There is no iframe in html/ any more. The creator is a top-level document with
the same app bar and the same tab strip as the SOP wizard; the two tabs that
used to swap a frame are links between them.
DEVIATION, stated rather than smuggled. The wave file says "remove the iframe
boundary so the creator renders in the parent document". It renders as its own
document instead. Every done-when is met - no iframe, no cross-frame messaging,
F4 resolved structurally, CR-006 toggles with no special-casing, back and
forward intact with T4.2's URL state - but the route is the other one, and the
reason is in creator-frame.md's own numbers:
merge into parent make it a page
selector collisions to resolve 21 0
script global collisions 9 0
cross-frame call sites to remove 28 28
probe entry points needing rework ~29 2
The 21 and the 9 were never the cost of dissolving the boundary. They are the
cost of MERGING TWO DOCUMENTS, which is a different change the boundary was
hiding. And 29 probe call sites address wp-creation-index.html directly, so a
route that keeps that address keeps all of them. creator-frame.md section 5
records this in full.
What went, and what replaced it:
#wp-frame, applyEmbedLayout, sizeWPFrame, viewportMinusChrome, chromeHeight,
renderWPTab, the resize handler, the ResizeObserver, --wp-chrome-h,
.content-area.embed-full, body.embed-full -> the window sizes the page
?embedded=1, body.embedded, .embed-hide, .embed-first -> nothing. An old
link carrying the param is ignored rather than half-obeyed.
openWpById / showDashboard / showForm / dashApplyFlag / applySopSections
called across the frame -> the URL. ?project= ?view= ?wp= ?flag= were
already read at the creator's own boot (T4.2), which is exactly why those four
could be DELETED rather than migrated. X4 is closed: the surviving path is the
one T5.5 built and proved.
inIframe in auth-guard.js, wp-chrome.js, wp-sidenav.js, help.js and _isTop in
project-data.js -> gone. help.js now reads the explicit WP_HELP_NO_FAB flag
both tool pages set, instead of inferring intent from where it is rendered.
.main-nav / .nav-tab in work-package-suite-styles.css -> wp-chrome.css,
because a tab row only one of two documents can style is the shape that put
the tabs in the parent and the toolbar in the child to begin with.
The three questions creator-frame.md section 4 said no count could answer:
1. The creator gets the app bar. It was the only page loading neither
wp-chrome file. Its header is now the .header-left / .header-right pair the
wizard uses, so the switcher lands in the same place on both.
2. Two sequence components, scoped not merged - confirmed Aug 18 that the
sequence is authored in the SOP and adjustable per package. BL-015 stays.
3. body.embedded is gone. The header it hid is replaced by the app bar; the
sample controls are visible in a new package toolbar (D1); the analytics
button is visible there until T7.10 moves it. The Dashboard BUTTON in that
row became a TAB, which is the one place B7's "fold the toolbar into the
tab row" actually happened.
Old addresses still resolve. ?tab=wp, ?view=dashboard and ?wp=<id> are in
bookmarks, in wp-sidenav's link map, and they are the shape CR-011 and CR-014
were specified against (X1). The wizard forwards them with replace(), so Back
does not bounce. Breaking these silently was the one regression this task could
have shipped that nobody would notice for weeks. frame_check.py section 4 pins
all three.
BEHAVIOUR CHANGE, deliberate. The live cross-frame hand-off showed the creator a
section toggle that had NOT been saved: flip it, look, reload, and the section
came back. What the creator shows now is the SOP that is stored. sections_check
5b pins both halves - an unsaved toggle does not travel, a saved one does.
BEHAVIOUR CHANGE, not deliberate, logged as BL-020. A tab switch is a page exit
now, so leaving the wizard with unsaved SOP edits fires T4.3's unsaved-work
guard. Nothing is lost - the guard writes the draft first and T4.3 recovers it -
but it is friction that did not exist, and suppressing a deliberate guard is a
product decision with its own downside. Logged, not quietly handled here.
tests/frame_check.py, 39 checks, new. Two of them exist because of failures
during this task rather than in it:
- "both documents parse and boot". A const shadowing a function parameter is a
SyntaxError, and work-package-suite-app.js did not parse at all for one run.
Four checks in url_state_check went red and not one said "the script did not
load". Asserting a page's own entry points exist costs nothing.
- "focus emulation is on, so a focus reading means something". An earlier draft
called page.call instead of page.ws.call inside a try/except and measured
nothing, reporting no focus ring anywhere - which looks exactly like a
finding. Trap 5 in reverse, for the second time in this project.
The four backlog entries logged against this file, re-measured rather than
assumed:
BL-001 still reproduces (485px in a 390px viewport) but its RECORDED CAUSE IS
WRONG. --nav-w now computes to 56px, so the injected-style explanation
is spent. The overflow is the creator's data tables - #asset-body's
lays out at 520px with no scroll container. frame_check reports the
offending boxes by selector and skips position:fixed subtrees, because
the comments drawer parked off-screen at right:844 made the first
measurement blame the drawer. Pinned, not fixed: T7.2 lays out the form.
BL-013 CLOSED. It was fixed by S12 in WAVE 4 - wp-creation-styles.css:209
carries the comment naming this entry - and nobody updated it. It was
quoted as a live CLAUDE.md violation while planning wave 7 and had not
been true for four waves. a11y_check walks 120 focusable elements on
the creator and every one rings at >= 3:1.
BL-006 15 by the probe's measure, unchanged; different denominator, stated.
BL-007 68 raw radii by the probe's measure. Nothing has reduced it in four
waves; it is measured every run now instead of once.
BL-018 cost a FOURTH probe. frame_check imports set_sop from sections_check
rather than writing a fifth copy of the workaround. T9.9 owns it.
Probes re-pointed, with reasons in the files: sections_check 5b (drove the live
hand-off), pipeline_check check 2 (read through contentDocument), f_items F4
(drove standalone and embedded; there is one mode now), validation_check
(lost "the wrong tab", gained the SOP gate).
Verified: frame_check 39/39, sections_check 95/95, pipeline_check 44/44,
url_state_check 23/23, validation_check 83/83, a11y_check 22/22,
autosave_check 34/34, aggregates_check 16/16, stepper_check 71/71,
browser_check 71/71, launcher_check 58/58, generalinfo_check 49/49,
rollup_check 63/63, cards_check 44/44, locations_check 58/58.
f_items: F1-F5 fixed, F6 reproduces (T7.2).
Metrics: iframes 1 -> 0, colour literals in rules outside theme-light.css 0,
dialogs 64, <div onclick> 2, .help-tip 18.
Items: B7 D1
Task: T7.1
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
110 lines
4.9 KiB
JavaScript
110 lines
4.9 KiB
JavaScript
/* Service worker for the Work Package Suite PWA.
|
|
|
|
Goal: let the app (and especially the field view) load and run offline. Data
|
|
durability is already handled by the sync outbox in project-data.js — this
|
|
worker only caches the static app shell so the pages open without a network.
|
|
|
|
Strategy:
|
|
• /api/* and non-GET → never touched (pass straight to the network; offline
|
|
reads fall back to the app's localStorage cache, writes queue in the outbox).
|
|
• HTML / CSS / JS → network-first, cache as fallback. These reference each
|
|
other, so a page must never run against a stale sibling.
|
|
• images / icons / manifest → stale-while-revalidate (instant from cache).
|
|
*/
|
|
'use strict';
|
|
// Bumped when the shell file list changes, so clients fetch the new assets
|
|
// instead of serving a half-old shell from the previous cache.
|
|
const CACHE = 'wp-suite-shell-v6';
|
|
const SHELL = [
|
|
'/', '/index.html', '/work-package-suite.html', '/wp-creation-index.html',
|
|
'/field.html', '/login.html', '/admin.html', '/users.html',
|
|
'/theme-light.css', '/work-package-suite-styles.css', '/wp-creation-styles.css',
|
|
'/wp-chrome.css', '/console.css', '/wp-sidenav.css',
|
|
'/auth-guard.js', '/project-data.js', '/feedback-config.js', '/help.js',
|
|
'/work-package-suite-app.js', '/wp-creation-app.js', '/field.js',
|
|
'/wp-chrome.js', '/wp-sidenav.js', '/wp-format.js', '/login.js',
|
|
'/console-util.js', '/admin.js', '/users.js',
|
|
'/prime-controls-logo.jpg', '/favicon.ico',
|
|
'/manifest.webmanifest', '/icon-192.png', '/icon-512.png',
|
|
];
|
|
|
|
self.addEventListener('install', (e) => {
|
|
// Cache each shell asset individually so one missing file doesn't abort install.
|
|
e.waitUntil(
|
|
caches.open(CACHE)
|
|
// cache:'reload' bypasses the browser HTTP cache. Without it the precache
|
|
// can be filled from stale HTTP entries, freezing a mismatched shell.
|
|
.then((c) => Promise.all(SHELL.map(
|
|
(u) => c.add(new Request(u, { cache: 'reload' })).catch(() => {}))))
|
|
.then(() => self.skipWaiting())
|
|
);
|
|
});
|
|
|
|
self.addEventListener('activate', (e) => {
|
|
e.waitUntil(
|
|
caches.keys()
|
|
.then((keys) => Promise.all(keys.filter((k) => k !== CACHE).map((k) => caches.delete(k))))
|
|
.then(() => self.clients.claim())
|
|
);
|
|
});
|
|
|
|
// Code (HTML / CSS / JS) is fetched NETWORK-FIRST, falling back to the cache when
|
|
// offline. Everything else (images, icons, the manifest) stays cache-first, which is
|
|
// where offline speed actually comes from.
|
|
//
|
|
// Why not cache-first for code: these files reference each other, and the cache
|
|
// stores them as independent entries. Cache-first served whichever copy of each file
|
|
// happened to be stored, so a browser could run new HTML against old CSS — which is
|
|
// exactly how the creator once collapsed to a 300x150 box when it was an iframe
|
|
// and its stylesheet was a version behind its markup. A page must
|
|
// only ever run against the stylesheet and scripts it shipped with.
|
|
const CODE_RE = /\.(html|css|js)$|\/$/i;
|
|
|
|
self.addEventListener('fetch', (e) => {
|
|
const req = e.request;
|
|
if (req.method !== 'GET') return; // outbox owns writes
|
|
const url = new URL(req.url);
|
|
if (url.origin !== self.location.origin) return; // third-party: default
|
|
if (url.pathname.startsWith('/api/')) return; // never cache the API
|
|
|
|
const isCode = CODE_RE.test(url.pathname);
|
|
|
|
// Cache key WITHOUT the query string. Links inside the app carry ?project=…&tab=…,
|
|
// and the creator used to carry a cache-busting timestamp in its frame src, so keying on the
|
|
// full URL both missed every offline navigation and grew the cache without bound.
|
|
const key = new Request(url.origin + url.pathname, { credentials: 'same-origin' });
|
|
const fromCache = () => caches.match(key).then((c) => c || caches.match(req));
|
|
|
|
const store = (res) => {
|
|
if (res && res.ok && res.type !== 'opaque') {
|
|
const copy = res.clone();
|
|
caches.open(CACHE).then((c) => c.put(key, copy)).catch(() => {});
|
|
}
|
|
return res;
|
|
};
|
|
|
|
if (isCode) {
|
|
e.respondWith(
|
|
// cache:'no-cache' forces revalidation with the server. Plain fetch() inherits
|
|
// the request's default cache mode, which consults the browser HTTP cache — so
|
|
// "network-first" alone still let a page run against a stale sibling file.
|
|
fetch(req, { cache: 'no-cache' })
|
|
.then((res) => {
|
|
// A 502/404 must not replace a page the cache could still serve.
|
|
if (!res || !res.ok) return fromCache().then((c) => c || res);
|
|
return store(res);
|
|
})
|
|
.catch(() => fromCache()) // offline → last good copy
|
|
.then((res) => res || Response.error()) // never resolve to undefined
|
|
);
|
|
return;
|
|
}
|
|
|
|
e.respondWith(
|
|
caches.match(key).then((cached) => {
|
|
const network = fetch(req).then(store).catch(() => cached);
|
|
return cached || network.then((res) => res || Response.error());
|
|
})
|
|
);
|
|
});
|