ocr-tech-k2 / tests /test_language_registry.py
prosperious's picture
Add 195 files (runs/hf-recognizer-mixed-v2-full/finalized/package/artifacts/model/model.safetensors ...)
1ef1de7 verified
Raw History Blame Contribute Delete
6.84 kB
from __future__ import annotations
import json
import struct
from copy import deepcopy
from pathlib import Path
from ocrtech.cli import main
from ocrtech.language_registry import audit_font_renderability, audit_language_registry
ROOT = Path(__file__).resolve().parents[1]
REGISTRY = ROOT / "languages" / "nepal-v1.json"
def test_nepal_language_registry_passes_core_gates(tmp_path: Path) -> None:
audit = audit_language_registry(REGISTRY, tmp_path / "audit")
assert audit.passed
assert audit.language_count >= 11
assert audit.counted_nepal_language_count >= 10
assert audit.first_target_ids == ["limbu"]
assert audit.primary_scripts["limbu"] == ["Limbu / Sirijonga"]
assert "english" in audit.primary_scripts
assert not audit.errors
assert (tmp_path / "audit" / "language-registry-audit.json").exists()
assert (tmp_path / "audit" / "language-registry-audit.md").exists()
def test_language_registry_rejects_non_limbu_first(tmp_path: Path) -> None:
payload = json.loads(REGISTRY.read_text(encoding="utf-8"))
payload["languages"] = [payload["languages"][1], payload["languages"][0], *payload["languages"][2:]]
registry = tmp_path / "bad-order.json"
registry.write_text(json.dumps(payload, ensure_ascii=False), encoding="utf-8")
audit = audit_language_registry(registry, tmp_path / "audit")
assert not audit.passed
assert "first language entry must have id 'limbu'" in audit.errors
def test_language_registry_rejects_too_few_counted_nepal_languages(tmp_path: Path) -> None:
payload = json.loads(REGISTRY.read_text(encoding="utf-8"))
reduced = deepcopy(payload)
counted = 0
for language in reduced["languages"]:
if language["counts_toward_nepal_language_target"]:
counted += 1
if counted > 3:
language["counts_toward_nepal_language_target"] = False
registry = tmp_path / "too-few.json"
registry.write_text(json.dumps(reduced, ensure_ascii=False), encoding="utf-8")
audit = audit_language_registry(registry, tmp_path / "audit", min_nepal_languages=10)
assert not audit.passed
assert any(error.startswith("registry counts 3 Nepal languages") for error in audit.errors)
def test_language_registry_rejects_english_counting_toward_target(tmp_path: Path) -> None:
payload = json.loads(REGISTRY.read_text(encoding="utf-8"))
for language in payload["languages"]:
if language["id"] == "english":
language["counts_toward_nepal_language_target"] = True
registry = tmp_path / "english-counted.json"
registry.write_text(json.dumps(payload, ensure_ascii=False), encoding="utf-8")
audit = audit_language_registry(registry, tmp_path / "audit")
assert not audit.passed
assert "english must not count toward the Nepal-language target" in audit.errors
def test_cli_audit_language_registry_writes_summary(tmp_path: Path, capsys) -> None:
exit_code = main(["audit-language-registry", str(REGISTRY), "--out", str(tmp_path / "audit")])
captured = capsys.readouterr()
payload = json.loads((tmp_path / "audit" / "language-registry-audit.json").read_text(encoding="utf-8"))
assert exit_code == 0
assert "passed=True" in captured.out
assert "first_targets=limbu" in captured.out
assert payload["counted_nepal_language_count"] >= 10
def test_font_renderability_audit_scores_cmap_coverage(tmp_path: Path) -> None:
font = tmp_path / "single-a.ttf"
_write_minimal_cmap_font(font, 0x0041)
registry = tmp_path / "registry.json"
registry.write_text(
json.dumps(
{
"languages": [
{
"id": "fixture",
"scripts": [
{
"script": "Latin",
"primary": True,
"unicode_ranges": ["U+0041"],
}
],
"font_candidates": [
{"name": "Fixture", "path": str(font), "status": "pending_render_audit"}
],
}
]
}
),
encoding="utf-8",
)
audit = audit_font_renderability(registry, tmp_path / "audit")
assert audit.passed
assert audit.script_count == 1
assert audit.candidates[0].coverage_ratio == 1.0
assert (tmp_path / "audit" / "font-renderability-audit.json").exists()
assert (tmp_path / "audit" / "font-renderability-audit.md").exists()
def test_cli_font_renderability_audit_fails_missing_coverage(tmp_path: Path, capsys) -> None:
font = tmp_path / "single-a.ttf"
_write_minimal_cmap_font(font, 0x0041)
registry = tmp_path / "registry.json"
registry.write_text(
json.dumps(
{
"languages": [
{
"id": "fixture",
"scripts": [
{
"script": "Latin",
"primary": True,
"unicode_ranges": ["U+0042"],
}
],
"font_candidates": [
{"name": "Fixture", "path": str(font), "status": "pending_render_audit"}
],
}
]
}
),
encoding="utf-8",
)
exit_code = main(["audit-font-renderability", str(registry), "--out", str(tmp_path / "audit")])
captured = capsys.readouterr()
payload = json.loads((tmp_path / "audit" / "font-renderability-audit.json").read_text(encoding="utf-8"))
assert exit_code == 1
assert "passed=False" in captured.out
assert payload["errors"] == ["fixture:Latin: no font candidate covers at least 95.00% of required codepoints"]
assert payload["candidates"][0]["missing_sample"] == ["U+0042"]
def _write_minimal_cmap_font(path: Path, codepoint: int) -> None:
seg_count = 2
end_codes = [codepoint, 0xFFFF]
start_codes = [codepoint, 0xFFFF]
id_deltas = [1, 1]
id_range_offsets = [0, 0]
subtable = b"".join(
[
struct.pack(">HHHHHHH", 4, 32, 0, seg_count * 2, 0, 0, 0),
struct.pack(">HH", *end_codes),
struct.pack(">H", 0),
struct.pack(">HH", *start_codes),
struct.pack(">hh", *id_deltas),
struct.pack(">HH", *id_range_offsets),
]
)
cmap = struct.pack(">HHHHL", 0, 1, 3, 1, 12) + subtable
offset_table = struct.pack(">LHHHH", 0x00010000, 1, 0, 0, 0)
table_record = b"cmap" + struct.pack(">LLL", 0, len(offset_table) + 16, len(cmap))
path.write_bytes(offset_table + table_record + cmap)