Add Test runs for Python 3.7 and remove 3.4

[crwilcox]

There are very few people using 3.4 and 3.7 is coming up. A few of the packages cannot use 3.7 yet

PubSub: #5292
Monitoring: #5293
BigQuery: #5294

[tswast]

Nice

[tseaver]

Are you sure about the 3.4 bit? It is still the default python3 for LTS Linux versions, AFAIK.

[crwilcox]

@tseaver: @jonparrott looked up the stats and said ~4%. We can leave it if you have strong feelings about it. It seemed a convenient time to not elongate the test runs is all.

[theacodes]

Let me double check that... I am bad at math.

…

On Thu, May 3, 2018, 1:47 PM Christopher Wilcox ***@***.***> wrote: @tartavull <https://github.com/tartavull> @jonparrott <https://github.com/jonparrott> looked up the stats and said ~4%. We can leave it if you have strong feelings about it. It seemed a convenient time to not elongate the test runs is all. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#5295 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAPUc1R6tt34CdAELzv52RBmSz11o7oUks5tu2zIgaJpZM4TxuS7> .

[theacodes]

Okay, looking over the last 6 months of download stats from BigQuery:

  SELECT
    details.python, COUNT(*) as total_downloads,
  FROM
    TABLE_DATE_RANGE(
      [the-psf:pypi.downloads],
      DATE_ADD(CURRENT_TIMESTAMP(), -6, "month"),
      CURRENT_TIMESTAMP()
    )
  WHERE
    file.project LIKE 'google-cloud%'
  GROUP BY details.python
  ORDER BY details.python DESC

Our packages have been installed a total of 54,568,855 times with 785,977 of those downloads originating from a Python 3.4 version. That means that Python 3.4 accounts for just 1.44% of our install base. By comparison, 3.4 accounts for 3.58% of the overall installs for all packages over the last 6 months.

Summoning @tswast to make sure I didn't do something stupid here.

[tswast]

Query looks right, though I'd filter out the "bandersnatch" installs with details.installer.name = 'pip' if I were you. In my experiments those are a significant percentage of downloads and don't actually count as installs since they mirror all of PyPI.

P.S. I could give you some help updating to standard SQL syntax if you'd like.

[theacodes]

Thanks @tswast. Limiting to just pip installs made the Py 3.4 percentage even lower at 1.43%. ;)

[tswast]

As pointed out by @tseaver in #5401, this also requires an update to all setup.py files.