BigQuery: removes support for python 3.4

[alixhami]

Addresses #4587

I queried the pypi dataset to see which versions of Python have been used with google-cloud-bigquery for the last 6 months.

python version	% of downloads
2.7	91.3%
3.6	4.1%
3.5	2.8%
3.4	1.5%
null	0.3%
other	0.01%

[dhermes]

@alixhami It's probably worth

• Updating the docs immediately
• Making a note in CHANGELOG.md in this PR (so that it's clear in the next release, but worded now)

[alixhami]

@dhermes What should be changed in the docs immediately? I thought we generated the changelog in the release PR with the names of PRs since the last release.

[dhermes]

What should be changed in the docs immediately?

Just add a note that 3.4 is no longer supported.

I thought we generated the changelog in the release PR with the names of PRs since the last release.

CHANGELOG.md is a living file, you can edit it whenever you like. We have not yet edited them outside of release PRs, but there's no hard and fast rule. You could just add a section with an ambiguous version (e.g. 0.x.y) and then put one bullet in there mentioning this PR

[dhermes]

LGTM but I'd feel better if @jonparrott had a peek

[lukesneeringer]

It makes me sad that Python 2.7 is still 91% of our usage.

I am not sure I am comfortable with this change; can we discuss it? In particular:

• I think we should have consistent support across APIs. So, if we are dropping Python 3.4 in BigQuery, we should drop it everywhere.
• Even with low usage, dropping a Python version is probably a semver-major change. (BigQuery is not 1.0 yet, but see above.)
• Ideally I would like to tie version support to the EOL dates on the Python versions. This is especially important for when it does come time to drop Python 2.7. I am going to want to point to that bright, shiny, "Python 2.7 is EOL in 2020" sign, and that case is weaker if we are willy nilly about other versions.
  • I get that this is tricky because the Python community is unfortunately amorphous about EOLs.

[theacodes]

It was my idea. ;)

[lukesneeringer]

I would like to discuss this before we do it.

[tseaver]

@alixhami I don't actually think those stats mean what you think they mean: it would be more in line with them to drop all Python3 support.

Is Pandas "core" to BigQuery's mission? If not, let's not have the tail wag the dog here.

[alixhami]

It sounds like the Python 3.4 support decision should be made at a higher level than the BigQuery client library, so I created PR #4681 to address just the BigQuery/Pandas side of things