Skip to content

TODO.md

Latest commit

 

History

History
63 lines (43 loc) · 5.28 KB

TODO.md

File metadata and controls

63 lines (43 loc) · 5.28 KB

TODO

Preview Streaming

  • Streaming search inside pager Large text previews should support /pattern search without loading everything. We’ll do literal-only search with SIMD fast paths (AVX2/AVX-512 on amd64, NEON on arm64) and chunked streaming fallback.
    Plan:

    • Input/UI: add / to enter search mode, show query + hit index in status bar, n/N (or arrows) to jump next/prev, Esc to exit.
    • Matcher: streaming literal search is in place; add SIMD fast paths (memmem) with chunk overlap and keep line/col offsets for hits (pending).
    • Highlighting: store hit spans per line; on render, map spans to wrapped rows and inject ANSI; binary mode skips search.
    • Navigation: maintain hit list + cursor; jumps update PreviewScrollOffset/PreviewWrapOffset to center the hit.
    • Performance/limits: cap buffered hits (e.g., 10k) and buffered lines (e.g., 20k); debounce query changes; fallback to pure Go bytes.Index when SIMD unavailable. (Optional tuning remains.)

  • UTF-16 pager streaming textPagerSource now streams UTF-16 LE/BE previews with shared decoder state, byte-offset aware line metadata, and copy-all support that loads the full file without re-decoding the head buffer.

  • Window resize on Windows pager
    Unix pagers now redraw immediately when the terminal resizes (SIGWINCH + select). Windows still requires a keypress because we never consume WINDOW_BUFFER_SIZE_EVENT. Implement a Windows-specific key reader that uses ReadConsoleInput/golang.org/x/sys/windows to listen for both KEY_EVENT and WINDOW_BUFFER_SIZE_EVENT, toggles raw mode via SetConsoleMode, and pushes synthetic resize events into the pager loop so p.updateSize()/render() fire without waiting for input.

Performance & Search

  • Token heuristics & order
    During AND queries we currently sort tokens only by length. Explore heuristics based on selectivity (rareness, occurrence counts) to run the most discriminating token first and avoid extra DP passes.

  • Lower-case name cache
    Case-insensitive filtering/search constantly calls strings.ToLower per filename/path. Cache folded names in AppState (and refresh when directories reload) to reduce CPU/allocs on large trees.

  • Result pooling
    GlobalSearchResult/FileEntry allocations still dominate AND benchmarks (~5 MB/op). Investigate pooling or avoiding os.FileInfo until a result is promoted by the collector.

Testing & QA

  • Preview pager PTY tests
    Add end-to-end tests that run PreviewPager.Run() against a pseudo-terminal created with the standard library (golang.org/x/term + os.Pipe/syscall), drive key presses (e.g., q, w, PgDn), and assert that the pager restores cursor visibility/DECAWM and renders expected headers/status for text and binary files. This would catch regressje typu “ukryty kursor po wyjściu” i dokumentować wymianę sekwencji sterujących.

  • Optional: auto-jump search anchor
    If we enable live jumping while typing, anchor to the view center on entering /, auto-center the first hit at/after that anchor as the query updates (debounced), and reset the anchor after manual jumps (n/N/Enter). Clearing (←) or Esc should restore the anchored view and clear results. Gate behind a preference (e.g., PreviewSearchAutoJump) to avoid surprising motion by default.

Binary Pager Improvements

  • Block jumps + offset status
    Add [ / ] for ±4 KB and { / } for ±64 KB jumps in PreviewPager for binaries; status shows current offset and file %. Pager-only; no preview panel change.

  • Binary search
    Remove the / blockade in binary mode (pager.go:1162-1168) after adding simple byte/ASCII search on binarySource. Support n/N, hit highlighting, respect searchMaxHits/Lines. Preview panel remains without search.

  • Adaptive hexdump width (pager)
    binaryPagerSource now emits 8/16/24 bytes per row based on pager width; offset/hex/ASCII columns adapt accordingly. Inline preview stays fixed at 16 B to remain lightweight.
    Follow-ups included: correct hit span mapping for 8/16/24, preserve byte position across resize/exit, and ensure hit-capped searches still highlight/navigate within the current viewport.

  • Dual-column hex + text
    Add a pager mode where each line shows hex plus parallel ASCII/UTF-8 decoding with non-printables masked; simple key toggle. Preview panel unchanged.

  • Export visible block
    Extend existing copyVisibleToClipboard so binary mode can copy the current view as hex+ASCII (e.g., the ~256 B on screen) and optionally save to temp/. Pager-only.

  • Lightweight format detection
    Detect signatures (PNG/ZIP/ELF/Mach-O/PDF) when loading binaries; show a terse header in preview (e.g., type + section sizes) and richer detail in pager. Minimal I/O: first few hundred bytes, no full parse.

Display & Theming

  • Terminal palette awareness / adaptive theming
    Detect terminal color capabilities and palette (truecolor vs 256, custom iTerm schemes), then adapt preview/pager styling accordingly: choose nearest palette colors when truecolor unavailable, or switch to palette-only SGR to eliminate tone drift. Expose a toggle/preference to force 24-bit or 256-color for consistent appearance across panels and pager.