fastcore.tools.view() generates OpenAI-incompatible tool schema for view_range

[bobrowadam]

Description

fastcore.tools.view() currently declares:

view_range: tuple[int, int] = None

When this is turned into a tool schema via toolslm.funccall.get_schema(), the generated JSON Schema for view_range uses prefixItems without items.

That works for some consumers, but OpenAI rejects it with:

Invalid schema for function 'view': In context=('properties', 'view_range'), array schema missing items.

This breaks downstream tools that expose fastcore.tools.view directly as an OpenAI function/tool, including shell-sage when configured to use OpenAI models.

Reproduction

Environment:

• macOS
• shell-sage 1.0.6
• fastcore 1.12.33
• also reproduced by inspecting latest fastcore 1.12.34 source on PyPI

Minimal repro:

from toolslm.funccall import get_schema
from fastcore.tools import view
import json

print(json.dumps(get_schema(view, pname="parameters"), indent=2))

Generated schema includes:

"view_range": {
  "type": "array",
  "default": null,
  "prefixItems": [
    {"type": "integer"},
    {"type": "integer"}
  ]
}

OpenAI then rejects the tool schema with:

Invalid schema for function 'view': In context=('properties', 'view_range'), array schema missing items.

Expected

The generated schema should be accepted by OpenAI function/tool calling.

Actual

The generated schema is rejected by OpenAI because the array schema has prefixItems but no items.

Local workaround

Changing the signature to:

view_range: list[int] = None

produces:

"view_range": {
  "type": "array",
  "items": {"type": "integer"},
  "default": null
}

and fixes the issue for OpenAI-backed shell-sage.

Suggested fix

Either:

1. change view_range in fastcore.tools.view() from tuple[int, int] to list[int], or
2. update the schema generator so tuple schemas emitted for OpenAI-compatible tool calling also include a valid items field.

I’m happy to test a fix against shell-sage + OpenAI.