Python 3.6. as well as Python 3.7 is EOL. For Pywikibot it is planed to drop Python 3.6 support soon but we should have a data base how often outdated Python releases are still used. It would be useful to have some statistics for that.
See older requests made in T242157, T286867 T312794 for previous release changes.
| Status | Subtype | Assigned | Task | ||
|---|---|---|---|---|---|
| Resolved | Dalba | T228375 Deprecate test_family and force wikipedia_family for test.wikipedia.org instead | |||
| Resolved | BUG REPORT | Xqt | T228335 multiple matches order of _code_fam_from_url depends on Python version | ||
| Resolved | Release | Xqt | T347020 Pywikibot 9 tasks (Tracking) | ||
| Open | None | T347031 Unintended leak of Proxy-Authorization header in requests | |||
| Resolved | JJMC89 | T340640 pkg_resources is deprecated as an API | |||
| Resolved | Xqt | T347026 Drop support for Python 3.6 | |||
| Resolved | Xqt | T350568 Provide some pywikibot usage statistics for Python3.6 and 3.7 |
I think the last step for you to get access to the data yourself is answering the T348520#9335272 question at the NDA task (and confirming the e-mail address you have currently set in LDAP). As far as I can see, all the relevant approvals are already recorded on the task. Sorry it is a complicated process – fortunately, it's something that only needs to be used once.
If there's actually more things to complete on the NDA task that I'm not seeing, I can take a look at generating the stats for you myself, but if the NDA access for you is as close as I think it is, then I'd prefer you answering that question and getting access to it yourself (if needed, I'd be more than happy to show you around, but the interface is fairly intuitive). Let me know.
stat1005 is not accessible to you at this point, but having access to the stat servers is not strictly needed to get a reasonable usage overview. Since the NDA request is now granted, you should be able to follow my steps from T242157#5785171, where I used the webrequest_sampled_128 dataset (1/128 of all Wikimedia webrequests) in https://turnilo.wikimedia.org/. While it will not give you data about all of the users, because of the sampling and because Turnilo only shows a limited amount of top user agents, I think for deprecation purposes it should be sufficient.
Here are some statistic data fro Pywikibot usage [%]:
| Pywikibot | 8 | 7 | 6 | 5 | 4 | 3 | Sum |
|---|---|---|---|---|---|---|---|
| ALL | 70 | 11 | 14 | 1 | 5 | 100 | |
| Python 3.11 | 12 | 12 | |||||
| Python 3.10 | 10 | 3 | 3 | 15 | |||
| Python 3.9 | 10 | 5 | 9 | 24 | |||
| Python 3.8 | 10 | 1 | 1 | 12 | |||
| Python 3.7 | 22 | 1 | 1 | 1 | 26 | ||
| Python 3.6 | 4 | 1 | 1 | 6 | |||
| Python 2.7 | 5 | 5 | |||||
There are still 4% using Python 3.6 with Pywikibot 8. Most of them are made by the https://www.marefa.org project reading data from commons (I guess) by using Pywikibot 8.4. Excluding these 200k requests the remaining usage for Python 3.6 with Pywikibot 8 is 0.6% only, most of them is caused by Pywikibot testing.
I haven't found any usage for Python 3.6 with Pywikibot 8.5 and 8.6 except Pywikibot-test requests. Therefore dropping Python 3.6 can be done now safely in pariculary because older Pywikibot releases aren't affected with such change.
@Urbanecm: I wasn't able to exclude Pywikibot-test from User Agent regex Pywikibot/8\.[56].+Python/3\.6 e.g. with (?!test) but get a 504 error. Any idea?
Besides filtering those agents out during post-processing, not really. 504 indicates a timeout, which makes sense, given negative lookahead is not exactly a cheap query. Granted, removing those agents during postprocessing means you lose certain UAs, but since Turnilo currently supports up to 10k of top user agents, and the least frequent UAs have 1 hit per last month, it should be good enough.