MAINT: Drop Python3.7 from supported versions.

[charris]

This drops Python 3.7 for NumPy 1.22.x. The following changes are made

• Python 3.7 is dropped from testing
• Setup.py is adjusted with Python 3.10 added
• Some small code cleanups

Note that testing of gcc48 is dropped. Does that matter anymore?

EDIT: I temporarily left the version check in setup.py to 3.7+ to work around LGTM. If LGTM isn't updated by release time we will need to drop it.

[charris]

LGTM is going to be a problem, it is stuck on ubuntu 19.10 which has Python 3.7 as the default. They seem to be falling behind the latest releases pretty badly and I can see this being a continuing problem going forward. Should we suspend it? That might cause a long wait if we enable it again, as I think it plays catch up by doing all the missed runs. Is there something better? Do we need it?

[mattip]

I think we should drop LGTM. Was there a PR that was improved due to its output over the past year or so?

[charris]

Was there a PR that was improved due to its output over the past year or so

I can't think of one :) The alerts are good for code cleanup, but I think there are standalone applications that might do as well. AFAICT, Semmle is more interested in their corporate version. Let me make a post on the mailing list to see if anyone complains.

[charris]

@jhelie Do you still have any connection with LGTM? If so, could you comment on the chance of having Python 3.8+ available for the C builds?

[jhelie]

Hi @charris, thanks for the ping. Yes, let me see if I can get a member of the Python team to get back to you on this.

[mattip]

Hopefully we can turn this back on with pypy-3.8 soonish

[charris]

Yes, that's why I commented it out instead of deleting.

[mattip]

Note that testing of gcc48 is dropped. Does that matter anymore?

I think gcc 4.8.5 is still the default version on CentOS7 (EOL 2024-07)see issue #14147. With that, it is pretty easy to get a more recent version of gcc. Maybe we should leave that one commented out for a while rather than remove it, just in case we get pushback.

[charris]

Maybe we should leave that one commented out for a while

Done.

[mattip]

I think this will break downstream packages wanting to test our latest builds against their packages, so they too will have to drop 3.7 right?

[charris]

I think this will break downstream packages wanting to test our latest builds

They can't use the latest builds running 3.7. Pandas, IIRC, is dropping 3.7 in their next release. By the schedule, anyone releasing after Dec 26 should be dropping 3.7.

EDIT: I already dropped 3.7 nightly builds, I should probably drop earlier builds so that we can elicit complaints.

[RasmusWL]

Do you still have any connection with LGTM? If so, could you comment on the chance of having Python 3.8+ available for the C builds?

Hi 👋 I'm part of the team at GitHub that works on CodeQL and maintains LGTM.com (GitHub bought Semmle, the company which ran LGTM.com, in 2019). We're planning to upgrade the dev environment to Ubuntu 20.04 at some point in this quarter, which will make Python 3.8 available. I have 2 ideas for workarounds until then:

1. You could manually install Python 3.8+ as part of each C++ analysis with an after_prepare step
2. You could use Code Scanning instead, which is the spiritual successor to LGTM.com fully integrated into GitHub (this is free to use for public repos). With a minor bit of configuration, it's possible to make it run the same queries that is used on LGTM.com. Since it's targeted for security alerts, it currently only allows access to the alerts for people with write permission (docs) -- I'm not sure how well that fits your main use-case of helping out with code cleanup, but that will be for you to decide 😊

[charris]

@RasmusWL Thanks for getting back to us. It sounds like waiting for the upgrade is the easiest path forward. What are your plans for future upgrades? The Python scientific stack has settled on supporting Python versions for around 42 months, so we will be dropping older versions on a yearly basis if Python continues with yearly releases.

[RasmusWL]

Indeed, you could wait until the upgrade to 20.04. To have greater flexibility and choice in the OSes that are used for the analysis, I would recommend that you switch to GitHub Code Scanning. It uses the same CodeQL engine and queries, and integrates straight into your development workflow on GitHub.

(sorry for the late reply, I was down with heavy fever after getting my covid vaccine)