Model/scripts/hyde/elmhurst_lib.py
Jun-te Kim 2f0eb49eee Checkpoint: UPRN 10093116543 Elmhurst build + devcontainer VNC/Playwright + perms
- Add SAP-accuracy sample for uprn_10093116543 (epc.json, elmhurst_inputs.md,
  summary/worksheet PDFs)
- Persist hyde viewer stack (xvfb/fluxbox/x11vnc/novnc/websockify) and Playwright
  chromium in the backend devcontainer; forward noVNC 6080
- Broaden .claude/settings.local.json allowlist (display/python/grep/tail)
- In-progress campaign mapper/cert_to_inputs work carried from prior cert

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-06-16 15:21:56 +00:00

228 lines
9.5 KiB
Python

"""Reusable Elmhurst RdSAP-10-Online automation helpers (campaign edition).
Why this exists: the per-cert build in the `expand-sap-accuracy-corpus` loop is
the slow, flaky part. `elmhurst_fill.py` hard-codes one cert's values and relies
on a *persisted* login session — but Elmhurst's auth is a **session cookie** that
Playwright's persistent context does NOT reliably restore across process launches.
This module sidesteps both problems:
* **Log in fresh each run** (creds from env `ELM_ACCESS`/`ELM_PWD`, else a
gitignored `scripts/hyde/.elmhurst-creds.json` {"access","pwd"}). Assessment
DATA persists server-side via *Save & Close*, so re-login per run is fine and
you can build/verify one page at a time across separate processes.
* **Headed on the noVNC display** (`DISPLAY=:99`; run `start_viewer.sh` first).
Hard-won Elmhurst-DOM gotchas baked into the helpers below (see SKILL.md too):
* Most controls AutoPostBack — set one field, wait for the postback, re-locate.
* Sub-tabs (Doors, Air Pressure Test, Lighting, Water-Heating FGHRS…) are
AjaxControlToolkit tabs: their fields are in the DOM but NOT visible until the
tab span `#__tab_<panelId>` is clicked. Use `click_tab()`.
* Delete is a Yes/No MODAL (not a native dialog): click the row delete, then
`#..._LinkButtonYes`. Its ModalBackground intercepts clicks — use JS .click().
* The window grid will NOT delete the LAST row, and **each Add wipes the
previously-added row's editable cells** — so multi-row window entry via the
add-form does not stick. Enter ONE combined window row (total glazed area,
one orientation); per worked-ref guidance that's an accepted approximation.
* Window glazing: "...known data" enables per-window U/g but then DEMANDS
Frame Factor + Data-Source + Location + Building Part per row (Recommendations
validation fails otherwise). Prefer a standard band, e.g.
"Double post or during 2022" (auto-derives U≈1.4/g≈0.72), and still set
Orientation + Location="External wall" + Building Part="Main".
* The "U-value known" override checkboxes (walls/roof/floor) don't reliably
toggle via automation — accept the natural RdSAP age-band U-value.
* The **Recommendations** page is the validation gate; it MUST show zero errors
before the Energy Report Summary / worksheet PDFs are downloadable.
Common required-but-blank fields: insulated-door U-value, air-pressure-test
certificate number (when method=Blower Door).
* Truth only after Save & Close + reload: the live (unsaved) grid mis-displays
older rows as 0.00 — never delete a row by its *live* width; reload first.
Usage:
import elmhurst_lib as E
with E.session() as (ctx, page):
E.goto(page, "Dimensions", "WebFormDimensions.aspx")
E.set_text(page, E.FP + "...TextBoxFloorAreaLowestFloor", "44.93")
E.save_close(page)
"""
from __future__ import annotations
import json
import os
from contextlib import contextmanager
from pathlib import Path
from typing import Callable, Generator, Optional
from playwright.sync_api import (
BrowserContext,
Page,
TimeoutError as PlaywrightTimeoutError,
sync_playwright,
)
HERE = Path(__file__).parent
SESSION_DIR = HERE / ".elmhurst-session"
CREDS_FILE = HERE / ".elmhurst-creds.json" # gitignored; {"access":..,"pwd":..}
# The single reusable campaign assessment ("Khalim-test"). Overwrite it per UPRN.
ASSESSMENT_GUID = "B44A0DB4-4C08-4241-B818-86F060172105"
ENTRY_URL = (
"https://rdsap10online.elmhurstenergy.co.uk/Processing/WebFormAddress.aspx"
f"?Guid={ASSESSMENT_GUID}"
)
FP = "ContentBody_ContentPlaceHolder1_" # control-id prefix
NAV = "ContentBody_buttonAction" # left-rail section nav id prefix
SETTLE_MS = 700
def _creds() -> tuple[str, str]:
if CREDS_FILE.exists():
d = json.loads(CREDS_FILE.read_text())
return d["access"], d["pwd"]
access = os.environ.get("ELM_ACCESS")
pwd = os.environ.get("ELM_PWD")
if not access or not pwd:
raise SystemExit(
"No Elmhurst creds: set ELM_ACCESS/ELM_PWD or write "
f"{CREDS_FILE} = {{'access':..,'pwd':..}}"
)
return access, pwd
@contextmanager
def session() -> Generator[tuple[BrowserContext, Page], None, None]:
"""Launch headed on :99, log in fresh, land on the assessment. Closes on exit."""
with sync_playwright() as p:
ctx = p.chromium.launch_persistent_context(
user_data_dir=str(SESSION_DIR),
headless=False,
accept_downloads=True,
viewport={"width": 1400, "height": 1000},
)
page = ctx.pages[0] if ctx.pages else ctx.new_page()
page.on("dialog", lambda d: d.accept())
page.goto(ENTRY_URL, wait_until="networkidle", timeout=60_000)
if "WebFormLogin" in page.url:
access, pwd = _creds()
page.fill("#ctl00_ctl00_ContentBody_ContentBody_TextBoxAccessCode", access)
page.fill("#ctl00_ctl00_ContentBody_ContentBody_TextBoxPassword", pwd)
with page.expect_navigation(wait_until="networkidle", timeout=60_000):
page.click("#ctl00_ctl00_ContentBody_ContentBody_ImageButtonEnter")
page.wait_for_timeout(1000)
try:
yield ctx, page
finally:
ctx.close()
def commit(page: Page, action: Callable[[], object]) -> None:
"""Run a control change that triggers an ASP.NET postback; wait for it."""
try:
with page.expect_navigation(wait_until="load", timeout=8_000):
action()
except PlaywrightTimeoutError:
page.wait_for_load_state("networkidle", timeout=8_000)
page.wait_for_timeout(SETTLE_MS)
def goto(page: Page, nav: str, url_fragment: str) -> None:
"""Navigate to a left-rail section (no-op if already on it — re-clicking
double-postbacks). Navigating saves the current page into the server session."""
if url_fragment in page.url:
return
with page.expect_navigation(wait_until="networkidle", timeout=60_000):
page.click(f"#{NAV}{nav}_Link", timeout=15_000)
page.wait_for_timeout(SETTLE_MS)
def click_tab(page: Page, panel_id: str) -> None:
"""Activate an AjaxControlToolkit sub-tab by its panel id, e.g.
'TabContainer_TabPanelDoorsPanel'. Client-side; no postback."""
page.evaluate(
"(id)=>{const e=document.getElementById('__tab_'+id); if(e)e.click();}",
f"{FP}{panel_id}",
)
page.wait_for_timeout(500)
def set_text(page: Page, suffix: str, value: str, autopostback: bool = True) -> None:
loc = page.locator(f"#{FP}{suffix}")
loc.fill(value)
if autopostback:
commit(page, lambda: loc.press("Tab"))
else:
page.wait_for_timeout(150)
def set_select(page: Page, suffix: str, value: str, autopostback: bool = True) -> None:
loc = page.locator(f"#{FP}{suffix}")
if autopostback:
commit(page, lambda: loc.select_option(value))
else:
loc.select_option(value)
def save_close(page: Page) -> None:
"""Save & Close — commits the whole server session to the DB (so the data
survives the next fresh-login run). The body button id-suffix avoids the
top-bar LinkButton copy. NEVER touches Submit."""
with page.expect_navigation(wait_until="networkidle", timeout=60_000):
page.click("#ContentBody_ButtonSaveClose", timeout=15_000)
page.wait_for_timeout(SETTLE_MS)
# --- Window grid helpers (Openings page) ---------------------------------
_WP = "TabContainer_TabPanelWindowsPanel_"
_GRID = "GridViewExtendedWidows"
def window_row_count(page: Page) -> int:
return page.evaluate(
f"""()=>document.querySelectorAll("[id*='{_GRID}_DeleteButton_']").length"""
)
def _jsclick(page: Page, element_id: str) -> None:
page.evaluate("(id)=>{const e=document.getElementById(id); if(e)e.click();}", element_id)
def add_combined_window(
page: Page, total_area_m2: float, orientation: str,
glazing: str = "Double post or during 2022",
) -> None:
"""Add ONE window row of `total_area_m2` (width=area, height=1.0). Sets the
required per-row fields (orientation/location/building-part) so the
Recommendations validation passes. Caller then deletes the prior rows."""
set_select(page, f"{_WP}DropDownListExtGlazing", glazing)
page.locator(f"#{FP}{_WP}TextBoxExtWidth").fill(str(total_area_m2))
h = page.locator(f"#{FP}{_WP}TextBoxExtHeight")
h.fill("1.00")
commit(page, lambda: h.press("Tab"))
page.locator(f"#{FP}{_WP}DropDownListExtOrientation").select_option(orientation)
page.locator(f"#{FP}{_WP}DropDownListExtBuildingPartId").select_option("Main")
page.locator(f"#{FP}{_WP}DropDownListExtLocation").select_option("External wall")
page.wait_for_timeout(200)
before = window_row_count(page)
_jsclick(page, f"{FP}{_WP}ButtonAddWindow")
for _ in range(20):
page.wait_for_timeout(200)
if window_row_count(page) > before:
break
def delete_first_window(page: Page) -> None:
"""Delete the first window row via the Yes/No modal. Will refuse to drop the
LAST row (Elmhurst blocks it) — so add the replacement BEFORE deleting."""
before = window_row_count(page)
first = page.evaluate(
f"""()=>{{const e=document.querySelector("[id*='{_GRID}_DeleteButton_']"); return e?e.id:null;}}"""
)
if not first:
return
_jsclick(page, first)
page.wait_for_selector(f"#{FP}DeleteWindowDialog_LinkButtonYes", state="visible", timeout=5_000)
_jsclick(page, f"{FP}DeleteWindowDialog_LinkButtonYes")
for _ in range(20):
page.wait_for_timeout(200)
if window_row_count(page) < before:
break