Switch to using the other style of environment markers#3796
Conversation
|
Note to self: branch-free setup.py is a prerequisite for shipping py3-stable-abi wheels. |
|
@dstufft uhhh, any idea why this would result in a package not getting installed under |
|
I know very little about |
9396cdf to
63e6719
Compare
|
@reaperhulk looking at the first 2 days of the month of august, we're down to 5.9% of installs being with pip 1.x |
|
Now with a week of data, it's down to 5.5%: SELECT
STRFTIME_UTC_USEC(timestamp, "%Y-%m") AS yyyymm,
ROUND(100 * SUM(CASE WHEN REGEXP_MATCH(details.installer.version, r"^1\.") THEN 1 ELSE 0 END) / COUNT(*), 1) AS percent_pip1,
COUNT(*) as download_count
FROM
TABLE_DATE_RANGE(
[the-psf:pypi.downloads],
DATE_ADD(CURRENT_TIMESTAMP(), -1, "year"),
CURRENT_TIMESTAMP()
)
WHERE
details.installer.name == "pip" AND file.project == "cryptography"
GROUP BY
yyyymm
ORDER BY
yyyymm DESC
LIMIT 100 |
|
What does this do to users who install via setuptools rather than via pip? |
|
setuptools version mumble mumble and above work fine. |
|
(IIRC, it's version 0.7, incidentally the same version that started verifying shit in TLS, so I don't care about 0.6 at all) |
|
So the setuptools version is actually substantially older than the pip version. 4 years to pip 6's coming-up-on-3 |
|
I believe that's correct, yes. |
|
@reaperhulk breaking setuptools 0.6 is arguably an important security measure :-) |
|
Running that query gives 6.4% for August now. I think I'm convinced, especially since this won't affect setuptools unless you're on a ludicrously old version and all that happens in the worst case is some deps don't get installed. |

No description provided.