| 1 | """Xquik X search source for the v3.0.0 last30days pipeline. |
| 2 | |
| 3 | Uses the Xquik REST API (https://xquik.com/api/v1) to search X/Twitter |
| 4 | with full engagement metrics (likes, retweets, replies, quotes, views, |
| 5 | bookmarks). Requires an API key from xquik.com. |
| 6 | """ |
| 7 | |
| 8 | from __future__ import annotations |
| 9 | |
| 10 | from datetime import datetime |
| 11 | from typing import Any, Dict, List, Optional |
| 12 | |
| 13 | from . import http, log |
| 14 | from .relevance import token_overlap_relevance as _compute_relevance |
| 15 | from .x_api import is_own_post |
| 16 | |
| 17 | # Per-process probe cache: (state, reason). state is "unset" until probed, then |
| 18 | # True (funded/working) | False (auth/payment failure) | None (inconclusive). |
| 19 | _probe_cache: tuple = ("unset", "") |
| 20 | |
| 21 | # Depth configurations: number of results to request per query |
| 22 | DEPTH_CONFIG = { |
| 23 | "quick": {"limit": 10, "queries": 1}, |
| 24 | "default": {"limit": 20, "queries": 2}, |
| 25 | "deep": {"limit": 40, "queries": 3}, |
| 26 | } |
| 27 | |
| 28 | _BASE_URL = "https://xquik.com/api/v1" |
| 29 | |
| 30 | |
| 31 | def _log(msg: str): |
| 32 | log.source_log("Xquik", msg, tty_only=False) |
| 33 | |
| 34 | |
| 35 | def _extract_core_subject(topic: str) -> str: |
| 36 | """Extract core subject for X search queries.""" |
| 37 | from .query import extract_core_subject |
| 38 | return extract_core_subject(topic, max_words=5, strip_suffixes=True) |
| 39 | |
| 40 | |
| 41 | def expand_xquik_queries(topic: str, depth: str) -> List[str]: |
| 42 | """Generate query variants based on depth. |
| 43 | |
| 44 | Args: |
| 45 | topic: Research topic |
| 46 | depth: "quick", "default", or "deep" |
| 47 | |
| 48 | Returns: |
| 49 | List of query strings (1 for quick, 2 for default, 3 for deep). |
| 50 | """ |
| 51 | core = _extract_core_subject(topic) |
| 52 | # Anti-bare-generic guard (#607): never let the core collapse to a single |
| 53 | # bare token when the topic carries more — a lone generic word floods X with |
| 54 | # off-topic collisions. Fall back to the full multi-word topic as the anchor. |
| 55 | topic_clean = topic.strip() |
| 56 | if len(core.split()) <= 1 and len(topic_clean.split()) > 1 and core.lower() != topic_clean.lower(): |
| 57 | core = topic_clean |
| 58 | queries = [core] |
| 59 | |
| 60 | # Add original topic if meaningfully different |
| 61 | if topic.lower().strip() != core.lower().strip(): |
| 62 | queries.append(topic.strip()) |
| 63 | |
| 64 | # Add compound term variant for deep searches |
| 65 | if len(queries) < 3: |
| 66 | from .query import extract_compound_terms |
| 67 | compounds = extract_compound_terms(topic) |
| 68 | if compounds: |
| 69 | or_parts = " OR ".join(f'"{t}"' for t in compounds[:3]) |
| 70 | queries.append(f"({or_parts})") |
| 71 | |
| 72 | cap = DEPTH_CONFIG.get(depth, DEPTH_CONFIG["default"])["queries"] |
| 73 | return queries[:cap] |
| 74 | |
| 75 | |
| 76 | def search_xquik( |
| 77 | topic: str, |
| 78 | from_date: str, |
| 79 | to_date: str, |
| 80 | depth: str = "default", |
| 81 | token: str = "", |
| 82 | ) -> Dict[str, Any]: |
| 83 | """Search X via Xquik REST API. |
| 84 | |
| 85 | Args: |
| 86 | topic: Search topic |
| 87 | from_date: Start date (YYYY-MM-DD) |
| 88 | to_date: End date (YYYY-MM-DD) |
| 89 | depth: Research depth - "quick", "default", or "deep" |
| 90 | token: Xquik API key |
| 91 | |
| 92 | Returns: |
| 93 | Dict with "items" list and optional "error" string. |
| 94 | """ |
| 95 | if not token: |
| 96 | return {"items": [], "error": "No XQUIK_API_KEY configured"} |
| 97 | |
| 98 | cfg = DEPTH_CONFIG.get(depth, DEPTH_CONFIG["default"]) |
| 99 | queries = expand_xquik_queries(topic, depth) |
| 100 | all_items: List[Dict[str, Any]] = [] |
| 101 | seen_ids: set[str] = set() |
| 102 | |
| 103 | for query_text in queries: |
| 104 | q = f"{query_text} since:{from_date} until:{to_date}" |
| 105 | items, auth_error = _execute_search( |
| 106 | q, cfg["limit"], token, |
| 107 | label=query_text, id_prefix="XQ", |
| 108 | seen_ids=seen_ids, relevance_query=query_text, |
| 109 | index_offset=len(all_items), |
| 110 | ) |
| 111 | if auth_error: |
| 112 | # Auth/payment failure is fatal for the whole source (e.g. 401/403, |
| 113 | # and 402-unpaid surfaced via U5 diagnose) — return it so the caller |
| 114 | # settles honestly instead of silently empty. |
| 115 | return {"items": [], "error": auth_error} |
| 116 | all_items.extend(items) |
| 117 | |
| 118 | return {"items": all_items} |
| 119 | |
| 120 | |
| 121 | def _execute_search( |
| 122 | q: str, |
| 123 | limit: int, |
| 124 | token: str, |
| 125 | *, |
| 126 | label: str, |
| 127 | id_prefix: str, |
| 128 | seen_ids: set[str], |
| 129 | relevance_query: str, |
| 130 | index_offset: int = 0, |
| 131 | failure_out: Optional[List[str]] = None, |
| 132 | ) -> tuple[List[Dict[str, Any]], str | None]: |
| 133 | """Run one Xquik search call and parse its tweets. |
| 134 | |
| 135 | Returns ``(items, auth_error)``. ``auth_error`` is a non-empty string only |
| 136 | on a fatal auth/payment failure (401/403); transient/HTTP errors log and |
| 137 | return ``([], None)`` so one bad lane never discards another's results. |
| 138 | |
| 139 | Non-fatal failures (429, 5xx, network error, unparseable payload) append a |
| 140 | reason to ``failure_out`` instead. They stay out of the return value |
| 141 | because the caller must keep going to the next handle, but a caller that |
| 142 | stopped there would report the lane as empty rather than as failed. |
| 143 | ``relevance_query`` (the topic) is what items are scored against — for the |
| 144 | handle lanes that differs from the search query (``from:handle``). |
| 145 | ``index_offset`` keeps item ids unique across multiple calls that share an |
| 146 | accumulator (multi-query topic search, per-handle lanes). |
| 147 | """ |
| 148 | full_url = f"{_BASE_URL}/x/tweets/search?q={_url_encode(q)}&queryType=Top&limit={limit}" |
| 149 | _log(f"Searching: {label}") |
| 150 | try: |
| 151 | request_headers = {"X-Api-Key": token} |
| 152 | response = http.get(full_url, headers=request_headers, timeout=30, retries=2) |
| 153 | except http.HTTPError as exc: |
| 154 | status = getattr(exc, "status_code", None) |
| 155 | if status == 402: |
| 156 | # Unpaid key — fatal for the source, and surfaced on the real search |
| 157 | # path (not just --diagnose) so a live run reports it instead of |
| 158 | # settling silently empty. The X retrieval branch classifies by |
| 159 | # message text only, so the detail carries the "payment required" |
| 160 | # marker that http.classify_failure maps to PAYMENT_REQUIRED. |
| 161 | return [], "Xquik key unpaid: payment required (402)" |
| 162 | if status in (401, 403): |
| 163 | return [], f"Xquik auth failed ({status})" |
| 164 | _log(f"HTTP error for '{label}': {exc}") |
| 165 | if failure_out is not None: |
| 166 | failure_out.append(f"Xquik HTTP error for '{label}': {exc}") |
| 167 | return [], None |
| 168 | except Exception as exc: |
| 169 | _log(f"Error for '{label}': {exc}") |
| 170 | if failure_out is not None: |
| 171 | failure_out.append(f"Xquik request failed for '{label}': {exc}") |
| 172 | return [], None |
| 173 | |
| 174 | tweets = response.get("tweets", []) |
| 175 | if not isinstance(tweets, list): |
| 176 | if failure_out is not None: |
| 177 | failure_out.append( |
| 178 | f"Xquik returned no tweets array for '{label}' " |
| 179 | f"(got {type(tweets).__name__})" |
| 180 | ) |
| 181 | return [], None |
| 182 | items: List[Dict[str, Any]] = [] |
| 183 | for tweet in tweets: |
| 184 | if not isinstance(tweet, dict): |
| 185 | continue |
| 186 | tweet_id = str(tweet.get("id", "")) |
| 187 | if tweet_id in seen_ids: |
| 188 | continue |
| 189 | seen_ids.add(tweet_id) |
| 190 | item = _parse_tweet(tweet, index_offset + len(items), relevance_query, id_prefix=id_prefix) |
| 191 | if item: |
| 192 | items.append(item) |
| 193 | return items, None |
| 194 | |
| 195 | |
| 196 | # True when a tweet URL is authored by ``handle`` (their own post). Used by |
| 197 | # the ABOUT lane to drop the subject's own tweets so only mentions *by |
| 198 | # others* remain; the implementation is ``x_api.is_own_post``. |
| 199 | _is_own = is_own_post |
| 200 | |
| 201 | |
| 202 | def search_handles( |
| 203 | handles: List[str], |
| 204 | topic: str, |
| 205 | from_date: str, |
| 206 | to_date: str, |
| 207 | *, |
| 208 | count_per: int = 8, |
| 209 | token: str = "", |
| 210 | failure_out: Optional[List[str]] = None, |
| 211 | ) -> List[Dict[str, Any]]: |
| 212 | """FROM lane: tweets authored BY each handle (their own timeline). |
| 213 | |
| 214 | The topic is NOT AND'd into the query (that was the from:-AND bug, #610) — |
| 215 | we pull the raw timeline and use ``topic`` for relevance ranking only. |
| 216 | Returns a flat list of item dicts (mirrors ``bird_x.search_handles``). |
| 217 | |
| 218 | When ``failure_out`` is provided, any failure reason is appended — the |
| 219 | fatal auth/payment one that stops the lane, and the per-handle transient |
| 220 | ones (429, 5xx, network, unparseable payload) that do not — so the caller |
| 221 | can distinguish a failed request from a handle that posted nothing. |
| 222 | """ |
| 223 | if not token or not handles: |
| 224 | return [] |
| 225 | items: List[Dict[str, Any]] = [] |
| 226 | seen_ids: set[str] = set() |
| 227 | for raw in handles: |
| 228 | handle = str(raw).lstrip("@").strip() |
| 229 | if not handle: |
| 230 | continue |
| 231 | q = f"from:{handle} since:{from_date} until:{to_date}" |
| 232 | got, auth_error = _execute_search( |
| 233 | q, count_per, token, |
| 234 | label=f"from:{handle}", id_prefix="XF", |
| 235 | seen_ids=seen_ids, relevance_query=topic, |
| 236 | index_offset=len(items), |
| 237 | failure_out=failure_out, |
| 238 | ) |
| 239 | if auth_error: |
| 240 | # Fatal auth/payment failure — stop, keep what we have. Surface the |
| 241 | # reason: an empty FROM lane is otherwise indistinguishable from a |
| 242 | # subject who simply did not post. |
| 243 | if failure_out is not None: |
| 244 | failure_out.append(auth_error) |
| 245 | break |
| 246 | items.extend(got) |
| 247 | return items |
| 248 | |
| 249 | |
| 250 | def search_mentions( |
| 251 | handles: List[str], |
| 252 | from_date: str, |
| 253 | to_date: str, |
| 254 | *, |
| 255 | topic: str = "", |
| 256 | count_per: int = 5, |
| 257 | token: str = "", |
| 258 | failure_out: Optional[List[str]] = None, |
| 259 | ) -> List[Dict[str, Any]]: |
| 260 | """ABOUT lane: tweets mentioning each handle, authored by OTHERS. |
| 261 | |
| 262 | Queries ``@handle`` then drops the handle's own tweets (``_is_own``) so only |
| 263 | third-party mentions remain. Returns a flat list of item dicts. |
| 264 | |
| 265 | When ``failure_out`` is provided, any failure reason is appended — the |
| 266 | fatal auth/payment one that stops the lane, and the per-handle transient |
| 267 | ones (429, 5xx, network, unparseable payload) that do not — so the caller |
| 268 | can distinguish a failed request from a handle nobody mentioned. |
| 269 | """ |
| 270 | if not token or not handles: |
| 271 | return [] |
| 272 | items: List[Dict[str, Any]] = [] |
| 273 | seen_ids: set[str] = set() |
| 274 | for raw in handles: |
| 275 | handle = str(raw).lstrip("@").strip() |
| 276 | if not handle: |
| 277 | continue |
| 278 | q = f"@{handle} since:{from_date} until:{to_date}" |
| 279 | got, auth_error = _execute_search( |
| 280 | q, count_per, token, |
| 281 | label=f"@{handle}", id_prefix="XA", |
| 282 | seen_ids=seen_ids, relevance_query=topic, |
| 283 | index_offset=len(items), |
| 284 | failure_out=failure_out, |
| 285 | ) |
| 286 | if auth_error: |
| 287 | if failure_out is not None: |
| 288 | failure_out.append(auth_error) |
| 289 | break |
| 290 | items.extend(it for it in got if not _is_own(it.get("url", ""), handle)) |
| 291 | return items |
| 292 | |
| 293 | |
| 294 | def probe_works(token: str, timeout: int = 8) -> Optional[bool]: |
| 295 | """Cheap runtime check that the xquik key actually returns data. |
| 296 | |
| 297 | Mirrors ``bird_x.probe_works`` for the key-based X path so ``--diagnose`` |
| 298 | reflects reality instead of static key presence. Returns True |
| 299 | (funded/working), False (a clear auth/payment failure — 401/403, or 402 |
| 300 | when the key is configured but unpaid), or None (inconclusive: timeout / |
| 301 | transient HTTP) so callers fail open. |
| 302 | The human-readable reason is available via ``probe_reason()``. Cached per |
| 303 | process so repeated diagnose calls don't re-probe. |
| 304 | """ |
| 305 | global _probe_cache |
| 306 | if _probe_cache[0] != "unset": |
| 307 | return _probe_cache[0] |
| 308 | if not token: |
| 309 | _probe_cache = (False, "no XQUIK_API_KEY configured") |
| 310 | return False |
| 311 | from datetime import timedelta, timezone |
| 312 | since = (datetime.now(timezone.utc) - timedelta(days=30)).strftime("%Y-%m-%d") |
| 313 | # @x (the platform's own account) posts frequently, so a no-error response |
| 314 | # means the key works even if this particular window is quiet. |
| 315 | q = f"from:x since:{since}" |
| 316 | full_url = f"{_BASE_URL}/x/tweets/search?q={_url_encode(q)}&queryType=Top&limit=1" |
| 317 | try: |
| 318 | request_headers = {"X-Api-Key": token} |
| 319 | http.get(full_url, headers=request_headers, timeout=timeout, retries=0) |
| 320 | except http.HTTPError as exc: |
| 321 | status = getattr(exc, "status_code", None) |
| 322 | if status == 402: |
| 323 | _probe_cache = (False, "xquik key unpaid: payment required (402)") |
| 324 | elif status in (401, 403): |
| 325 | _probe_cache = (False, f"xquik auth failed ({status})") |
| 326 | else: |
| 327 | # 5xx / unexpected status — inconclusive, don't report a false-down. |
| 328 | _probe_cache = (None, f"xquik probe inconclusive ({status})") |
| 329 | return _probe_cache[0] |
| 330 | except Exception as exc: |
| 331 | _probe_cache = (None, f"xquik probe inconclusive ({type(exc).__name__})") |
| 332 | return None |
| 333 | _probe_cache = (True, "ok") |
| 334 | return True |
| 335 | |
| 336 | |
| 337 | def probe_reason() -> str: |
| 338 | """Human-readable reason for the last ``probe_works`` result (or '').""" |
| 339 | return _probe_cache[1] |
| 340 | |
| 341 | |
| 342 | def search_and_enrich( |
| 343 | topic: str, |
| 344 | from_date: str, |
| 345 | to_date: str, |
| 346 | depth: str = "default", |
| 347 | token: str = "", |
| 348 | ) -> Dict[str, Any]: |
| 349 | """Search X via Xquik and return results. |
| 350 | |
| 351 | Xquik API returns full engagement data by default, so no separate |
| 352 | enrichment step is needed. |
| 353 | """ |
| 354 | return search_xquik(topic, from_date, to_date, depth=depth, token=token) |
| 355 | |
| 356 | |
| 357 | def parse_xquik_response(response: Dict[str, Any]) -> List[Dict[str, Any]]: |
| 358 | """Extract items from search response. |
| 359 | |
| 360 | Args: |
| 361 | response: Response dict from search_xquik() |
| 362 | |
| 363 | Returns: |
| 364 | List of normalized item dicts. |
| 365 | """ |
| 366 | return response.get("items", []) |
| 367 | |
| 368 | |
| 369 | def _parse_tweet( |
| 370 | tweet: Dict[str, Any], index: int, query: str, id_prefix: str = "XQ" |
| 371 | ) -> Dict[str, Any] | None: |
| 372 | """Parse a single tweet from the API response into the standard item format.""" |
| 373 | author = tweet.get("author") or {} |
| 374 | username = str(author.get("username", "")).lstrip("@") |
| 375 | tweet_id = str(tweet.get("id", "")) |
| 376 | |
| 377 | # Build URL |
| 378 | url = "" |
| 379 | if username and tweet_id: |
| 380 | url = f"https://x.com/{username}/status/{tweet_id}" |
| 381 | if not url: |
| 382 | return None |
| 383 | |
| 384 | # Parse date |
| 385 | date = None |
| 386 | created_at = tweet.get("createdAt") or "" |
| 387 | if created_at: |
| 388 | try: |
| 389 | if len(created_at) > 10 and created_at[10] == "T": |
| 390 | dt = datetime.fromisoformat(created_at.replace("Z", "+00:00")) |
| 391 | else: |
| 392 | dt = datetime.strptime(created_at, "%a %b %d %H:%M:%S %z %Y") |
| 393 | date = dt.strftime("%Y-%m-%d") |
| 394 | except (ValueError, TypeError): |
| 395 | pass |
| 396 | |
| 397 | text = str(tweet.get("text", "")).strip()[:500] |
| 398 | |
| 399 | # Leading-run @mentions = who the post is directed at (reply target). Shared |
| 400 | # parser with bird so the first-party interaction signal fires for xquik too. |
| 401 | from .query import leading_mentions |
| 402 | mentioned_handles = leading_mentions(text) |
| 403 | |
| 404 | # Build engagement dict with full metrics |
| 405 | engagement = { |
| 406 | "likes": _safe_int(tweet.get("likeCount")), |
| 407 | "reposts": _safe_int(tweet.get("retweetCount")), |
| 408 | "replies": _safe_int(tweet.get("replyCount")), |
| 409 | "quotes": _safe_int(tweet.get("quoteCount")), |
| 410 | "views": _safe_int(tweet.get("viewCount")), |
| 411 | "bookmarks": _safe_int(tweet.get("bookmarkCount")), |
| 412 | } |
| 413 | |
| 414 | return { |
| 415 | "id": f"{id_prefix}{index + 1}", |
| 416 | "text": text, |
| 417 | "url": url, |
| 418 | "author_handle": username, |
| 419 | "date": date, |
| 420 | "engagement": engagement, |
| 421 | "mentioned_handles": mentioned_handles, |
| 422 | "relevance": _compute_relevance(query, text) if query else 0.7, |
| 423 | "why_relevant": "", |
| 424 | } |
| 425 | |
| 426 | |
| 427 | def _safe_int(value: Any) -> int | None: |
| 428 | """Convert value to int, returning None on failure.""" |
| 429 | if value is None: |
| 430 | return None |
| 431 | try: |
| 432 | return int(value) |
| 433 | except (ValueError, TypeError): |
| 434 | return None |
| 435 | |
| 436 | |
| 437 | def _url_encode(text: str) -> str: |
| 438 | """URL-encode a string using stdlib.""" |
| 439 | from urllib.parse import quote |
| 440 | return quote(text, safe="") |
| 441 |