Better match algorithm

[YodaEmbedding]

smart-open is better at frecency than previous plugins. However (with fzy and defaults), it may underweight exact matches or lack smarter filepath matching heuristics.

---

smartopen.nvim:

╭──────────────────────── Results ────────────────────────╮
│     base.py demo/pipeline/scheduler                    │
│     smart.py demo/pipeline/scheduler                   │
│     timing.py demo/pipeline/component                  │
│     simple.py demo/pipeline/component                  │
│     parallel.py demo/pipeline/component                │
│     base.py demo/pipeline/component                    │
│     component/__init__.py demo/pipeline                │
│     client_to_server_ping_item.py demo/serialization   │
│     server_to_client_ping.proto protos                 │
│     server_to_client_ping_item.py demo/serialization   │
│     client_to_server_ping.proto protos                 │
│     ping.py demo/io/network                            │
│     pipeline/__init__.py demo                          │
│>    proto/__init__.py demo                             │
╰─────────────────────────────────────────────────────────╯
╭───────────────── Search Files By Name ──────────────────╮
│> pip                                                   0│
╰─────────────────────────────────────────────────────────╯

Seems to overweight frecency.

---

fff.nvim:

┌ Files ──────────────────────────────────────────────────┐
│   history.py demo/pipeline/scheduler                   │
│   __init__.py demo/pipeline/component                  │
│   __init__.py demo/pipeline/scheduler                  │
│   smart.py demo/pipeline/scheduler                     │
│   frame_id_map.py demo/pipeline                        │
│   defaults.py demo/pipeline/scheduler                  │
│   simple.py demo/pipeline/component                    │
│   parallel.py demo/pipeline/component                  │
│   timing.py demo/pipeline/component                    │
│   base.py demo/pipeline/scheduler                      │
│   base.py demo/pipeline/component                      │
┌─────────────────────────────────────────────────────────┐
│🪿 pip                                          175/180 •│
└─────────────────────────────────────────────────────────┘

Weighs exact matches more? (Disclaimer: minimal use, though I opened demo/proto/init.py a bunch of times to try to bias the frecency.)

---

fff.nvim uses:

• frizbee

Related:

• [Feature Request] match_algorithm add support for zf #61
• zf

zf is a fuzzy finder that excels at filtering filepaths:

• because filenames are usually unique, matches on filenames are prioritized
• when the query resembles a file path, zf uses heuristics for a more accurate match

The goal of zf is to be more accurate than other fuzzy finders when filtering filepaths, but it also functions as a general-purpose fuzzy finder.

Try zf online!

---

Config:

return {
  "danielfalk/smart-open.nvim",

  dependencies = {
    "kkharji/sqlite.lua",
    "nvim-telescope/telescope.nvim",

    -- Optional. Only required if using match_algorithm fzf.
    { "nvim-telescope/telescope-fzf-native.nvim", build = "make" },

    -- Optional. If installed, native fzy will be used when match_algorithm is fzy.
    { "nvim-telescope/telescope-fzy-native.nvim" },
  },

  lazy = true,

  config = function()
    require("telescope").load_extension("smart_open")
  end,
}

[danielfalk]

What you're seeing may not be related to weights at all. This is probably because of the "virtual file name". See Notes for details, but this works by treating the parent directory + the filename as the filename in certain cases.

So in your example, proto/__init__.py is treated as the filename. Why is this? In some projects such as python, many of the files are named __init__.py, so trying to find the right __init__.py can be challenging. So smart-open expands the concept of the filename to include the parent directory too, just for files named __init__.py (or index.js etc) so there's more uniqueness to it.

As a consequence, your pip above matches [p]roto/__[i]nit__.[p]y, and of course it matches [pip]eline/__init__.py as well.

Frecency is still treated as a weight, though how much weight it gets depends on how often you select the frecent file in practice. If you select those files often, they get more weight. If you don't, they get less weight. It's intended to be self-tuning.