Files
github-spec-kit/tests/conftest.py
Copilot f0998348be feat: Config-driven opt-in authentication registry with multi-platform support (#2393)
* Initial plan

* feat: add authentication provider registry (GitHub + Azure DevOps)

Agent-Logs-Url: https://github.com/github/spec-kit/sessions/da7ecfd0-e1c9-48dc-b692-27be0879e976

Co-authored-by: mnriem <15701806+mnriem@users.noreply.github.com>

* feat: add try-each-provider HTTP helper and wire all catalog fetches through auth registry

- Add authentication/http.py with open_url() that tries each configured
  provider in registry order, falling through on 401/403 to the next,
  and finally to unauthenticated
- Add build_request() for one-shot request construction
- Add configured_providers() to registry __init__
- Remove api_base_url() from AuthProvider ABC (unused)
- Remove hosts attribute from providers (no host matching)
- Replace _github_http.py usage in ExtensionCatalog and PresetCatalog
- Wire IntegrationCatalog and WorkflowCatalog through open_url (were unauthenticated)
- Wire _fetch_latest_release_tag() through open_url
- Wire all inline --from-url downloads through open_url
- Fix unused stub variable flagged by code-quality bot
- 49 auth tests (positive + negative), 1805 total tests passing

* fix: address review — fix stale docstrings, restore Accept header, add extra_headers to open_url

- Fix _open_url() docstrings in extensions.py and presets.py that
  incorrectly claimed redirect stripping behavior
- Add extra_headers parameter to open_url() so callers can pass
  additional headers (e.g. Accept) that persist across retries
- Restore Accept: application/vnd.github+json header in
  _fetch_latest_release_tag() via extra_headers

* feat: config-driven opt-in auth via ~/.specify/auth.json

Security-first redesign: no credentials are sent unless the user
explicitly creates ~/.specify/auth.json mapping hosts to providers.

- Add authentication/config.py: loads and validates auth.json with
  host-to-provider mappings, supports token/token_env/azure-ad/azure-cli
- Refactor AuthProvider ABC: auth_headers(token, scheme) + resolve_token(entry)
- Refactor GitHubAuth: bearer scheme only, token from config entry
- Refactor AzureDevOpsAuth: 4 schemes (basic-pat, bearer, azure-cli, azure-ad)
  with dynamic token acquisition for azure-cli and azure-ad
- Rewrite authentication/http.py: host matching, redirect stripping,
  provider fallthrough on 401/403, unauthenticated fallback
- Add docs/reference/authentication.md with full reference and template
- 1823 tests passing (67 auth-specific)

* fix: address review — unused imports, host normalization, provider+scheme validation, security hardening

- Remove unused imports (os, field, Any) in config.py
- Normalize hosts during load (strip + lowercase)
- Validate token/token_env are non-empty strings during load
- Validate provider+scheme compatibility during load
- Fix extra_headers order: auth headers applied last, cannot be overridden
- Remove unused 'tried' variable in http.py
- Warn (once) on malformed auth.json instead of silent fallback
- URL-encode OAuth2 client credentials body in azure_devops.py
- Update 403 message to mention auth.json configuration
- Fix registry leak in test_register_duplicate (try/finally)
- Fix import style consistency in test_authentication.py
- Add azure-cli and azure-ad token acquisition tests (mock subprocess/urlopen)
- Add autouse fixture to isolate upgrade tests from real auth.json
- 1829 tests passing

* fix: reject unknown providers, validate azure-ad fields, strip Authorization from extra_headers

- Reject unknown provider keys during auth.json load with clear error message
- Validate azure-ad tenant_id/client_id/client_secret_env as non-empty strings
- Strip Authorization from extra_headers in both build_request and open_url
  to prevent accidental or intentional bypass of provider-configured auth
- Add tests for unknown provider and incompatible scheme validation
- 1831 tests passing

* fix: extract shared auth test helpers, global config isolation, align docstring

- Move _inject_github_config / make_github_auth_entry to tests/auth_helpers.py
  to eliminate duplication across test_extensions, test_presets, test_upgrade
- Move auth config isolation fixture to global conftest.py (autouse) so ALL
  tests are isolated from ~/.specify/auth.json, not just test_upgrade
- Align load_auth_config docstring with actual behavior: ValueError may be
  caught by higher-level HTTP helpers that warn and continue unauthenticated
- 1831 tests passing

* fix: preserve auth header across multi-hop redirect chains

- Read Authorization from both headers and unredirected_hdrs in
  _StripAuthOnRedirect to survive multi-hop chains within allowed hosts
- Add test_multi_hop_redirect_within_hosts_preserves_auth
- 1832 tests passing

* fix: use resolved config path in warning/error messages and patch build_opener in no-network test

Agent-Logs-Url: https://github.com/github/spec-kit/sessions/86df9557-54f1-4fe4-a25f-9501cb2356cf

Co-authored-by: mnriem <15701806+mnriem@users.noreply.github.com>

* fix: assert full resolved config path in rate-limit output test

Agent-Logs-Url: https://github.com/github/spec-kit/sessions/86df9557-54f1-4fe4-a25f-9501cb2356cf

Co-authored-by: mnriem <15701806+mnriem@users.noreply.github.com>

* fix: close HTTPError on 401/403, remove _VALID_AUTH_SCHEMES, catch TimeoutExpired, skip POSIX test on Windows, remove unused import

Agent-Logs-Url: https://github.com/github/spec-kit/sessions/a1e29737-dd6e-4287-96c1-509e0c96fb21

Co-authored-by: mnriem <15701806+mnriem@users.noreply.github.com>

* fix: use stable ~/.specify/auth.json in rate-limit message, skip POSIX permission check on Windows

Agent-Logs-Url: https://github.com/github/spec-kit/sessions/4636bcdb-87ae-45d6-9545-a40e4effd617

Co-authored-by: mnriem <15701806+mnriem@users.noreply.github.com>

* fix: validate host patterns, cache auth config per-process

Agent-Logs-Url: https://github.com/github/spec-kit/sessions/889b58a7-7f8c-47e2-8056-931ebcc671cc

Co-authored-by: mnriem <15701806+mnriem@users.noreply.github.com>

* fix: clarify _is_valid_host_pattern docstring, clean up test sentinel type

Agent-Logs-Url: https://github.com/github/spec-kit/sessions/889b58a7-7f8c-47e2-8056-931ebcc671cc

Co-authored-by: mnriem <15701806+mnriem@users.noreply.github.com>

* fix: improve _is_valid_host_pattern docstring and test observability

Agent-Logs-Url: https://github.com/github/spec-kit/sessions/889b58a7-7f8c-47e2-8056-931ebcc671cc

Co-authored-by: mnriem <15701806+mnriem@users.noreply.github.com>

* Potential fix for pull request finding

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

* Potential fix for pull request finding

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

---------

Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com>
Co-authored-by: mnriem <15701806+mnriem@users.noreply.github.com>
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
2026-05-07 12:51:20 -05:00

84 lines
2.9 KiB
Python

"""Shared test helpers for the Spec Kit test suite."""
import os
import re
import shutil
import subprocess
import sys
import pytest
_ANSI_ESCAPE_RE = re.compile(r"\x1b\[[0-?]*[ -/]*[@-~]")
def _has_working_bash() -> bool:
"""Check whether a functional native bash is available.
On Windows, ``subprocess.run(["bash", ...])`` uses CreateProcess,
which searches System32 *before* PATH — so it may find the WSL
launcher even when Git-for-Windows bash appears first in PATH via
``shutil.which``. We therefore probe with bare ``"bash"`` (the
same way test helpers invoke it) to get an accurate result.
On Windows, only Git-for-Windows bash (MSYS2/MINGW) is accepted.
The WSL launcher is rejected because it runs in a separate Linux
filesystem and cannot handle native Windows paths used by the
test fixtures.
Set SPECKIT_TEST_BASH=1 to force-enable bash tests regardless.
"""
if os.environ.get("SPECKIT_TEST_BASH") == "1":
return True
if shutil.which("bash") is None:
return False
# Probe with bare "bash" — same as the test helpers — so that
# Windows CreateProcess resolution order is respected.
try:
r = subprocess.run(
["bash", "-c", "echo ok"],
capture_output=True, text=True, timeout=5,
)
if r.returncode != 0 or "ok" not in r.stdout:
return False
except (OSError, subprocess.TimeoutExpired):
return False
# On Windows, verify we have MSYS/MINGW bash (Git for Windows),
# not the WSL launcher which can't handle native paths.
if sys.platform == "win32":
try:
u = subprocess.run(
["bash", "-c", "uname -s"],
capture_output=True, text=True, timeout=5,
)
kernel = u.stdout.strip().upper()
if not any(k in kernel for k in ("MSYS", "MINGW", "CYGWIN")):
return False
except (OSError, subprocess.TimeoutExpired):
return False
return True
requires_bash = pytest.mark.skipif(
not _has_working_bash(), reason="working bash not available"
)
def strip_ansi(text: str) -> str:
"""Remove ANSI escape codes from Rich-formatted CLI output."""
return _ANSI_ESCAPE_RE.sub("", text)
# ---------------------------------------------------------------------------
# Auth config isolation — prevents tests from reading ~/.specify/auth.json
# ---------------------------------------------------------------------------
@pytest.fixture(autouse=True)
def _isolate_auth_config(monkeypatch):
"""Ensure no test reads the real ~/.specify/auth.json."""
from specify_cli.authentication import http as _auth_http
monkeypatch.setattr(_auth_http, "_config_override", [])
# Also clear the per-process cache so tests that unset _config_override
# won't see a previously cached real-file result.
monkeypatch.setattr(_auth_http, "_config_cache", None)