Medios-Macina/docs/plugin_authoring.md

# Plugin authoring: ResultTable and plugin adapters

This short guide explains how to write plugins that integrate with the strict
ResultTable API: adapters yield `ResultModel` instances, and plugins register
via `SYS.result_table_adapters.register_plugin` with columns and a
`selection_fn`.

The public terminology is plugin-first, even though some internal classes and
metadata fields still use `Provider` naming.

---

## Quick summary
- Plugins register a plugin adapter, a `columns` definition, and a `selection_fn`.
- `selection_fn` returns CLI args for a selected row.
- For HTML table or list scraping, prefer `TablePluginMixin` from `SYS.plugin_helpers`.

## Runtime dependency policy
- Treat required runtime dependencies such as Playwright as mandatory: import them unconditionally and let missing dependencies fail fast.
- Use guarded imports only for truly optional dependencies such as `pandas`.
- Keep plugin code minimal and explicit: fail early and document required runtime dependencies in README and installation notes.

---

## Minimal plugin template

```py
# plugins/my_plugin.py
from typing import Any, Dict, Iterable, List

from SYS.result_table_api import ResultModel, ColumnSpec, title_column
from SYS.result_table_adapters import register_plugin

SAMPLE_ITEMS = [
    {
        "name": "Example File.pdf",
        "path": "https://example.com/x.pdf",
        "ext": "pdf",
        "size": 1024,
        "source": "myplugin",
    },
]


def adapter(items: Iterable[Dict[str, Any]]) -> Iterable[ResultModel]:
    for it in items:
        title = it.get("name") or it.get("title") or str(it.get("path") or "")
        yield ResultModel(
            title=str(title),
            path=str(it.get("path")) if it.get("path") else None,
            ext=str(it.get("ext")) if it.get("ext") else None,
            size_bytes=int(it.get("size")) if it.get("size") is not None else None,
            metadata=dict(it),
            source=str(it.get("source")) if it.get("source") else "myplugin",
        )


def columns_factory(rows: List[ResultModel]) -> List[ColumnSpec]:
    cols = [title_column()]
    if any((row.metadata or {}).get("size") for row in rows):
        cols.append(ColumnSpec("size", "Size", lambda row: row.size_bytes or ""))
    return cols


def selection_fn(row: ResultModel) -> List[str]:
    if row.path:
        return ["-path", row.path]
    return ["-title", row.title or ""]


register_plugin("myplugin", adapter, columns=columns_factory, selection_fn=selection_fn)
```

---

## Table scraping with `TablePluginMixin`

If your plugin scrapes HTML tables or list-like results, use `TablePluginMixin`:

```py
from PluginCore.base import Provider
from SYS.plugin_helpers import TablePluginMixin


class MyTablePlugin(TablePluginMixin, Provider):
    URL = ("https://example.org/search",)

    def validate(self) -> bool:
        return True

    def search(self, query: str, limit: int = 50, **kwargs):
        url = f"{self.URL[0]}?q={quote_plus(query)}"
        return self.search_table_from_url(url, limit=limit)
```

`TablePluginMixin.search_table_from_url` returns
`PluginCore.base.SearchResult` entries. If you want to integrate the plugin
with the strict `ResultTable` registry, add a small adapter that converts
`SearchResult` to `ResultModel` and register it using `register_plugin`.

---

## Columns and selection
- `columns` may be a static `List[ColumnSpec]` or a factory that inspects sample rows.
- `selection_fn` must accept a `ResultModel` and return a `List[str]` representing CLI args.
- For downloadable file rows, prefer explicit URL args such as `['-url', row.path]` so downstream downloaders interpret the row unambiguously.
- Ensure `ResultModel.source` is set directly or falls back to the registered plugin name during serialization.

---

## Optional pandas support
`SYS.html_table.extract_records` prefers a pure-lxml path but can fall back to
`pandas.read_html` when pandas is installed and the helper detects it works for
the input table. This is optional. Document whether your plugin requires
`pandas` and emit a clear error or log message when it is missing.

---

## Testing and examples
- Write `tests/test_plugin_<name>.py` or follow the repo's older naming conventions when extending existing tests.
- Verify `plugin.build_table(...)` produces a `ResultTable` with rows and columns.
- Verify `serialize_rows()` yields `_selection_args`, `_selection_action` when applicable, and `source`.
- When you need an exact CLI stage sequence, call `table.set_row_selection_action(row_index, tokens)` so replay uses the row action verbatim.
- For table-oriented plugins, test `search_table_from_url` with a local HTML fixture or a mocked `HTTPClient`.

Example test skeleton:

```py
from SYS.result_table_adapters import get_plugin
from plugins import example_plugin


def test_example_plugin_registration():
    plugin = get_plugin("example")
    rows = list(plugin.adapter(example_plugin.SAMPLE_ITEMS))
    assert rows and rows[0].title
    cols = plugin.get_columns(rows)
    assert any(col.name == "title" for col in cols)
    table = plugin.build_table(example_plugin.SAMPLE_ITEMS)
    assert table.provider == "example" and table.rows
```

---

## References and examples
- Read `plugins/example_plugin.py` for a compact example of a strict adapter and dynamic columns.
- Read `plugins/vimm/__init__.py` for a table-oriented plugin that uses `TablePluginMixin` and converts `SearchResult` to `ResultModel` for registration.
- See `docs/plugin_guide.md` for a broader plugin development checklist.