diff: add a true in-place "inline" overlay (virtual text) as the future layout = "inline"
- Dominant language
- Lua
- Stars
- 3.1k
- Forks
- 216
- PR merge metrics
- No merged PRs in 30d
Description
## Summary
Follow-up to #293 / #195. The `diff_opts.layout = "unified"` option (renamed from `"inline"` before v0.4.0) renders a unified diff in a **separate vsplit pane** — a single read-only buffer with interleaved red/green lines. This issue tracks adding a **true in-place inline overlay**: the diff shown as virtual text layered onto the file in the **same window**, the way users typically picture a "VS Code-style inline" diff (cf. mini.diff, gitsigns preview, sidekick NES, Cursor inline edits).
When implemented, this mode is the one that should claim `layout = "inline"` — which is exactly why we reserved that name in #293.
## "unified" vs the proposed "inline"
- **`"unified"` (shipped):** single read-only buffer in a `rightbelow vsplit`, deleted (red/strikethrough) + added (green) lines interleaved. Compact, but it's a separate pane and not editable before accept.
- **`"inline"` (this issue):** no extra window — proposed changes rendered in place over the real file.
## Suggested approach (keeps pre-accept editability)
The unified/read-only path forfeits editing the proposal before accepting. A prototype pattern that keeps it (the codecompanion/sidekick model):
- **Proposed/added lines = real, editable buffer lines**, highlighted (e.g. `ClaudeCodeInlineDiffAdd`) with a `+` sign.
- **Deleted lines = virtual text** via `nvim_buf_set_extmark` `virt_lines` (strikethrough/red), so they're visible but not part of the editable content.
- On **accept**, the buffer content already *is* the new file (deleted lines were never real text), so edits made during review are captured. On **reject**, clear the extmarks / restore.
A pure-Lua, zero-dependency engine (not mini.diff/diffview — see #169) keeps this aligned with the project's philosophy. Bonus: a single-buffer extmark view sets no `&diff` windows, which sidesteps the `closeAllDiffTabs` foreign-diff class (#277) and the stale-diff repaint issues (#205).
## Notes
- The current `"unified"` path doesn't fire the `ClaudeCodeDiffOpened` / `ClaudeCodeDiffClosed` `User` autocmds that the native diff does — worth wiring up for any new mode.
- @wookayin expressed interest ("once we implement that") on #195 — collaboration welcome.
Refs #293, #195, #82
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Contributor guide
No contributing guide indexed for this repository
Assessment
This issue has not been assessed yet.