Testing client

[ahopkins]

New test_client based upon request-async for usage with #1475

[yunstanford]

Looks like request-async only supports Python 3.6+.

We'll have to drop Py3.5 support if merging this.

[ahopkins]

Yeah, I saw that last night after I saw the first test fail. I was thinking of taking a look at it and then hoping that (1) it was a simple fix like F strings and (2) @tomchristie would accept a PR to make it 3.5. If not, then we @huge-success/sanic-core-devs have a decision to make.

[tomchristie]

Great work @ahopkins!

The biggest 3.5 blocker will be async generators syntax. https://www.python.org/dev/peps/pep-0525/

async for item in ...:

It looks to me like sanic doens't yet use this syntax in any of it's API, since I can't see any (older style). __aiter__ or __anext__, and since the streaming request/response docs use loop style APIs rather than async iterator styles.

It'd be a bit of a pain to backport to 3.5 for those. It's feasible, but it'd likely be something that I'd still want to drop as soon as possible.

Personally I'd probably advocate for this being a good time to consider a version with 3.6+ support, but that's for you folks to take a call on.

[yunstanford]

I'm fine with dropping py3.5 support, but probably should announce that first or shows deprecation time.

[ahopkins]

I too would be in favor of dropping 3.5, but it does seem to account for a decent portion of downloads.

This is based off stats from 2019-01-01 thru today.

python_version	downloads
3.6	102,431	41%
3.7	77,564	31%
3.5	42,955	17%
null	25,451	10%
2.7	2,426	1%
3.4	247	0%
3.8	49	0%
2.6	3	0%
TOTAL	251,126

SELECT
  REGEXP_EXTRACT(details.python, r"([0-9]+\.[0-9]+)") AS python_version,
  COUNT(*) AS downloads
FROM TABLE_DATE_RANGE(
  [the-psf:pypi.downloads],
  TIMESTAMP("2019-01-01"),
  CURRENT_TIMESTAMP()
)
WHERE file.project="sanic"
GROUP BY python_version
ORDER BY downloads DESC

[tomchristie]

Are you able to limit the results to only include downloads against the current sanic release?

That'd be more indicative of what Python versions are typically being used by new projects (rather than including deployments that may be pinned to older versions anyways).

[tomchristie]

(I'm arguing backwards there, because I think it'd be a good version jump time anyways, but hey. 😆)

[ahopkins]

When I filter for 18.12.0 and 19.3.1, it does drop from 17% to 9% You are right.

python_version	downloads
3.7	57852	40%
3.6	55204	38%
null	17201	12%
3.5	12368	9%
2.7	1779	1%
3.4	151	0%
3.8	48	0%
2.6	2	0%
TOTAL	144,605

SELECT
  REGEXP_EXTRACT(details.python, r"([0-9]+\.[0-9]+)") AS python_version,
  COUNT(*) AS downloads
FROM TABLE_DATE_RANGE(
  [the-psf:pypi.downloads],
  TIMESTAMP("2019-01-01"),
  CURRENT_TIMESTAMP()
)
WHERE file.project="sanic"
AND (file.version="18.12.0" OR file.version="19.3.1")
GROUP BY python_version
ORDER BY downloads DESC

[ahopkins]

See further discussion about dropping 3.5 support here: https://community.sanicframework.org/t/should-we-bump-the-minimum-python-required-version-to-3-6/238/6?u=ahopkins

[andreymal]

Debian stable don't have py3.6+ yet, so I think this is not a good idea to drop py3.5

[ahopkins]

@andreymal The current direction we are leaning towards is to drop 3.5 going forward, but extending our 18.12LTS support thru the 3.5 end of life in September 2020.

[codecov]

Codecov Report

Merging #1562 into master will decrease coverage by 0.08%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master    #1562      +/-   ##
==========================================
- Coverage    91.4%   91.31%   -0.09%     
==========================================
  Files          18       18              
  Lines        1804     1820      +16     
  Branches      344      347       +3     
==========================================
+ Hits         1649     1662      +13     
- Misses        131      132       +1     
- Partials       24       26       +2

Impacted Files	Coverage Δ
sanic/testing.py	96.03% <100%> (+0.74%)	⬆️
sanic/request.py	97.36% <0%> (-1.32%)	⬇️
sanic/router.py	95.43% <0%> (-0.46%)	⬇️
sanic/app.py	92.47% <0%> (+0.23%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 6a4a3f6...ccd4c96. Read the comment docs.

[yunstanford]

LGTM, but let’s squash commits and add concrete commit msg, so that it helps generate changelog.

[seemethere]

Agree with @yunstanford, let's squash these commits then it's a LGTM for me as well.

[ahopkins]

Will do. But there are a few more changes I want to make. Some tests use aiohttp directly and need to be refactored or else we will need both libraries, which seems bloated to me. Still WIP for current moment.

I'll finish shortly.

[ahopkins]

@yunstanford and @seemethere This should be ready to roll assuming all of the tests pass. 🤞

[ahopkins]

FYI - To help you in your review since everything is squashed now...

The files that changed since the last time are tests/test_keep_alive_timeout.py and tests/test_request_timeout.py, with a minor change to sanic/testing.py.