

We dropped both Python 2.6 and 3.3 from Numba. Our conda package download numbers indicated that 3.3 was the least popular, but Python 2.6 was the most annoying to support, so we deprecated both, and then dropped them the following release.

seibert commented Jun 7, 2016

We dropped both Python 2.6 and 3.3 from Numba. Our conda package download numbers indicated that 3.3 was the least popular, but Python 2.6 was the most annoying to support, so we deprecated both, and then dropped them the following release.

@bryevdv

This comment has been minimized.

Show comment
Hide comment
@bryevdv

bryevdv Jun 7, 2016

FWIW Bokeh is also only tested on py 2.7 and 3.4+

bryevdv commented Jun 7, 2016

FWIW Bokeh is also only tested on py 2.7 and 3.4+

@seibert

This comment has been minimized.

Show comment
Hide comment
@seibert

seibert Jun 7, 2016

I should note that we never found a situation where Python 3.3 and 3.4 behaved differently in a way we had to work around, so support for Python 3.3 likely costs you very little effort. We just wanted to tidy up our 3D configuration matrix.

seibert commented Jun 7, 2016

I should note that we never found a situation where Python 3.3 and 3.4 behaved differently in a way we had to work around, so support for Python 3.3 likely costs you very little effort. We just wanted to tidy up our 3D configuration matrix.

@ianozsvald

This comment has been minimized.

Show comment
Hide comment
@ianozsvald

ianozsvald Jun 8, 2016

I surveyed our meetup audience (200 folk) at PyDataLondon last night (and I do this most months), we're about 50/50 for Py 2.7 vs 3.4+ in the audience with a slight favour still to using Py2.7. Very few folk in the room use Py2.6 and one or two might be stuck on 2.4 (e.g. old banking systems). For Py 3 most people are on 3.4 or 3.5. Dask is cutting edge, you've got a lot of latitude I'd argue to drop older versions that cause you a drag. Dask was our "package of the month", most of our audience hadn't heard of it (our PoTM highlights stuff we reckon our audience ought to have a clue about).

It might also be worth raising the question - is Py2.7 support, given the Jan 1st 2020 sunset date, a worthwhile support target given the size of the existing userbase? Probably the answer is yes (do you have figures?) but just maybe you've got a majority of Python 3+ users and supporting 2.7 is a drag. I don't expect this to be true in 2016, but I figure you're a young enough project that the question about added inertia for Py2+3 support is worth raising.

ianozsvald commented Jun 8, 2016

I surveyed our meetup audience (200 folk) at PyDataLondon last night (and I do this most months), we're about 50/50 for Py 2.7 vs 3.4+ in the audience with a slight favour still to using Py2.7. Very few folk in the room use Py2.6 and one or two might be stuck on 2.4 (e.g. old banking systems). For Py 3 most people are on 3.4 or 3.5. Dask is cutting edge, you've got a lot of latitude I'd argue to drop older versions that cause you a drag. Dask was our "package of the month", most of our audience hadn't heard of it (our PoTM highlights stuff we reckon our audience ought to have a clue about).

It might also be worth raising the question - is Py2.7 support, given the Jan 1st 2020 sunset date, a worthwhile support target given the size of the existing userbase? Probably the answer is yes (do you have figures?) but just maybe you've got a majority of Python 3+ users and supporting 2.7 is a drag. I don't expect this to be true in 2016, but I figure you're a young enough project that the question about added inertia for Py2+3 support is worth raising.

@mrocklin

This comment has been minimized.

Show comment
Hide comment
@mrocklin

mrocklin Jun 8, 2016

Member

Thanks for the poll. That's quite helpful.

On Wed, Jun 8, 2016 at 1:13 AM, Ian Ozsvald notifications@github.com
wrote:

I surveyed our meetup audience (200 folk) at PyDataLondon last night (and
I do this most months), we're about 50/50 for Py 2.7 vs 3.4+ in the
audience with a slight favour still to using Py2.7. Very few folk in the
room use Py2.6 and one or two might be stuck on 2.4 (e.g. old banking
systems). For Py 3 most people are on 3.4 or 3.5. Dask is cutting edge,
you've got a lot of latitude I'd argue to drop older versions that cause
you a drag.


You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub
#1245 (comment), or mute
the thread
https://github.com/notifications/unsubscribe/AASszElqzCW2KGFN846txf5as6RO25Tnks5qJnmagaJpZM4IwY8d
.

Member

mrocklin commented Jun 8, 2016

Thanks for the poll. That's quite helpful.

On Wed, Jun 8, 2016 at 1:13 AM, Ian Ozsvald notifications@github.com
wrote:

I surveyed our meetup audience (200 folk) at PyDataLondon last night (and
I do this most months), we're about 50/50 for Py 2.7 vs 3.4+ in the
audience with a slight favour still to using Py2.7. Very few folk in the
room use Py2.6 and one or two might be stuck on 2.4 (e.g. old banking
systems). For Py 3 most people are on 3.4 or 3.5. Dask is cutting edge,
you've got a lot of latitude I'd argue to drop older versions that cause
you a drag.


You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub
#1245 (comment), or mute
the thread
https://github.com/notifications/unsubscribe/AASszElqzCW2KGFN846txf5as6RO25Tnks5qJnmagaJpZM4IwY8d
.

@ianozsvald

This comment has been minimized.

Show comment
Hide comment
@ianozsvald

ianozsvald Jun 9, 2016

I'll also note that I ran a similar poll at PyDataBerlin conf 2 weeks back, it was more like a 40/60 split for Python 2/3 (and the majority of each were on 2.7 or 3.4+). I've also been gathering evidence for the dropping of 2.6 support and the switch to 3.4+ only:
http://ianozsvald.com/2016/02/29/will-we-see-module-on-python-3-4-is-free-but-only-paid-support-for-python-2-7/

I'll also note that I ran a similar poll at PyDataBerlin conf 2 weeks back, it was more like a 40/60 split for Python 2/3 (and the majority of each were on 2.7 or 3.4+). I've also been gathering evidence for the dropping of 2.6 support and the switch to 3.4+ only:
http://ianozsvald.com/2016/02/29/will-we-see-module-on-python-3-4-is-free-but-only-paid-support-for-python-2-7/

@seibert

This comment has been minimized.

Show comment
Hide comment
@seibert

seibert Jun 9, 2016

Just to be clear, dropping Python 2.6 is a very different question than dropping Python 2.7. Unless a project absolutely depends on a Python 3 feature (or is purely a hobby, so do what makes you happy), I think Python 2.7 support is still a really good idea, and not onerous with the help of CI.

seibert commented Jun 9, 2016

Just to be clear, dropping Python 2.6 is a very different question than dropping Python 2.7. Unless a project absolutely depends on a Python 3 feature (or is purely a hobby, so do what makes you happy), I think Python 2.7 support is still a really good idea, and not onerous with the help of CI.

@mrocklin mrocklin referenced this issue Jun 9, 2016

Merged

Drop 2.6 #1264

@mrocklin mrocklin closed this in #1264 Jun 9, 2016

@sinhrks sinhrks added this to the 0.10.0 milestone Jun 14, 2016

@ianozsvald

This comment has been minimized.

Show comment
Hide comment
@ianozsvald

ianozsvald Jun 20, 2016

Hello again. I was curious about the question for Python 2vs3 popularity for data scientists. I queried my PyDataLondon audience (3,400+ members) last week, by sending 4 emails I got 13% of them to respond to my survey. Here are the results:
http://ianozsvald.com/2016/06/20/results-for-which-version-of-python-2vs3-do-london-data-scientists-use/

In short - Python 2.7 dominates in business, Python 3.4 dominates outside of work. I hypothesise that we'll hit 50% Python 3.x usage by this time next year for PyDataLondon users. Almost nobody uses Python <=2.6 or Python <=3.3.

Hello again. I was curious about the question for Python 2vs3 popularity for data scientists. I queried my PyDataLondon audience (3,400+ members) last week, by sending 4 emails I got 13% of them to respond to my survey. Here are the results:
http://ianozsvald.com/2016/06/20/results-for-which-version-of-python-2vs3-do-london-data-scientists-use/

In short - Python 2.7 dominates in business, Python 3.4 dominates outside of work. I hypothesise that we'll hit 50% Python 3.x usage by this time next year for PyDataLondon users. Almost nobody uses Python <=2.6 or Python <=3.3.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment