Python version support

[AA-Turner]

We're probably coming to the point where even for flit_core I'll drop support for some older Python versions, but let's work that out separately. I generally find it easier to support a pretty broad range of Python versions than to deal with people getting stuck on an old version of a package I maintain because I dropped support for older Python.

Originally posted by @takluyver in #728 (comment)

Flit 3.11 supports Python 3.8+, and flit-core supports 3.6+.

Latest statistics 1 for flit-core downloads for show that 3.6 represents <0.05%, 3.7 represents <0.1%, and 3.8 <2%. 3.9 is the overwhelming majority at >50%.

For flit 2, Python 3.9 is again the overwhelming majority with >80%. The numbers are a little suspicious as 3.9 specifically skyrockets from late October/November onwards. I will try to look at a more granular level.

This probably suggests that dropping 3.6 and 3.7 for flit-core is a reasonable move, but other than that Flit should maintain support for non end-of-life Python releases.

A

[AA-Turner]

Running uvx pypinfo --all --limit 1000 --markdown flit pyversion version, and filtering to only Python 3.6+ and Flit 3.7+, I get the following results. There's a clear outlier with Python 3.9 and Flit-core 3.10.1, which the second sets of table clips to 50,000.

Note that this is for the last 30 days, and Flit 3.11 was only released on the 19th.

Flit-core

Download counts

	3.6	3.7	3.8	3.9	3.10	3.11	3.12	3.13	3.14	Total
3.7.0	0	4	2	18	3	5	3	3	0	38
3.7.1	0	489	19,604	499	901	69	19,799	15	0	41,376
3.8.0	1	442	953	602	2,227	8,080	761	48	0	13,114
3.9.0	6	2,432	6,185	6,290	15,845	30,465	303,682	495	4	365,404
3.10.0	0	1	4	28	54	5	72	32	0	196
3.10.1	2,389	19,651	122,668	4,415,456	943,626	617,362	1,104,001	185,074	1,698	7,411,925
3.11.0	300	2,955	23,444	572,580	170,023	182,477	99,939	60,592	837	1,113,147
Total	2,696	25,974	172,860	4,995,473	1,132,679	838,463	1,528,257	246,259	2,539	8,945,200

Download proportions

	3.6	3.7	3.8	3.9	3.10	3.11	3.12	3.13	3.14	Total
3.7.0	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%
3.7.1	0.00%	0.01%	0.22%	0.01%	0.01%	0.00%	0.22%	0.00%	0.00%	0.46%
3.8.0	0.00%	0.00%	0.01%	0.01%	0.02%	0.09%	0.01%	0.00%	0.00%	0.15%
3.9.0	0.00%	0.03%	0.07%	0.07%	0.18%	0.34%	3.39%	0.01%	0.00%	4.08%
3.10.0	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%
3.10.1	0.03%	0.22%	1.37%	49.36%	10.55%	6.90%	12.34%	2.07%	0.02%	82.86%
3.11.0	0.00%	0.03%	0.26%	6.40%	1.90%	2.04%	1.12%	0.68%	0.01%	12.44%
Total	0.03%	0.29%	1.93%	55.85%	12.66%	9.37%	17.08%	2.75%	0.03%	100%

Download counts (no Python 3.9 / Flit 3.10.1 outlier)

	3.6	3.7	3.8	3.9	3.10	3.11	3.12	3.13	3.14	Total
3.7.0	0	4	2	18	3	5	3	3	0	38
3.7.1	0	489	19,604	499	901	69	19,799	15	0	41,376
3.8.0	1	442	953	602	2,227	8,080	761	48	0	13,114
3.9.0	6	2,432	6,185	6,290	15,845	30,465	303,682	495	4	365,404
3.10.0	0	1	4	28	54	5	72	32	0	196
3.10.1	2,389	19,651	122,668	50,000	943,626	617,362	1,104,001	185,074	1,698	3,046,469
3.11.0	300	2,955	23,444	572,580	170,023	182,477	99,939	60,592	837	1,113,147
Total	2,696	25,974	172,860	630,017	1,132,679	838,463	1,528,257	246,259	2,539	4,579,744

Download proportions (no Python 3.9 / Flit 3.10.1 outlier)

	3.6	3.7	3.8	3.9	3.10	3.11	3.12	3.13	3.14	Total
3.7.0	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%
3.7.1	0.00%	0.01%	0.43%	0.01%	0.02%	0.00%	0.43%	0.00%	0.00%	0.90%
3.8.0	0.00%	0.01%	0.02%	0.01%	0.05%	0.18%	0.02%	0.00%	0.00%	0.29%
3.9.0	0.00%	0.05%	0.14%	0.14%	0.35%	0.67%	6.63%	0.01%	0.00%	7.98%
3.10.0	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%
3.10.1	0.05%	0.43%	2.68%	1.09%	20.60%	13.48%	24.11%	4.04%	0.04%	66.52%
3.11.0	0.01%	0.06%	0.51%	12.50%	3.71%	3.98%	2.18%	1.32%	0.02%	24.31%
Total	0.06%	0.57%	3.77%	13.76%	24.73%	18.31%	33.37%	5.38%	0.06%	100%

Flit

Download counts

	3.6	3.7	3.8	3.9	3.10	3.11	3.12	3.13	3.14	Total
3.7.0	0	0	1	5	5	5	5	1	0	22
3.7.1	9	221	19,327	300	373	20	208	15	0	20,473
3.8.0	2	40	96	40	1,571	3,013	3,137	9	0	7,908
3.9.0	216	2,842	762	938	1,967	2,026	12,127	135	0	21,013
3.10.0	0	0	1	16	24	9	13	0	0	63
3.10.1	0	1	6,047	2,773,411	122,244	74,641	26,640	10,530	100	3,013,614
3.11.0	0	0	1,770	47,344	13,526	8,204	3,312	1,927	52	76,135
Total	227	3,104	28,004	2,822,054	139,710	87,918	45,442	12,617	152	3,139,228

Download proportions

	3.6	3.7	3.8	3.9	3.10	3.11	3.12	3.13	3.14	Total
3.7.0	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%
3.7.1	0.00%	0.01%	0.62%	0.01%	0.01%	0.00%	0.01%	0.00%	0.00%	0.65%
3.8.0	0.00%	0.00%	0.00%	0.00%	0.05%	0.10%	0.10%	0.00%	0.00%	0.25%
3.9.0	0.01%	0.09%	0.02%	0.03%	0.06%	0.06%	0.39%	0.00%	0.00%	0.67%
3.10.0	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%
3.10.1	0.00%	0.00%	0.19%	88.35%	3.89%	2.38%	0.85%	0.34%	0.00%	96.00%
3.11.0	0.00%	0.00%	0.06%	1.51%	0.43%	0.26%	0.11%	0.06%	0.00%	2.43%
Total	0.01%	0.10%	0.89%	89.90%	4.45%	2.80%	1.45%	0.40%	0.00%	100%

Download counts (no Python 3.9 / Flit 3.10.1 outlier)

	3.6	3.7	3.8	3.9	3.10	3.11	3.12	3.13	3.14	Total
3.7.0	0	0	1	5	5	5	5	1	0	22
3.7.1	9	221	19,327	300	373	20	208	15	0	20,473
3.8.0	2	40	96	40	1,571	3,013	3,137	9	0	7,908
3.9.0	216	2,842	762	938	1,967	2,026	12,127	135	0	21,013
3.10.0	0	0	1	16	24	9	13	0	0	63
3.10.1	0	1	6,047	50,000	122,244	74,641	26,640	10,530	100	290,203
3.11.0	0	0	1,770	47,344	13,526	8,204	3,312	1,927	52	76,135
Total	227	3,104	28,004	98,643	139,710	87,918	45,442	12,617	152	415,817

Download proportions (no Python 3.9 / Flit 3.10.1 outlier)

	3.6	3.7	3.8	3.9	3.10	3.11	3.12	3.13	3.14	Total
3.7.0	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	0.01%
3.7.1	0.00%	0.05%	4.65%	0.07%	0.09%	0.00%	0.05%	0.00%	0.00%	4.92%
3.8.0	0.00%	0.01%	0.02%	0.01%	0.38%	0.72%	0.75%	0.00%	0.00%	1.90%
3.9.0	0.05%	0.68%	0.18%	0.23%	0.47%	0.49%	2.92%	0.03%	0.00%	5.05%
3.10.0	0.00%	0.00%	0.00%	0.00%	0.01%	0.00%	0.00%	0.00%	0.00%	0.02%
3.10.1	0.00%	0.00%	1.45%	12.02%	29.40%	17.95%	6.41%	2.53%	0.02%	69.79%
3.11.0	0.00%	0.00%	0.43%	11.39%	3.25%	1.97%	0.80%	0.46%	0.01%	18.31%
Total	0.05%	0.75%	6.73%	23.72%	33.60%	21.14%	10.93%	3.03%	0.04%	100%

[gotmax23]

We still maintain a flit-core RPM in EPEL 8 (Python 3.6) which is supported until 2029 but not reflected in the PyPI statistics, so take that for what it's worth.

[AA-Turner]

By 2029 Python 3.13 will be end-of-life! Does your package attempt to update to new versions of flit(-core)?

I imagine it would be fairly easy to patch out 3.9+ features (we won't be using match/case soon), but my personal opinion would be not to factor downstream redistributors into the support decision -- they have their own policies for their own audiences.

A

[takluyver]

Thanks, that's good to know about, though we're not likely to keep supporting Python 3.6 upstream for another 5 years.

What's your general plan when packages drop the Python version you're on? I think I'm already something of an outlier in maintaining support for as long as I do (compare Numpy's policy), so you've presumably had this for a lot of packages already. Do you stick with an old version and try to backport specific changes that seem important enough? Or do you take newer versions and modify them to get around compatibility issues? Or a mixture?

[gotmax23]

Thanks, that's good to know about, though we're not likely to keep supporting Python 3.6 upstream for another 5 years.

Definitely. I just wanted to point out that there were other people still using 3.6.

What's your general plan when packages drop the Python version you're on?

Generally, we avoid updating Python libraries in stable releases unless there's a strong reason to do so (e.g., security fixes, major bugfixes, and update that's needed by another package, user-requested features that don't come with other incompatible changes). By the time an upstream drops compatibility for the Python version we are using, we usually just manually backport specific changes, but this is not always straightforward.

[cdce8p]

IMO it would make sense to drop support for 3.6 and 3.7. If we say that support for flit-core should be longer than flit, would it make sense to say

• flit -> all non-EOL Python versions
• flit-core -> all non-EOL Python version + 1 / 2 minor releases?

[takluyver]

I think based on the numbers @AA-Turner showed, it's definitely reasonable to drop 3.6 at this point, and probably 3.7 as well.

I'd prefer to go by usage numbers like these - and considering how much additional effort it takes to test & maintain support for older Python versions - rather than a rule about how many versions are supported or for how long. Although I can't imagine we'd ever drop flit-core's support for a Python version before its EOL. Perhaps 1% or 0.5% is a reasonable, if arbitrary threshold to consider - as in, we don't normally drop a Python version before it reaches that level, not we automatically drop it as soon as it does.

As an aside: I'm not sure it's meaningful to squash the 'outlier' numbers for Python 3.9 & Flit 3.10.1. It looks like there are consistently a lot of downloads on Python 3.9 for whatever reason, not just a one-off glitch in the numbers. Of course, that's very likely something highly automated which it might be easy to change: download counts are not a great proxy for how much pain would be caused by dropping support for a given Python version. (I am pleased that we're thinking about the numbers and how to make sense of them, not just accepting them as gospel, though 👍 )

Another rough indicator is what Python versions are in major Linux distros. In this case, no still supported Ubuntu is affected (20.04 already had Python 3.8), nor officially supported Debian (although versions with third-party extended long term support are), while RHEL 7 (two major versions behind current, in 'extended lifecycle support') & derivatives would be affected. That also suggests it's reasonable to drop 3.6 & 3.7 now.