Skip to content

Codebase intelligence for AI and humans: code health scores, auto-generated docs, git analytics, dead code detection, and architectural decisions via MCP.

3.8 KB
Raw Blame
"""The repository overview's prose budget, and the report row that checks it.

The overview is the page every reader meets first, and it had grown to roughly
a thousand words of prose saying what four hundred said.  The budget is asked
for in the prompt; this is the check that says whether the run honoured it.
"""

from __future__ import annotations

from repowise.core.generation.models import GeneratedPage
from repowise.core.generation.prose import prose_word_count
from repowise.core.generation.report import (
    ORIENTATION_PROSE_WORD_BUDGET,
    GenerationReport,
)

BUDGET = ORIENTATION_PROSE_WORD_BUDGET


def _page(content: str, page_type: str = "repo_overview") -> GeneratedPage:
    return GeneratedPage(
        page_id=f"{page_type}:.",
        page_type=page_type,
        title="Repository Overview",
        content=content,
        source_hash="",
        model_name="mock",
        provider_name="mock",
        input_tokens=0,
        output_tokens=0,
        cached_tokens=0,
        generation_level=6,
        target_path="",
        created_at="2026-07-29T00:00:00Z",
        updated_at="2026-07-29T00:00:00Z",
    )


def _overview_of(word_count: int) -> GeneratedPage:
    body = " ".join(f"word{index}" for index in range(word_count - 1))
    return _page(f"# Overview\n\n{body}\n")


# ---------------------------------------------------------------------------
# Counting
# ---------------------------------------------------------------------------


def test_prose_word_count_ignores_code_blocks_and_tables() -> None:
    content = """# Overview

The loader reads three files.

```python
these words are quoted source and are not prose the reader wades through
```

| Field | Type |
| --- | --- |
| path | string |
"""
    # "Overview" plus "The loader reads three files."
    assert prose_word_count(content) == 6


def test_prose_word_count_ignores_bare_punctuation() -> None:
    assert prose_word_count("- one — two\n\n* three\n") == 3


# ---------------------------------------------------------------------------
# The budget
# ---------------------------------------------------------------------------


def test_overview_at_the_cap_is_within_budget() -> None:
    report = GenerationReport.from_pages([_overview_of(BUDGET)])

    assert report.overview_prose_words == BUDGET
    assert not report.overview_over_budget
    assert report.overview_length_summary() == f"{BUDGET} / {BUDGET} words"


def test_overview_just_under_the_cap_is_within_budget() -> None:
    report = GenerationReport.from_pages([_overview_of(BUDGET - 1)])

    assert report.overview_prose_words == BUDGET - 1
    assert not report.overview_over_budget


def test_overview_just_over_the_cap_is_flagged() -> None:
    report = GenerationReport.from_pages([_overview_of(BUDGET + 1)])

    assert report.overview_prose_words == BUDGET + 1
    assert report.overview_over_budget
    assert "over budget" in report.overview_length_summary()


def test_a_run_without_an_overview_says_so_rather_than_reading_as_a_pass() -> None:
    report = GenerationReport.from_pages([_page("# Layer\n\nwords.", page_type="layer_page")])

    assert report.overview_prose_words is None
    assert not report.overview_over_budget
    assert report.overview_length_summary() == "no overview in this run"


# ---------------------------------------------------------------------------
# Prompt and check must agree
# ---------------------------------------------------------------------------


def test_the_prompt_states_the_same_cap_the_report_checks() -> None:
    """A budget the prompt never mentions would only ever be reported, never met."""
    from pathlib import Path

    template = (
        Path(__file__).parents[3]
        / "packages/core/src/repowise/core/generation/templates/repo_overview.j2"
    )
    assert str(ORIENTATION_PROSE_WORD_BUDGET) in template.read_text()