Orchestrator/bahn/teamlandkarte-mcp/tests/test_reference_partner_name_pbt.py

# Feature: llm-fulltext-matching, Property 2b:
# Referenzen mit leerem Partner-Name behalten projects, ohne Partner-Token
"""Property-based tests for reference rendering with empty partner names.

These tests cover the deterministic serializer
``serialize_capacity_profile`` defined in
``teamlandkarte_mcp.matching.profiles``. They focus specifically on the
``Referenzen:`` block and verify three invariants for any list of
:class:`CapacityReferenceEntry` values that mixes empty and non-empty
``partner_name`` values:

(a) The number of rendered reference lines equals the number of input
    entries with non-whitespace ``projects``.
(b) Each line that corresponds to an entry with empty ``partner_name``
    starts with ``Projekte:`` and does NOT contain a ``Partner:`` token.
(c) Each line that corresponds to an entry with non-empty
    ``partner_name`` contains both ``Partner: <name>`` and
    ``Projekte: <projects>``.

Because the serializer classifies entries via ``if entry.partner_name:``
(truthiness), this test mirrors that exact logic and uses
``bool(entry.partner_name)`` to decide whether the ``Partner:`` token is
expected.

The strategies deliberately exclude line-break-like characters
(everything ``str.splitlines`` treats as a line boundary) so the
line-based extraction of the rendered ``Referenzen:`` block remains
unambiguous. The property under test is the *content* of the rendered
references, not how the serializer would behave on text containing
literal newlines.

**Validates: Requirements 2.8, 3.4, 3.6**
"""

from __future__ import annotations

from hypothesis import given, settings
from hypothesis import strategies as st

from teamlandkarte_mcp.matching.profiles import (
    CapacityProfile,
    CapacityReferenceEntry,
    serialize_capacity_profile,
)


# ---------------------------------------------------------------------------
# Strategies
# ---------------------------------------------------------------------------

# Alphabet for partner_name/projects: any unicode codepoint that
# ``str.splitlines`` does NOT treat as a line boundary, so the rendered
# output remains a stable, single-line-per-entry block. We exclude:
#
# * ``Cs`` (surrogates) — invalid in well-formed strings.
# * ``Cc`` (control characters) — includes ``\n``, ``\r``, ``\v``,
#   ``\f``, ``\x1c``-``\x1f``, ``\x85`` etc.
# * ``Zl`` (line separator, ``\u2028``).
# * ``Zp`` (paragraph separator, ``\u2029``).
_safe_chars = st.characters(
    blacklist_categories=("Cs", "Cc", "Zl", "Zp"),
)

# ``partner_name``: either empty (NULL ``partner_id`` or join mismatch,
# see requirement 2.8) or a short non-empty string. The serializer uses
# truthy-checking, so any non-empty string (including whitespace-only)
# counts as "has partner" — we keep this strategy aligned with that
# logic and classify entries by ``bool(entry.partner_name)`` in the
# assertions below.
_partner_name = st.one_of(
    st.just(""),
    st.text(alphabet=_safe_chars, min_size=1, max_size=20),
)

# ``projects``: non-whitespace text so the serializer never drops the
# entry defensively. This lets us assert the stronger invariant
# "input entry count == rendered line count".
_projects = st.text(
    alphabet=_safe_chars, min_size=1, max_size=30
).filter(lambda s: s.strip() != "")

_reference = st.builds(
    CapacityReferenceEntry,
    partner_name=_partner_name,
    projects=_projects,
)

_reference_list = st.lists(_reference, min_size=1, max_size=8)


# ---------------------------------------------------------------------------
# Helpers
# ---------------------------------------------------------------------------


def _extract_reference_lines(serialized: str) -> list[str]:
    """Extract the rendered reference lines from a serialized profile.

    The serializer renders the ``Referenzen:`` block as a heading line
    followed by either a bullet list (one ``- <line>`` per entry) or
    `` (keine)`` on the same line as the heading. The next field
    heading is always ``Zertifikate:`` at the start of its own line,
    which we use as the terminator. The leading ``- `` prefix is
    stripped from each line so callers can assert directly on the
    rendered content.
    """
    lines: list[str] = []
    in_references = False
    for raw_line in serialized.splitlines():
        if raw_line.startswith("Zertifikate:"):
            break
        if raw_line.startswith("Referenzen:"):
            in_references = True
            continue
        if in_references and raw_line.startswith("- "):
            lines.append(raw_line[2:])
    return lines


# ---------------------------------------------------------------------------
# Tests
# ---------------------------------------------------------------------------


@settings(max_examples=100, deadline=None)
@given(refs=_reference_list)
def test_references_with_empty_partner_name_keep_projects_without_partner_token(
    refs: list[CapacityReferenceEntry],
) -> None:
    """Property 2b: empty ``partner_name`` keeps ``projects`` without token.

    For any list of :class:`CapacityReferenceEntry` values with mixed
    ``partner_name``:

    (a) The number of rendered reference lines equals the number of
        input entries (the strategy guarantees non-whitespace
        ``projects``, so no entry is dropped).
    (b) Every line for an entry with empty ``partner_name`` starts with
        ``Projekte:`` and contains no ``Partner:`` token.
    (c) Every line for an entry with non-empty ``partner_name`` contains
        both ``Partner: <name>`` and ``Projekte: <projects>``.
    """
    profile = CapacityProfile(
        id="x",
        owner_name="o",
        role_name="r",
        competences=[],
        description="",
        references=list(refs),
        certificates=[],
    )

    serialized = serialize_capacity_profile(profile)
    lines = _extract_reference_lines(serialized)

    # (a) One rendered line per input entry, in input order.
    assert len(lines) == len(refs), (
        f"Expected {len(refs)} reference lines, got {len(lines)}: "
        f"{lines!r}"
    )

    # (b) and (c): per-entry assertions, mirroring the serializer's
    # ``if entry.partner_name:`` truthiness check.
    for entry, line in zip(refs, lines):
        projects_clean = entry.projects.strip()
        if entry.partner_name:
            # (c) Non-empty partner name → both tokens present.
            assert f"Partner: {entry.partner_name}" in line, (
                f"Expected 'Partner: {entry.partner_name}' in line: "
                f"{line!r}"
            )
            assert f"Projekte: {projects_clean}" in line, (
                f"Expected 'Projekte: {projects_clean}' in line: "
                f"{line!r}"
            )
        else:
            # (b) Empty partner name → starts with ``Projekte:`` and no
            # ``Partner:`` token at all.
            assert line.startswith("Projekte:"), (
                f"Expected line to start with 'Projekte:', got: {line!r}"
            )
            assert "Partner:" not in line, (
                f"Did not expect 'Partner:' token in line: {line!r}"
            )
            assert line == f"Projekte: {projects_clean}", (
                f"Expected exact 'Projekte: {projects_clean}', got: "
                f"{line!r}"
            )