# TODO

## Barbar sidebar offset + alignment still broken

**Symptoms:**
- Oil sidebar label shows literal `Files` (static), not the expected fish-shortened
  cwd (`~/P/t/neovimrc`).
- Sidebar width does not shrink to fit longest entry — stays at default ~35 cols
  even when the longest filename is much shorter.
- Bar colors (`BufferCurrent*`, `BufferOffset`) visibly mismatch after darkbox
  edits and restart.

**Context:**
- `lua/timmypidashev/plugins/oil.lua` has `fish_cwd()` and `fit_width()` helpers
  wired to `bufferline.api.set_offset(width, text)`. Expected to be called on
  `BufWinEnter` / `BufReadPost` / `TextChanged` for oil buffers, plus retry loop
  `maybe_fit()` polling for buffer populate.
- `lua/timmypidashev/plugins/barbar.lua` sets `sidebar_filetypes.oil` +
  `exclude_ft = { "oil" }`.
- Darkbox `BufferOffset` was updated to `fg = foreground, bold = true` to match
  `BufferCurrent`; separator glyph changed from `▎` to `│`.

**Hypotheses to verify:**
1. `task dev` was not run, so the copied `~/.config/nvim/lua/…` lacks the rtp
   prepend + oil changes. Check:
   `:lua =vim.api.nvim_get_runtime_file("colors/darkbox.lua", true)[1]`
   must show `~/Projects/timmypidashev/darkbox.nvim/…`.
2. Leftover pack clone of darkbox is winning over local rtp:
   `~/.local/share/nvim/site/pack/core/opt/darkbox.nvim` may need removal.
3. `fit_width` / `maybe_fit` never fire because the autocmd's buffer filter
   or pattern isn't matching oil buffers as expected.
4. `bufferline.api.set_offset` text arg is silently ignored or `api` module
   name differs across barbar versions.

**Next steps:**
- Instrument with `vim.notify` inside `maybe_fit`, `fit_width`, and `set_offset`
  to confirm they run and with what values.
- Try `require("barbar").api` vs `require("bufferline.api")` — the module path
  has changed between barbar releases.
- Consider dropping barbar's offset mechanism and rendering the label inline at
  the top of the oil buffer using extmarks instead.

---

## Opening markdown files has ~2s blank / freeze

**Symptoms:**
- Opening `README.md` (and likely other markdown files) shows a blank buffer for
  ~2 seconds before content/syntax renders.

**Likely causes (to investigate in order):**
1. **marksman** LSP spawning and blocking initial render. Check
   `:LspLog` timestamps on attach.
2. **markdownlint-cli2** lint run on `BufReadPost` via `nvim-lint` —
   `lua/timmypidashev/plugins/lint.lua` triggers `try_lint` on `BufReadPost`.
   This can shell out synchronously if misconfigured.
3. **Treesitter** `markdown` + `markdown_inline` parsing both on the same buffer
   — our `FileType` autocmd starts treesitter; main branch install is async,
   but re-parse on large markdown can be slow first time.
4. **conform.nvim** format_on_save preload of `prettierd` binary (less likely
   on read, but worth eliminating).
5. **blink.cmp** Lua fuzzy impl cold-startup (we disabled rust fallback).

**Diagnostic steps:**
- `:lua vim.lsp.set_log_level("debug")` then `:LspLog` after opening the file.
- Compare times: `nvim --clean README.md` vs current config.
- Disable plugins one-by-one (`marksman`, `nvim-lint`, treesitter) to isolate.
- `:Profile` (via `:h profile`) around `:e README.md`.

**Likely fix candidates:**
- Switch `nvim-lint` trigger from `BufReadPost` to `BufWritePost` only.
- Detach marksman if no `.marksman.toml` / `.git` root — or remove it from the
  enabled LSP list entirely if unused.
- Debounce treesitter highlight start for markdown.