team-tryouts/app/app.py

"""Team Tryouts Application - Flask Application Factory.

This module provides the application factory for creating and configuring
the Flask application instance with comprehensive security hardening.
"""

import os
import secrets

import markupsafe
from dotenv import load_dotenv
from flask import Flask, g, jsonify, redirect, render_template, request, url_for
from flask_cors import CORS
from sqlalchemy import text
from werkzeug.exceptions import HTTPException

from app import i18n
from app.extensions import babel, csrf, db, limiter, login_manager

load_dotenv()


def nl2br(value):
    """Convert newlines to HTML line breaks.

    Args:
        value: String value to convert.

    Returns:
        Markup: HTML-safe string with line breaks.
    """
    if value:
        # Markup('<br>').join() escapes each segment before joining.
        # Markup('<br>'.join(...)) would mark attacker-controlled text as safe.
        return markupsafe.Markup('<br>').join(str(value).splitlines())
    return ''


def normalise_database_url(url):
    """Name the PostgreSQL driver explicitly in a connection URL.

    `postgresql://…` does not mean "whichever driver is installed": it means
    psycopg2, which SQLAlchemy imports at create_engine() time. requirements
    .txt pins psycopg 3 (`psycopg[binary]`) and no psycopg2, so a clean
    install starting against the URL Render hands out — and the one this
    project's own documentation shows — raises

        ModuleNotFoundError: No module named 'psycopg2'

    before the first request. Anything with a driver already spelled out
    (`postgresql+psycopg://`, `postgresql+psycopg2://`) is left alone, so
    naming psycopg2 stays possible for an environment that has it.

    `postgres://` is the legacy alias several hosts still emit; SQLAlchemy
    dropped it in 1.4.

    Args:
        url: Value of DATABASE_URL, or None.

    Returns:
        str | None: The URL, with a driver named when it was PostgreSQL.
    """
    if not url:
        return url
    scheme, separator, rest = url.partition('://')
    if not separator or '+' in scheme:
        return url
    if scheme in ('postgres', 'postgresql'):
        return f'postgresql+psycopg://{rest}'
    return url


def build_csp(*, allow_inline_script, nonce=None):
    """Assemble the Content-Security-Policy header.

    Two mutually exclusive modes, and they really are exclusive.

    Under CSP level 3, a browser that understands nonces **ignores
    'unsafe-inline' entirely as soon as a nonce is present**. Emitting both
    would therefore not be a gentle transition: it would drop every inline
    script and every onclick attribute at once, in modern browsers only.
    The switch has to be atomic, which is why one flag drives it.

    While allow_inline_script is true no nonce is emitted at all, so adding
    nonce="{{ csp_nonce }}" to a template ahead of the switch is harmless.

    Flipping the flag requires every inline event handler to be gone first.
    A nonce cannot authorise an onclick attribute — nonces apply to script
    elements, never to handler attributes. See tests/test_csp.py, which
    tracks how many are left.

    Args:
        allow_inline_script: Keep 'unsafe-inline' in script-src.
        nonce: Per-request nonce, used only when inline script is not allowed.

    Returns:
        str: The header value.
    """
    if allow_inline_script:
        script_src = "'self' 'unsafe-inline' https://cdn.jsdelivr.net"
    else:
        script_src = f"'self' 'nonce-{nonce}' https://cdn.jsdelivr.net"

    return '; '.join(
        [
            "default-src 'self'",
            f'script-src {script_src}',
            # style-src is a separate migration: inline style="" attributes are
            # spread across the templates and are not an XSS vector on their own.
            "style-src 'self' 'unsafe-inline' https://cdnjs.cloudflare.com https://cdn.jsdelivr.net",
            "font-src 'self' https://cdnjs.cloudflare.com",
            "img-src 'self' data: https://cdn.discordapp.com",
            "connect-src 'self'",
            "frame-ancestors 'none'",
            "base-uri 'self'",
            "form-action 'self'",
        ]
    )


def create_app(config=None):
    """Create and configure the Flask application.

    Args:
        config: Optional mapping of configuration overrides, applied after the
            environment defaults and before validation. This is what makes the
            factory usable from tests: pass a throwaway database URI, a dummy
            secret, and turn off the Discord bot, without touching os.environ.

    Initializes Flask with:
        - Secret key for session security
        - Database configuration
        - CSRF protection
        - CORS with restricted origins
        - Login manager
        - Rate limiting
        - All route blueprints
        - Security headers and HTTPS redirects
        - Custom error handlers
        - Health check endpoint
        - Structured logging

    Handles database initialization and seeding with sample data if empty.

    Returns:
        Flask: Configured Flask application instance.
    """
    app = Flask(__name__)

    # Cache-busting stamps for static files, filled lazily by the url_defaults
    # hook below. Per application instance, so the test suite does not carry
    # one app's mtimes into the next.
    _static_stamps = {}

    # --- defaults from the environment ------------------------------------
    app.config['SECRET_KEY'] = os.getenv('SECRET_KEY')
    app.config['SQLALCHEMY_DATABASE_URI'] = os.getenv('DATABASE_URL')
    app.config['SQLALCHEMY_TRACK_MODIFICATIONS'] = False
    app.config['WTF_CSRF_ENABLED'] = True
    app.config['CORS_ALLOWED_ORIGINS'] = os.getenv('CORS_ALLOWED_ORIGINS', '')
    app.config['FORCE_HTTPS'] = os.getenv('FORCE_HTTPS', 'true').lower() == 'true'

    # Now false: every inline event handler has been replaced by a
    # data-action attribute dispatched from main.js, so script-src no longer
    # needs 'unsafe-inline'. Inline <script> blocks carry a per-request
    # nonce. The escape hatch remains for a deployment that hits an
    # overlooked handler — but leaving it on gives up the protection that
    # would have blocked SEC-XSS-001.
    app.config['CSP_ALLOW_INLINE_SCRIPT'] = (
        os.getenv('CSP_ALLOW_INLINE_SCRIPT', 'false').lower() == 'true'
    )

    # Internationalisation. French is the site's primary language.
    app.config['BABEL_DEFAULT_LOCALE'] = i18n.DEFAULT_LOCALE
    app.config['BABEL_TRANSLATION_DIRECTORIES'] = os.path.join(
        os.path.dirname(os.path.abspath(__file__)), 'translations'
    )

    # Side effects of create_app(), both on by default so that production and
    # development behave exactly as before. Tests turn them off.
    app.config['AUTO_CREATE_TABLES'] = os.getenv('AUTO_CREATE_TABLES', 'true').lower() == 'true'
    app.config['ENABLE_DISCORD_BOT'] = os.getenv('ENABLE_DISCORD_BOT', 'true').lower() == 'true'

    # --- caller overrides win ---------------------------------------------
    if config:
        app.config.update(config)

    # --- validation, after overrides so tests can supply their own ---------
    if not app.config['SECRET_KEY']:
        raise RuntimeError('SECRET_KEY environment variable must be set for security')
    if not app.config['SQLALCHEMY_DATABASE_URI']:
        raise RuntimeError(
            'DATABASE_URL environment variable must be set to a PostgreSQL connection string'
        )

    # After the overrides, so a caller-supplied URL is normalised too.
    app.config['SQLALCHEMY_DATABASE_URI'] = normalise_database_url(
        app.config['SQLALCHEMY_DATABASE_URI']
    )

    # File upload size limit (16 MB)
    app.config['MAX_CONTENT_LENGTH'] = 16 * 1024 * 1024

    # Secure session cookie settings
    app.config['SESSION_COOKIE_SECURE'] = (
        os.getenv('SESSION_COOKIE_SECURE', 'true').lower() == 'true'
    )
    app.config['SESSION_COOKIE_HTTPONLY'] = True
    app.config['SESSION_COOKIE_SAMESITE'] = 'Lax'
    app.config['PERMANENT_SESSION_LIFETIME'] = 3600  # 1 hour session timeout

    # Configure CORS - restrict to specific origins in production
    allowed_origins = str(app.config.get('CORS_ALLOWED_ORIGINS') or '').split(',')
    allowed_origins = [origin.strip() for origin in allowed_origins if origin.strip()]

    # CORS is only configured when origins are named explicitly.
    #
    # The previous else-branch called CORS(app, supports_credentials=True)
    # with no origins argument. flask-cors then defaults to '*' and, because
    # credentials are allowed, echoes back whatever Origin the caller sent
    # together with Access-Control-Allow-Credentials: true — the opposite of
    # the "allow all (development) or none (production)" the comment claimed.
    #
    # Exploitation was blocked by SESSION_COOKIE_SAMESITE = 'Lax', which stops
    # the browser attaching the session cookie to a cross-site fetch. That is
    # a single setting standing between a misconfiguration and a cross-origin
    # data leak. This application renders server-side HTML on one origin and
    # needs no CORS policy at all.
    if allowed_origins:
        CORS(
            app,
            origins=allowed_origins,
            supports_credentials=True,
            methods=['GET', 'POST', 'PUT', 'PATCH', 'DELETE', 'OPTIONS'],
            max_age=3600,  # Cache preflight for 1 hour
        )

    db.init_app(app)
    login_manager.init_app(app)
    csrf.init_app(app)
    limiter.init_app(app)
    babel.init_app(app, locale_selector=i18n.select_locale)

    # Exposed to every template so the language switcher can render itself
    # without each view having to pass the list along.
    @app.before_request
    def generate_csp_nonce():
        # Only meaningful once inline script is disallowed; generated
        # unconditionally so templates can carry nonce="" beforehand.
        g.csp_nonce = secrets.token_urlsafe(16)

    @app.before_request
    def assign_request_id():
        """Give this request a name, so its log lines can be found (OBS-005).

        Every record emitted while handling it carries this id — see
        RequestIdFilter — which is what turns "an error happened around
        14:32" into the six lines that led to it. It goes back in
        X-Request-Id and onto the 500 page, so that a report of "it broke
        when I clicked save" is enough to find the trace.

        Generated here, never taken from an inbound header: with no trusted
        proxy settled (OPS-002), an accepted header lets any caller write
        arbitrary text — newlines included — into the log file.
        """
        g.request_id = secrets.token_hex(8)

    @app.after_request
    def expose_request_id(response):
        response.headers['X-Request-Id'] = g.get('request_id', '-')
        return response

    @app.url_defaults
    def version_static_urls(endpoint, values):
        """Stamp every static URL with the file's modification time.

        Without this, nginx cannot be allowed to cache style.css and main.js:
        their URLs never change, so a 30-day expiry means a 30-day-old stylesheet
        with no way to invalidate it short of telling people to hard-refresh.
        With it, a deployed file gets a new URL and the old entry simply stops
        being asked for — which is what makes the `immutable` in nginx.conf
        true rather than merely fast (PERF-006).

        The stamp is computed once per file per process. The process restarts
        on deploy, which is exactly when a file can have changed.
        """
        if endpoint != 'static' or 'filename' not in values:
            return
        filename = values['filename']
        stamp = _static_stamps.get(filename)
        if stamp is None:
            try:
                stamp = str(int(os.stat(os.path.join(app.static_folder, filename)).st_mtime))
            except OSError:
                # A missing file is the template's problem, not this hook's:
                # let the URL build and let the 404 say so.
                stamp = ''
            _static_stamps[filename] = stamp
        if stamp:
            values['v'] = stamp

    @app.context_processor
    def inject_csp_nonce():
        return {
            'csp_nonce': '' if app.config['CSP_ALLOW_INLINE_SCRIPT'] else g.get('csp_nonce', '')
        }

    @app.context_processor
    def inject_locales():
        from flask_babel import get_locale

        return {
            'current_locale': str(get_locale() or i18n.DEFAULT_LOCALE),
            'supported_locales': i18n.SUPPORTED_LOCALES,
            'locale_names': i18n.LOCALE_NAMES,
        }

    # Configure structured logging
    from app.logging_config import configure_logging

    configure_logging(app)

    from app.routes.auth import auth_bp
    from app.routes.evaluations import evaluations_bp
    from app.routes.main import main_bp
    from app.routes.matches import matches_bp
    from app.routes.team_matches import team_matches_bp
    from app.routes.teams import teams_bp
    from app.routes.tryouts import tryouts_bp
    from app.routes.users import users_bp

    app.register_blueprint(auth_bp)
    app.register_blueprint(tryouts_bp)
    app.register_blueprint(evaluations_bp)
    app.register_blueprint(users_bp)
    app.register_blueprint(main_bp)
    app.register_blueprint(teams_bp)
    app.register_blueprint(matches_bp)
    app.register_blueprint(team_matches_bp)

    # Register custom Jinja filters
    app.jinja_env.filters['nl2br'] = nl2br

    # =========================================================================
    # Security Headers
    # =========================================================================
    @app.after_request
    def add_security_headers(response):
        """Add security headers to all responses.

        Implements defense-in-depth with comprehensive HTTP security headers.
        These complement the headers set by Nginx in production.

        HSTS is only sent in production (non-debug) to avoid breaking
        local development over plain HTTP.
        """
        # X-XSS-Protection is deliberately not set: the auditor it addressed
        # has been removed from every current browser, and its last versions
        # introduced vulnerabilities of their own. CSP frame-ancestors and
        # X-Frame-Options cover the remaining ground.
        response.headers['X-Content-Type-Options'] = 'nosniff'
        response.headers['X-Frame-Options'] = 'DENY'
        response.headers['Referrer-Policy'] = 'strict-origin-when-cross-origin'
        response.headers['Permissions-Policy'] = (
            'camera=(), microphone=(), geolocation=(), interest-cohort=(), payment=(), usb=()'
        )
        response.headers['Cross-Origin-Opener-Policy'] = 'same-origin'
        response.headers['Content-Security-Policy'] = build_csp(
            allow_inline_script=app.config['CSP_ALLOW_INLINE_SCRIPT'],
            nonce=g.get('csp_nonce'),
        )

        # Only enable HSTS when HTTPS is actually being used
        # (either direct TLS or behind a proxy that terminates TLS)
        is_https = request.is_secure or request.headers.get('X-Forwarded-Proto') == 'https'
        if is_https:
            response.headers['Strict-Transport-Security'] = (
                'max-age=31536000; includeSubDomains; preload'
            )

        return response

    # =========================================================================
    # HTTPS Redirect (Production only)
    # =========================================================================
    @app.before_request
    def force_https():
        """Redirect all HTTP requests to HTTPS in production.

        Respects the X-Forwarded-Proto header from reverse proxies.
        Can be disabled via FORCE_HTTPS environment variable.

        Returns:
            Response | None: A redirect, or None to let the request through.
        """
        if not app.debug and app.config['FORCE_HTTPS']:
            already_secure = (
                request.is_secure or request.headers.get('X-Forwarded-Proto') == 'https'
            )
            if not already_secure:
                return redirect(request.url.replace('http://', 'https://'), code=301)
        return None

    # =========================================================================
    # Health Check Endpoint
    # =========================================================================
    @app.route('/health')
    def health_check():
        """Health check endpoint for monitoring and load balancers.

        Verifies database connectivity and application health.
        Returns 200 with basic status info or 503 if unhealthy.

        Returns:
            Response: JSON health status.
        """
        health_data = {
            'status': 'healthy',
            'app': 'team-tryouts',
            'version': '1.0.0',
        }

        # The bot runs in a daemon thread inside this process. When it dies
        # the site keeps serving pages and every notification stops, with
        # nothing to see from outside — which is how it stayed unnoticed.
        # Reported, not fatal: a club without Discord reminders is degraded,
        # not down, and a 503 here would take the site out of the load
        # balancer for it (OPS-012).
        if app.config['ENABLE_DISCORD_BOT']:
            from app.discord_bot import bot_status

            health_data['discord_bot'] = bot_status()

        # Check database connectivity
        try:
            db.session.execute(text('SELECT 1'))
            health_data['database'] = 'connected'
        except Exception:
            # Never echo the driver error: it routinely carries the host,
            # database name and user of the connection string, and /health
            # is unauthenticated.
            app.logger.error('Health check: database unreachable', exc_info=True)
            health_data['status'] = 'unhealthy'
            health_data['database'] = 'error'
            return jsonify(health_data), 503

        return jsonify(health_data), 200

    # =========================================================================
    # Custom Error Handlers
    # =========================================================================
    @app.errorhandler(400)
    def bad_request(error):
        """Handle 400 Bad Request errors.

        Args:
            error: The error object.

        Returns:
            Response: Rendered error page or JSON for API requests.
        """
        if (
            request.path.startswith('/users/disponibilities')
            or request.path.startswith('/users/coach-availability')
            or request.path.startswith('/users/api/')
        ):
            return jsonify({'error': 'Bad request', 'message': str(error)}), 400
        return render_template('errors/400.html', error=error), 400

    @app.errorhandler(401)
    def unauthorized(error):
        """Handle 401 Unauthorized errors.

        Args:
            error: The error object.

        Returns:
            Response: Redirect to login for pages, JSON for API.
        """
        if request.path.startswith('/users/disponibilities') or request.path.startswith(
            '/users/api/'
        ):
            return jsonify({'error': 'Unauthorized'}), 401
        from flask import flash as _flash

        _flash('Please log in to access this page.', 'warning')
        return redirect(url_for('auth.login'))

    @app.errorhandler(403)
    def forbidden(error):
        """Handle 403 Forbidden errors.

        Args:
            error: The error object.

        Returns:
            Response: Rendered error page or JSON for API requests.
        """
        if request.path.startswith('/users/disponibilities') or request.path.startswith(
            '/users/api/'
        ):
            return jsonify({'error': 'Forbidden', 'message': str(error)}), 403
        return render_template('errors/403.html', error=error), 403

    @app.errorhandler(404)
    def not_found(error):
        """Handle 404 Not Found errors.

        Args:
            error: The error object.

        Returns:
            Response: Rendered error page or JSON for API requests.
        """
        if request.path.startswith('/users/disponibilities') or request.path.startswith(
            '/users/api/'
        ):
            return jsonify({'error': 'Not found'}), 404
        return render_template('errors/404.html', error=error), 404

    @app.errorhandler(429)
    def too_many_requests(error):
        """Handle 429 Too Many Requests errors.

        Args:
            error: The error object.

        Returns:
            Response: JSON error for API or rendered page.
        """
        if request.path.startswith('/users/disponibilities') or request.path.startswith(
            '/users/api/'
        ):
            return jsonify(
                {'error': 'Too many requests', 'message': 'Please try again later.'}
            ), 429
        return render_template('errors/429.html', error=error), 429

    @app.errorhandler(500)
    def internal_error(error):
        """Handle 500 Internal Server Error.

        Never exposes stack traces to users. Logs the full error internally.

        Args:
            error: The error object.

        Returns:
            Response: Generic error page or JSON.
        """
        # Log the full error for debugging
        app.logger.error('Internal Server Error: %s', str(error), exc_info=True)

        # Roll back any failed database session
        db.session.rollback()

        # The id is the only thing that connects a user saying "it broke when
        # I clicked save" to the stack trace in errors.log. It identifies one
        # request and nothing else — no session, no account, nothing an
        # attacker can use — so showing it costs nothing (OBS-005).
        request_id = g.get('request_id', '-')

        if request.path.startswith('/users/disponibilities') or request.path.startswith(
            '/users/api/'
        ):
            return jsonify(
                {
                    'error': 'Internal server error',
                    'message': 'An unexpected error occurred. Please try again later.',
                    'request_id': request_id,
                }
            ), 500
        return render_template('errors/500.html', request_id=request_id), 500

    @app.errorhandler(HTTPException)
    def handle_http_exception(error):
        """Catch-all handler for any unhandled HTTP exceptions.

        Args:
            error: The HTTPException object.

        Returns:
            Response: JSON error for API, re-raises for others.
        """
        if request.path.startswith('/users/disponibilities') or request.path.startswith(
            '/users/api/'
        ):
            return jsonify(
                {'error': error.name, 'message': error.description, 'code': error.code}
            ), error.code
        return error

    # =========================================================================
    # Database Initialization
    # =========================================================================
    with app.app_context():
        import app.models as models  # noqa: F401 — registers all models with SQLAlchemy

        # NOTE: create_all() only ever creates missing tables. It never adds a
        # column to an existing one, so a model change is silently absent from
        # any database that already has the table. Replacing this with Alembic
        # is tracked as DB-002/DB-004; until then the behaviour is preserved.
        if app.config['AUTO_CREATE_TABLES']:
            db.create_all()

    # Start the Discord bot for notifications
    if app.config['ENABLE_DISCORD_BOT']:
        try:
            from app.discord_bot import start_bot

            start_bot(flask_app=app)
        except Exception:  # the site must come up even if the bot cannot
            # With the message alone, the two ways this fails — a bad token
            # and a broken import in discord_bot — read identically, and
            # neither is diagnosable from one line. Notifications are down
            # either way, so the traceback is the whole value of the log.
            # Error, not warning: a club that receives no reminders has lost
            # a feature, and the old level put that next to the deprecation
            # notices.
            app.logger.error(
                'Could not start the Discord bot. The site is up; no notification '
                'will be sent until this is fixed.',
                exc_info=True,
            )

    return app


# No __main__ block here on purpose. There used to be one, and with run.py
# and wsgi.py that made three ways to start the application, each with its
# own host, port and debug default — `python app/app.py` bound 0.0.0.0:10000
# while `python run.py` bound 127.0.0.2:5000 with the debugger on. This
# module defines the factory; run.py starts it for development, wsgi.py for
# production (ARCH-007).