gimme-job/gimme_job/adapters/tribalhealth.py

"""Tribal Health job board adapter — Elementor WordPress loop, numbered-page pagination.

Tribal Health (tribalhealth.com) is a clinician staffing agency that places providers
at tribal / IHS sites. Its job board renders an Elementor loop of EVERY open job
(heavily PT/OT/RN); it exposes no usable keyword search, so we collect every card and
keyword-filter to dentist/orthodontist roles here (mirrors the nativehealth pattern).

The dental roles are sparse and never guaranteed to land on page 1, so collect_cards
walks ALL board pages itself rather than relying on the orchestrator's page loop —
which stops paginating the moment a page yields zero (post-filter) cards. paginate()
therefore always returns False.
"""
from __future__ import annotations

from loguru import logger

from gimme_job.adapters.base import ManifestDrivenAdapter
from gimme_job.adapters.registry import register
from gimme_job.models.dto import JobPostingCandidate, RawJobCard

_DEFAULT_KEYWORDS = ["orthodontist", "orthodontic", "dentist"]
_COMPANY = "Tribal Health"
_BOARD_URL = "https://tribalhealth.com/job-board/"
_MAX_SCROLLS = 4  # Elementor loop lazy-renders cards as you scroll

_CARD_JS = """() => {
    const cards = Array.from(document.querySelectorAll('.e-loop-item'));
    return cards.map(c => {
        const cls = c.className || '';
        const a = c.querySelector("a[href*='/jobs/']");
        const h = c.querySelector('.elementor-heading-title');
        // Term list items render as [specialty, "City, State"].
        const terms = Array.from(
            c.querySelectorAll('.elementor-post-info__terms-list-item')
        ).map(t => t.textContent.trim()).filter(Boolean);
        const location = terms.find(t => t.includes(',')) || terms[terms.length - 1] || null;
        const postId = (cls.match(/post-(\\d+)/) || [])[1] || null;
        return {
            title: h ? h.textContent.trim() : (a ? a.textContent.trim() : null),
            location: location,
            url: a ? a.href : null,
            external_job_id: postId,
        };
    }).filter(j => j.title && j.url);
}"""


@register("tribalhealth")
class TribalHealthAdapter(ManifestDrivenAdapter):

    def _keywords(self) -> list[str]:
        kws = self.manifest.post_filters.include_title_keywords or _DEFAULT_KEYWORDS
        return [k.lower() for k in kws]

    def _scrape_current_page(self, page) -> list[dict]:
        """Wait for cards, scroll to force lazy render, return raw card dicts."""
        try:
            page.wait_for_selector(".e-loop-item", timeout=12000, state="attached")
        except Exception:
            return []
        for _ in range(_MAX_SCROLLS):
            try:
                page.mouse.wheel(0, 4000)
                page.wait_for_timeout(400)
            except Exception:
                break
        try:
            return page.evaluate(_CARD_JS) or []
        except Exception as e:
            logger.debug(f"[tribalhealth] card evaluate failed: {e}")
            return []

    def collect_cards(self, page, manifest) -> list[RawJobCard]:
        """Walk every board page, gather all cards, then keyword-filter by title."""
        max_pages = max(1, manifest.pagination.max_pages)
        raw_data: list[dict] = []
        seen_urls: set[str] = set()

        for page_num in range(1, max_pages + 1):
            if page_num > 1:
                # Page 1 was already navigated to by the orchestrator.
                try:
                    page.goto(f"{_BOARD_URL}{page_num}/", timeout=30000, wait_until="domcontentloaded")
                except Exception:
                    break
            page_raw = self._scrape_current_page(page)
            if not page_raw:
                break  # empty page = end of board
            new_on_page = 0
            for item in page_raw:
                if item["url"] in seen_urls:
                    continue
                seen_urls.add(item["url"])
                raw_data.append(item)
                new_on_page += 1
            logger.debug(f"[tribalhealth] page {page_num}: {new_on_page} new raw cards")
            if new_on_page == 0:
                break  # pagination wrapped around / no new cards

        keywords = self._keywords()
        cards: list[RawJobCard] = []
        for item in raw_data:
            title = item["title"]
            if not any(kw in title.lower() for kw in keywords):
                continue
            cards.append(RawJobCard(
                title=title,
                company=_COMPANY,
                location=item["location"] or None,
                url=item["url"],
                external_job_id=item["external_job_id"],
            ))

        logger.info(
            f"[tribalhealth] Keyword filter: {len(cards)}/{len(raw_data)} cards matched "
            f"across {min(page_num, max_pages)} page(s)"
        )
        return cards

    def paginate(self, page, page_index: int, config) -> bool:
        """All pages are walked inside collect_cards — no orchestrator-level paging."""
        return False

    def normalize(self, raw: RawJobCard) -> JobPostingCandidate:
        return super().normalize(raw)