Files

125 lines
5.0 KiB
Python

#!/usr/bin/env python3
import os
import sys
from pathlib import Path
# Endpoints returning binary (blob) instead of JSON. These cannot use handleApiResponse
# or parseApiJsonResponse. Add path segment or exact path that matches api.fetchApi URL.
# - "download": matches /meld/api/download/zip, /meld/api/download/raw (ZIP/image binary)
BLOB_ENDPOINTS = ("download",)
def get_files_to_check() -> list[str]:
"""Search for .ts and .tsx files using api.fetchApi or .json()"""
base = Path("ui/src")
files = []
for ext in ("*.ts", "*.tsx"):
for p in base.rglob(ext):
if p.name.endswith(".d.ts"):
continue
# Skip test files and mocks
posix = str(p).replace("\\", "/")
if (
"/test/" in posix
or "/__mocks__/" in posix
or p.name.endswith(".test.ts")
or p.name.endswith(".test.tsx")
):
continue
text = p.read_text(encoding="utf-8", errors="replace")
if "api.fetchApi" in text or (".json()" in text and "api" in text):
files.append(posix)
return sorted(set(files))
def check_frontend_api_usage() -> int:
files = get_files_to_check()
errors = []
for file_path in files:
if not os.path.isfile(file_path):
continue
with open(file_path, encoding="utf-8") as f:
lines = f.read().splitlines()
for i, line in enumerate(lines):
# Inline ignore
if "frontend-api-check-ignore" in line:
continue
# Exclude handleResponse/parseJsonResponse/handleApiResponse/parseApiJsonResponse definition itself
if "export" in line:
import re
if re.search(r"\bexport\s+(const|let|var|function|default|type)\b.*\bhandleResponse\b", line):
continue
if re.search(r"\bexport\s+(const|let|var|function|default|type)\b.*\bparseJsonResponse\b", line):
continue
if re.search(r"\bexport\s+(const|let|var|function|default|type)\b.*\bhandleApiResponse\b", line):
continue
if re.search(r"\bexport\s+(const|let|var|function|default|type)\b.*\bparseApiJsonResponse\b", line):
continue
# Rule 1: Check .json() usage
if ".json()" in line:
# In ui/src/api.ts, .json() is allowed (implementation of handleResponse/parseJsonResponse)
if file_path.endswith("ui/src/api.ts"):
continue
errors.append(
f"{file_path}:{i + 1}: Forbidden direct .json() call. Use handleApiResponse(res) or parseApiJsonResponse(res) instead."
)
# Rule 2: If api.fetchApi is used, one of the response handlers should be used
if "api.fetchApi" in line:
found_handle = False
for j in range(i, min(i + 15, len(lines))):
if (
"handleResponse" in lines[j]
or "parseJsonResponse" in lines[j]
or "handleApiResponse" in lines[j]
or "parseApiJsonResponse" in lines[j]
):
found_handle = True
break
if not found_handle:
import re
# Extract the URL from api.fetchApi(URL, ...) - handle multiline
buffer = " ".join(lines[i : min(i + 5, len(lines))])
url_match = re.search(r'api\.fetchApi\(\s*["\'`]([^"\'`]+)["\'`]', buffer)
url = url_match.group(1) if url_match else None
if url:
# Check against BLOB_ENDPOINTS strictly (exact match or path segment)
if any(url == b or url.startswith(f"{b}/") or f"/{b}/" in f"/{url}/" for b in BLOB_ENDPOINTS):
continue
# Determine error message: use extracted URL if available, otherwise check line
is_meld = (url and "/meld/" in f"/{url}/") or any(
q in line for q in ['"/meld/', "'/meld/", "`/meld/"]
)
if is_meld:
errors.append(
f"{file_path}:{i + 1}: api.fetchApi call to /meld/ should use handleApiResponse(res) or parseApiJsonResponse(res)."
)
else:
errors.append(
f"{file_path}:{i + 1}: api.fetchApi call should use handleApiResponse or parseApiJsonResponse, or add to BLOB_ENDPOINTS if binary."
)
if errors:
print("Frontend API Usage Checks Failed:")
for error in errors:
print(f" {error}")
return 1
print("Frontend API Usage Checks Passed.")
return 0
if __name__ == "__main__":
sys.exit(check_frontend_api_usage())