All commit authors signed the Contributor License Agreement.

Most changes to Pythonrequire a NEWS entry. Add one using theblurb_it web app or theblurb command-line tool.

If this change has little impact on Python users, wait for a maintainer to apply theskip news label instead.

I think we need a separate nodule for this because if we want to have a smarter completion later the code is better to be isolated, so something like sqlite3._completer.

If possible, can we have anautogenerated list of keywords? and those keywords could also be stored in a (private) module-level list.

Please add tests

Yes, tests is coming. I will try to find a way to autogenerate list of keywords.

You can mark the pr as a draft.

Added tests and moved keyword list to module level.

Keywords seems not easy to auto-generate, theSQLite document says they can be accessed bysqlite3_keyword_count() andsqlite3_keyword_name() using C, but I am bad at C and didn't figure out how to call these two at Python level. I think these two functions are not exposed to Python?

Hi@picnixz, as Stan mentioned, beta freeze is around the corner, but I really hope this can be shipped in 3.14, do you think there is a chance to merge it before beta freeze?

On my side it looks great but I'll also want@erlend-aasland to have a look as he's the main maintainer. Maybe he'l ask for some stylistic changes (in which case I'm sorry if I suggested something wrong).

Thank you so much for your review. Really appreciate your guidance!

Thetest_completion_order() fails. TheSQLITE_KEYWORDS given bysqlite3_keyword_name() andsqlite3_keyword_count() is not in alphabetical order, but GNU Readline always sorts completions before displaying them.

Unfortunately I didn't find adisable-completion-sorting-like option in GNU Readline's document. There is anrl_sort_completion_matches in Readline's source code but it's not exposed as a config flag.

Can we remove thistest_completion_order(), as the order of displayed candidates is controlled by GNU Readline.

Let's just check that the completion of nothing returns all keywords. You can sort theSQLITE_KEYWORDS list before checking for the equality (I think it's better not to sort it at the C level as it could be helpful in the future for someone interacting with python and C)

🤖 New build scheduled with the buildbot fleet by@encukou for commit8d4f659 🤖

Results will be shown at:

https://buildbot.python.org/all/#/grid?branch=refs%2Fpull%2F133393%2Fmerge

If you want to schedule another build, you need to add the🔨 test-with-buildbots label again.

There are failures on FreeBSD, and onUbuntu and someFedora/RHEL configurations :(
@tanloong, do you want to investigate these?

Yes, I will look into it.

FreeBSD: When usingrun_pty() on FreeBSD, thexyzzy\t\t\b\b\b\b\b.quit\n does not quit the session. Instead of doing nothing on\t\t when there are no matching completions, two actual tab characters are inserted. Because of these extra tabs, the five backspaces\b\b\b\b\b delete characters to the left but leave the first two charactersxy intact. This results in the commandxy.quit\n, which is seen not as a dot command (due to not starting with dot) but an incomplete SQLite statement (due to the lack of trailing semicolon),.quit does not take effect and the session hangs.

It is mentioned in other issues as well that therun_pty() has platform specific problems (issue1,issue2). How about we skiptest_nothing_to_complete() on FreeBSD?

Ubuntu and Fedora/RHEL: Could not reproduce on my virtual machines. I suspect that theenv={"NO_COLOR": "1"} may not disable color output on buildbots, so thesqlite> prompt has surrounding control sequences, causingline.startswith(f"{self.PS1}xyzzy") to always returnFalse.

The subprocess created byrun_pty() might inherit some related environment variables from the main process. I want to open a terminal on buildbots for manual debugging but have not found a way to do that yet. I added the@force_not_colorized_test_class decorator to theCompletionTest class to disable coloring in the main process and will see if that helps.