PyREPL: Different Completion Behavior with GNU Readline

[tanloong]

Bug report

Bug description:

The current completion logic in Lib/_pyrepl/completing_reader.py uses a suffix-insertion strategy (source code):

In the contrast, Readline completely replaces user input with the candidate (source code).

This will lead to unwanted behavior in case-insensitive completion scenario. If user input 'xy', they might get 'xyZZY' instead of the expected 'XYZZY'.

CPython versions tested on:

CPython main branch

Operating systems tested on:

Linux

Linked PRs

• gh-145392: Fix _pyrepl completion to replace input stem instead of appending #145393

[encukou]

Do you have a reproducer?
Is this needed/helpful for SQLite?

[tanloong]

Yes, this will be helpful for SQLite.

After seeing #145379, I wondered if the SQLite CLI could also use PyREPL. That would greatly improve the experience on Windows. I had an LLM create a quick demo of the SQLite CLI using PyREPL.

In that demo, when I type sel and press Tab, it completes to selECT instead of the correct SELECT.
I will push the demo branch as a reproducer soon.

The demo is far from ready-to-use, and there are also syntax highlighting issues (maybe could not solved until #141619 is merged). But I believe the SQLite CLI will eventually move to PyREPL (we might not need to modify the sqlite module directly but instead update the underlying code module to use PyREPL). Changing the completion to use a replace-strategy instead of suffix-insertion will help with this. I'm fine with pausing this issue until the SQLite CLI moves to PyREPL, and then treat it as a SQLite bug only after that.

[tanloong]

This is the demo branch as a reproducer: https://github.com/tanloong/cpython/tree/sqlite-pyrepl

In the screen recording, when typing sel and press Tab, it completes to selECT instead of SELECT.

simplescreenrecorder-2026-04-01_21.46.35.mp4

[encukou]

Thanks! That makes sense.

I'm fine with pausing this issue until the SQLite CLI moves to PyREPL, and then treat it as a SQLite bug only after that.

Let's do that. There is no issue in current use of pyrepl, and if sqlite3 starts using it, it'll only be a small inconvenience.

[encukou]

I'll close for now; let's reopen when this becomes helpful.