|
9 | 9 | - Users can decide to switch between autocomplete and next edit. |
10 | 10 | - Next edit triggers at the same time autocomplete is triggered, via vscode's inline completion provider. |
11 | 11 | - The following happens after the trigger: |
12 | | - - User's current cursor position is captured. |
13 | | - - We define an editable range, ±5 lines from the current cursor position. |
14 | | - - User's most recent edit is captured as a unified diff. (this is currently buggy) |
15 | | - - This is sent to the model, which returns a new editable range with next edit predictions. |
16 | | - - We display this new editable range in a SVG decoration. |
17 | | - - User can either tab to accept or esc to reject. |
18 | | - - On accept, the old editable range will be replaced by the new editable region. The cursor will be moved to the last line containing some change. |
19 | | - - On reject, nothing happens. |
| 12 | + - User's current cursor position is captured. |
| 13 | + - We define an editable range, ±5 lines from the current cursor position. |
| 14 | + - User's most recent edit is captured as a unified diff. (this is currently buggy) |
| 15 | + - This is sent to the model, which returns a new editable range with next edit predictions. |
| 16 | + - We display this new editable range in a SVG decoration. |
| 17 | + - User can either tab to accept or esc to reject. |
| 18 | + - On accept, the old editable range will be replaced by the new editable region. The cursor will be moved to the last line containing some change. |
| 19 | + - On reject, nothing happens. |
20 | 20 |
|
21 | 21 | ## What needs to be worked on? |
22 | 22 |
|
23 | 23 | - User edit captures. |
24 | 24 | - Find a better way to trigger next edit (this links back to the diff capture problem). |
25 | | - - We can see that next edit triggers as soon as the user accepts a change. This is because autocomplete runs the same way. |
26 | | - - I think autocomplete has some filter logic that doesn't display the ghost text under some conditions, which I am guessing are the following: |
27 | | - - The model does not have any more completions to create. |
28 | | - - The prediction at the cursor location has been cached. |
| 25 | + - We can see that next edit triggers as soon as the user accepts a change. This is because autocomplete runs the same way. |
| 26 | + - I think autocomplete has some filter logic that doesn't display the ghost text under some conditions, which I am guessing are the following: |
| 27 | + - The model does not have any more completions to create. |
| 28 | + - The prediction at the cursor location has been cached. |
29 | 29 | - JetBrains integration. |
0 commit comments