[BUG] autocomplete for tables with schema behaves unexpectedly

[nktnet1]

Description

When typing in the query, if we use autocomplete for tables with a schema, the schema gets duplicated (appears twice).

Steps to Reproduce

1. Set up the database:

   CREATE SCHEMA IF NOT EXISTS sample;
   
   CREATE TABLE IF NOT EXISTS sample.demo (
       id INT PRIMARY KEY,
       name TEXT
   );
   
   INSERT INTO sample.demo VALUES
       (1, 'Test');
   
   SELECT *
   FROM sample.demo;

2. Now type in the query

   SELECT * FROM sample.dem

   

3. Perform the autocomplete by clicking the suggestion or hitting TAB/ENTER on your keyboard

4. The query becomes

   SELECT * FROM sample.sample.demo

   

   i.e. the schema name is duplicated

Expected Behavior

It should autocomplete to sample.demo

Actual Behavior

It autocompleted to sample.sample.demo

Environment

• LibreDB Studio Version: (e.g., 0.15.0)

[mikevillari]

I am taking this completion fix. I will reproduce the accepted completion text from the provider replacement range, cover schema-qualified table prefixes and existing table/alias column completions, and keep the change within the completion provider and its tests. AI-assisted work; I will report validation results and environment limitations in the PR.

[cevheri]

Reproduced, and your report turned out to describe two defects rather than one.

The duplication you screenshotted happens because Monaco replaces only the word under the cursor, dem, while the provider inserts the full sample.demo, so the already typed sample. survives in front of it. The second one is easier to miss: the suggestion list filters on that same bare word, so sample.de and sample.dem match nothing at all and the popup goes empty. Your screenshot at step 2 shows exactly that state, with a single entry left.

@mikevillari has a fix in review at #715. It is not merged yet: reviewing it surfaced a case where table labels carry no schema, which several engines produce, and that path needs to keep working before this lands.

Nothing for you to do here, this is just so the thread is not silent. Reporting the intermediate keystrokes and not only the final result is what made the second defect visible, so thank you for the detail.