vocero-client/vocero-api/vocero_api/providers/caller.py

from __future__ import annotations

import json
from typing import Any
from urllib.parse import urlencode

import httpx

from vocero_api.core import config as settings
from vocero_api.core import logging as app_logging


async def _fetch_reports_list(phone_number: str | None = None, auth_token: str | None = None) -> list | dict | None:
    """Fetch scheduled reports list from S2S_REPORT_LIST.

    Completely generic / deploy-agnostic: performs GET ${S2S_REPORT_LIST} and appends
    ?phoneNumber=<x> if a phone is provided. Accepts JSON (optional {"result": [...]}
    wrapper) and returns the list or unwrapped list. On any error logs and returns None
    (never fails the proxy / reasoning tunnel).

    USER FIX 2026-09-11: remove mass injection of scheduled reports.
      • If S2S_INJECT_REPORTS is FALSE (DEFAULT) we short-circuit to None at the
        very top — NO HTTP call is made to S2S_REPORT_LIST at all. The upstream
        endpoint is never hit, not even for a "cheap empty list"; combined with
        _resolve_reports_context → "" above, this guarantees ZERO scheduled-report
        contents reach any LLM prompt anywhere (WS session instructions patch,
        /api/reasoning user_context, reasoning instructions augment, quick reports
        lookup response).
      • S2S_INJECT_REPORTS=true restores prior behaviour (both the HTTP fetch AND
        the full injection into every prompt path).
    """
    # ── MASS-INJECTION KILL SWITCH (user request) ─────────────────────────
    if not settings.S2S_INJECT_REPORTS:
        return None
    # ───────────────────────────────────────────────────────────────────────
    if not settings.S2S_REPORT_LIST:
        return None
    if phone_number:
        sep = "&" if "?" in settings.S2S_REPORT_LIST else "?"
        target_url = f"{settings.S2S_REPORT_LIST}{sep}{urlencode({'phoneNumber': phone_number})}"
    else:
        target_url = settings.S2S_REPORT_LIST
    app_logging.logger.info(
        "[INJECT-REPORTS] fetch upstream URL=%s phoneLen=%s bearer=%s",
        settings.S2S_REPORT_LIST,
        len(phone_number) if phone_number else "none",
        "yes" if auth_token else "no",
    )
    try:
        headers = {}
        if auth_token:
            headers["Authorization"] = f"Bearer {auth_token}"
        async with httpx.AsyncClient(timeout=15.0, follow_redirects=True) as http:
            resp = await http.get(target_url, headers=headers or None)
        if resp.status_code != 200:
            app_logging.logger.warning(
                "[INJECT-REPORTS] ⛔ upstream HTTP %s response=%.250s",
                resp.status_code,
                resp.text,
            )
            return None
        text = (resp.text or "").strip()
        if not text:
            app_logging.logger.info("[INJECT-REPORTS] ⏭️ upstream returned empty body.")
            return None
        try:
            payload = resp.json()
        except Exception:
            app_logging.logger.warning(
                "[INJECT-REPORTS] ⛔ upstream returned non-JSON body len=%d; cannot parse list. response=%.250s",
                len(text),
                text,
            )
            return None
    except Exception as exc:
        app_logging.logger.warning(
            "[INJECT-REPORTS] ⛔ fetch unreachable error=%r",
            exc,
        )
        return None
    # Unwrap {"result": [...]} or other common list-wrapping dict keys
    # (e.g. {"data": [...], "reports": [...], "items": [...], "content": [...], "value": [...]}).
    unwrapped: object = payload
    if isinstance(payload, dict):
        # Log full structure for debugging if enabled
        if settings.S2S_DEBUG_PROMPT:
            try:
                keys_preview = ",".join(sorted(str(k) for k in payload.keys()))
                preview = json.dumps(payload, ensure_ascii=False)
                preview = preview[:4000]
                app_logging.logger.info(
                    "[DEBUG-REPORTS][fetch] upstream response dict keys=[%s] full PREVIEW (len=%d):\n%s\n",
                    keys_preview,
                    len(preview),
                    preview,
                )
            except Exception:
                pass
        # Preferred explicit "result" key first
        if "result" in payload and isinstance(payload["result"], list):
            unwrapped = payload["result"]
        else:
            # Scan all common list-wrapper keys (case-insensitive match on the key)
            wrapper_keys = {
                k.lower()
                for k in (
                    "data",
                    "reports",
                    "reportList",
                    "report_list",
                    "report",
                    "items",
                    "itemList",
                    "item_list",
                    "list",
                    "content",
                    "scheduled",
                    "scheduledReports",
                    "scheduled_reports",
                    "podcastList",
                    "podcast_list",
                    "podcasts",
                    "entries",
                    "records",
                    "rows",
                    "pageContent",
                    "page_content",
                    "value",
                )
            }
            for k, v in payload.items():
                if isinstance(k, str) and isinstance(v, list) and k.lower() in wrapper_keys:
                    unwrapped = v
                    break
            # Fallback: if it's still a dict with only ONE list-valued key,
            # assume that's the list (common pattern like {"data": [...]} with
            # unusual key name we don't hardcode).
            if not isinstance(unwrapped, list):
                list_valued_keys = [k for k, v in payload.items() if isinstance(v, list) and len(v) > 0]
                if len(list_valued_keys) == 1:
                    unwrapped = payload[list_valued_keys[0]]
    # Validate array shape
    if isinstance(unwrapped, list):
        n_items = len(unwrapped)
        n_dict = sum(1 for x in unwrapped if isinstance(x, dict))
        app_logging.logger.info(
            "[INJECT-REPORTS] ✅ reports list loaded OK (list len=%d dictItems=%d)",
            n_items,
            n_dict,
        )
        return unwrapped
    app_logging.logger.info(
        "[INJECT-REPORTS] ⏭️ upstream returned type=%s (not list); nothing to inject.",
        type(unwrapped).__name__,
    )
    if settings.S2S_DEBUG_PROMPT and isinstance(payload, dict):
        # Log final top-level keys to help diagnose wrapper shape
        app_logging.logger.info(
            "[DEBUG-REPORTS][fetch] wrapper not detected — top-level keys of dict response: [%s] "
            "(list values under keys: %s)",
            ",".join(sorted(str(k) for k in payload.keys())),
            ",".join(sorted(str(k) + "=list[" + str(len(v)) + "]" for k, v in payload.items() if isinstance(v, list)))
            or "none",
        )
    return None


def _extract_login_token(login_payload: dict | None) -> str | None:
    """Given the parsed response from the S2S_LOGIN_BY_PHONE endpoint (already
    {"result":…}-unwrapped), try to extract an auth / session token.
    Looks under common names (case-insensitive). Returns None if nothing found.
    """
    if not login_payload or not isinstance(login_payload, dict):
        return None
    # Unwrap one level of nesting if result is still an dict: handled by caller already.
    candidates = {
        c.lower()
        for c in (
            "token",
            "jwt",
            "access_token",
            "accessToken",
            "auth_token",
            "authToken",
            "session_token",
            "sessionToken",
            "id_token",
            "idToken",
            "bearer",
            "bearer_token",
            "bearerToken",
            "authentication_token",
        )
    }
    for k, v in login_payload.items():
        if isinstance(k, str) and k.lower() in candidates and isinstance(v, str) and v.strip():
            return v.strip()
    # Second pass: nested dict {"data": {"token": …}} or {"auth": {"jwt": …}}
    for _k, v in login_payload.items():
        if isinstance(v, dict):
            nested = _extract_login_token(v)
            if nested:
                return nested
    # Third pass: list like [{"token": ...}] (uncommon but safe)
    for _k, v in login_payload.items():
        if isinstance(v, list) and v:
            for elem in v:
                if isinstance(elem, dict):
                    nested = _extract_login_token(elem)
                    if nested:
                        return nested
    return None


async def _login_by_phone(phone_number: str) -> tuple[str | None, dict | None]:
    """Perform login-by-phone: POST S2S_LOGIN_BY_PHONE with JSON
    {"phoneNumber": phone_number}. Returns (bearer_token_or_None, response_obj_or_None).

    Never fails: on any HTTP/parse error logs a warning and returns (None, None)
    so subsequent profile/summary calls simply proceed without a Bearer token.
    """
    if not settings.S2S_LOGIN_BY_PHONE or not phone_number:
        return None, None
    body = {"phoneNumber": phone_number}
    app_logging.logger.info(
        "[INJECT-LOGIN] POST upstream URL=%s phoneLen=%d",
        settings.S2S_LOGIN_BY_PHONE,
        len(phone_number),
    )
    try:
        async with httpx.AsyncClient(timeout=10.0, follow_redirects=True) as http:
            resp = await http.post(
                settings.S2S_LOGIN_BY_PHONE,
                json=body,
                headers={"Content-Type": "application/json"},
            )
        if resp.status_code != 200:
            app_logging.logger.warning(
                "[INJECT-LOGIN] ⛔ upstream HTTP %s phoneLen=%d response=%.250s",
                resp.status_code,
                len(phone_number),
                resp.text,
            )
            return None, None
        text = (resp.text or "").strip()
        if not text:
            app_logging.logger.info("[INJECT-LOGIN] ⏭️ upstream returned empty body; no login token available.")
            return None, None
        try:
            payload = resp.json()
        except Exception:
            app_logging.logger.warning(
                "[INJECT-LOGIN] ⛔ upstream returned non-JSON body len=%d; cannot extract token. response=%.250s",
                len(text),
                text,
            )
            return None, None
    except Exception as exc:
        app_logging.logger.warning(
            "[INJECT-LOGIN] ⛔ fetch unreachable phoneLen=%d error=%r",
            len(phone_number),
            exc,
        )
        return None, None

    # Unwrap {"result": …} OR {"data": …} if present (common wrapper conventions).
    unwrapped: dict | None = None
    if isinstance(payload, dict):
        for _wrap_key in ("result", "data", "payload"):
            if isinstance(payload.get(_wrap_key), dict):
                unwrapped = payload[_wrap_key]
                break
        if unwrapped is None:
            unwrapped = payload
    token = (
        _extract_login_token(unwrapped)
        if isinstance(unwrapped, dict)
        else _extract_login_token(payload if isinstance(payload, dict) else None)
    )
    # Always log login-token diagnostics, even without S2S_DEBUG_PROMPT:
    # token length, first eight characters, response keys and wrapper layer.
    _all_keys: list[str] = []
    if isinstance(payload, dict):
        _all_keys.extend(str(k) for k in payload.keys())
    _unwrap_keys: list[str] = []
    if isinstance(unwrapped, dict):
        _unwrap_keys.extend(str(k) for k in unwrapped.keys())
    if isinstance(unwrapped, dict):
        _t_preview = ""
        _t_len = 0
        if isinstance(token, str) and token.strip():
            _t_len = len(token.strip())
            if _t_len > 8:
                _t_preview = token.strip()[:8] + "..."
            else:
                _t_preview = token.strip()
        _fmt_note = ""
        if _t_len >= 120:
            _fmt_note = " [JWT-like]"
        elif _t_len > 0:
            _fmt_note = " [SHORT]"
        _wrap_layer = "ROOT"
        if isinstance(payload, dict):
            for _wk in ("result", "data", "payload"):
                if isinstance(payload.get(_wk), dict):
                    _wrap_layer = "result/data/payload"
                    break
        _all_keys_s = ",".join(sorted(_all_keys)) if _all_keys else "-"
        _unwrap_keys_s = ",".join(sorted(_unwrap_keys)) if _unwrap_keys else "-"
        _t_info = "MISSING"
        if _t_len:
            _t_info = "YES len=" + str(_t_len) + " head=" + _t_preview
        msg = f"[INJECT-LOGIN] ✅ login OK: wrapLayer={_wrap_layer} topKeys={_all_keys_s} unwrappedKeys={_unwrap_keys_s} token={_t_info}{_fmt_note}"
        app_logging.logger.info("%s", msg)
        return token, unwrapped
    # Fallback: non-dict payload (string/list/number) — still log it, but no token expected.
    app_logging.logger.info(
        "[INJECT-LOGIN] ⏭️ login response parsed but type=%s (not dict); no token extracted.",
        type(payload).__name__,
    )
    if settings.S2S_DEBUG_PROMPT:
        app_logging.logger.info(
            "[DEBUG-LOGIN][response] FULL login payload (len=%d):\n%s\n",
            len(str(payload)),
            payload,
        )
    return None, None


async def _fetch_latest_summary_for_phone(phone_number: str, auth_token: str | None = None) -> Any:
    """Fetch the latest caller summary via S2S_LATEST_SUMMARY URL.

    Generic / completely deploy-agnostic: plain GET
    S2S_LATEST_SUMMARY?phoneNumber=<phone>; accepts JSON (optional {"result": …}
    wrapper) or plain text. Never fails the proxy / reasoning; returns None on
    any error and logs it. If `auth_token` is provided it is sent as
    Authorization: Bearer <auth_token>.

    OVERRIDE (2026-09-23): when S2S_LATEST_SUMMARY_TEXT is set (non-empty) it
    takes absolute priority over S2S_LATEST_SUMMARY. No HTTP call is made at all;
    the raw plain-text value is returned directly. Designed for operators who want
    a clean, pre-curated, essential summary snippet without any upstream API
    dependency. Works even without a phone_number (static snippet).
    """
    if settings.S2S_LATEST_SUMMARY_TEXT:
        override_len = len(settings.S2S_LATEST_SUMMARY_TEXT)
        app_logging.logger.info(
            "[INJECT-SUMMARY] S2S_LATEST_SUMMARY_TEXT OVERRIDE active (len=%d chars). Skipping HTTP fetch; injecting raw static plain-text summary directly.",
            override_len,
        )
        if settings.S2S_DEBUG_PROMPT:
            app_logging.logger.info(
                "[INJECT-SUMMARY] [DEBUG-PROMPT] S2S_LATEST_SUMMARY_TEXT raw value (first 800 chars):\n%s",
                settings.S2S_LATEST_SUMMARY_TEXT[:800],
            )
        return settings.S2S_LATEST_SUMMARY_TEXT
    if not settings.S2S_LATEST_SUMMARY or not phone_number:
        return None
    target_url = f"{settings.S2S_LATEST_SUMMARY}?{urlencode({'phoneNumber': phone_number})}"
    app_logging.logger.info(
        "[INJECT-SUMMARY] fetch upstream URL=%s phoneLen=%d bearer=%s",
        settings.S2S_LATEST_SUMMARY,
        len(phone_number),
        "yes" if auth_token else "no",
    )
    try:
        headers = {}
        if auth_token:
            headers["Authorization"] = f"Bearer {auth_token}"
        async with httpx.AsyncClient(timeout=10.0, follow_redirects=True) as http:
            resp = await http.get(target_url, headers=headers or None)
        if resp.status_code != 200:
            app_logging.logger.warning(
                "[INJECT-SUMMARY] ⛔ upstream HTTP %s phoneLen=%d response=%.200s",
                resp.status_code,
                len(phone_number),
                resp.text,
            )
            return None
        # Accept JSON first, then fall back to plain text (some summary APIs
        # return text/plain).
        text = (resp.text or "").strip()
        if not text:
            return ""
        try:
            payload = resp.json()
        except Exception:
            app_logging.logger.info(
                "[INJECT-SUMMARY] upstream returned non-JSON response (len=%d); treating as plain string summary.",
                len(text),
            )
            return text
    except Exception as exc:
        app_logging.logger.warning(
            "[INJECT-SUMMARY] ⛔ fetch unreachable phoneLen=%d error=%r",
            len(phone_number),
            exc,
        )
        return None
    # Unwrap {"result": …} just like profile fetch (common wrapper pattern).
    if isinstance(payload, dict) and "result" in payload:
        payload = payload["result"]
    # Filter out empty dict/list / empty string after unwrap.
    empty = (
        payload is None
        or (isinstance(payload, str) and not payload.strip())
        or (isinstance(payload, dict) and len(payload) == 0)
        or (isinstance(payload, list) and len(payload) == 0)
    )
    if empty:
        app_logging.logger.info("[INJECT-SUMMARY] ⏭️  upstream returned empty summary; nothing to inject.")
        return None
    if isinstance(payload, dict):
        app_logging.logger.info(
            "[INJECT-SUMMARY] ✅ summary loaded OK (object) keys=%s",
            ",".join(sorted(str(k) for k in payload.keys())),
        )
    elif isinstance(payload, list):
        app_logging.logger.info(
            "[INJECT-SUMMARY] ✅ summary loaded OK (list) len=%d",
            len(payload),
        )
    else:
        app_logging.logger.info(
            "[INJECT-SUMMARY] ✅ summary loaded OK (string) len=%d",
            len(str(payload)),
        )
    return payload


async def _fetch_podcast_episode(episode_id: str) -> Any:
    """Fetch a single podcast episode by ID from the Aquachile AzurianLab API.

    Completely generic / deploy-agnostic: plain GET
    https://aquachile.azurianlab.com/api/v1/podcast/episodes/<episode_id>
    Accepts JSON (optional {"result": …} wrapper) or plain text. Never fails
    the proxy / reasoning; returns None on any error and logs it.
    """
    if not episode_id:
        return None
    target_url = f"https://aquachile.azurianlab.com/api/v1/podcast/episodes/{episode_id}"
    app_logging.logger.info(
        "[INJECT-EPISODE] fetch upstream URL=%s episodeId=%.80s",
        target_url,
        episode_id,
    )
    try:
        async with httpx.AsyncClient(timeout=15.0, follow_redirects=True) as http:
            resp = await http.get(target_url)
        if resp.status_code != 200:
            app_logging.logger.warning(
                "[INJECT-EPISODE] ⛔ upstream HTTP %s episodeId=%.80s response=%.200s",
                resp.status_code,
                episode_id,
                resp.text,
            )
            return None
        text = (resp.text or "").strip()
        if not text:
            return ""
        try:
            payload = resp.json()
        except Exception:
            app_logging.logger.info(
                "[INJECT-EPISODE] upstream returned non-JSON response (len=%d); treating as plain string.",
                len(text),
            )
            return text
    except Exception as exc:
        app_logging.logger.warning(
            "[INJECT-EPISODE] ⛔ fetch unreachable episodeId=%.80s error=%r",
            episode_id,
            exc,
        )
        return None
    if isinstance(payload, dict) and "result" in payload:
        payload = payload["result"]
    empty = (
        payload is None
        or (isinstance(payload, str) and not payload.strip())
        or (isinstance(payload, dict) and len(payload) == 0)
        or (isinstance(payload, list) and len(payload) == 0)
    )
    if empty:
        app_logging.logger.info("[INJECT-EPISODE] ⏭️  upstream returned empty episode; nothing to inject.")
        return None
    if isinstance(payload, dict):
        app_logging.logger.info(
            "[INJECT-EPISODE] ✅ episode loaded OK (object) keys=%s",
            ",".join(sorted(str(k) for k in payload.keys())),
        )
    elif isinstance(payload, list):
        app_logging.logger.info(
            "[INJECT-EPISODE] ✅ episode loaded OK (list) len=%d",
            len(payload),
        )
    else:
        app_logging.logger.info(
            "[INJECT-EPISODE] ✅ episode loaded OK (string) len=%d",
            len(str(payload)),
        )
    return payload


async def _fetch_user_profile_for_phone(phone_number: str, auth_token: str | None = None) -> dict | None:
    """Fetch the profile JSON from VOCERO_USER_PHONE_AUTH_URL for a phone.
    Returns the raw {"result": ...}-unwrapped object or None. Errors are logged
    then suppressed (never fail the proxy / reasoning because of this).

    When `auth_token` is provided it is forwarded as Authorization: Bearer <auth_token>
    to the upstream profile endpoint.
    """
    if not settings.VOCERO_USER_PHONE_AUTH_URL or not phone_number:
        return None
    target_url = f"{settings.VOCERO_USER_PHONE_AUTH_URL}?{urlencode({'phoneNumber': phone_number})}"
    app_logging.logger.info(
        "[INJECT-PROFILE] fetch upstream URL=%s phoneLen=%d bearer=%s",
        settings.VOCERO_USER_PHONE_AUTH_URL,
        len(phone_number),
        "yes" if auth_token else "no",
    )
    try:
        headers = {}
        if auth_token:
            headers["Authorization"] = f"Bearer {auth_token}"
        async with httpx.AsyncClient(timeout=10.0, follow_redirects=True) as http:
            resp = await http.get(target_url, headers=headers or None)
        if resp.status_code != 200:
            app_logging.logger.info(
                "[INJECT-PROFILE] ⛔ upstream HTTP %s phoneLen=%d response=%.200s",
                resp.status_code,
                len(phone_number),
                resp.text,
            )
            return None
        payload = resp.json()
    except Exception as exc:
        app_logging.logger.warning(
            "[INJECT-PROFILE] ⛔ fetch unreachable phoneLen=%d error=%r",
            len(phone_number),
            exc,
        )
        return None
    if isinstance(payload, dict) and "result" in payload and isinstance(payload["result"], dict):
        profile_obj = payload["result"]
    elif isinstance(payload, dict) and not isinstance(payload, list):
        profile_obj = payload
    else:
        app_logging.logger.info(
            "[INJECT-PROFILE] ⛔ upstream returned non-object JSON type=%s keys=?",
            type(payload).__name__,
        )
        return None
    app_logging.logger.info(
        "[INJECT-PROFILE] ✅ profile loaded OK keys=%s",
        ",".join(sorted(profile_obj.keys())),
    )
    return profile_obj