A host that serves the tree with Last-Modified and no Cache-Control lets a browser run its cached files for
hours. Measured: a typed-URL revisit after a deploy requested none of the gallery, the manifest or the bundles,
so a returning reader ran the old files, or an old manifest against a new bundle.
- The manifest is fetched with cache 'no-cache', so it is always revalidated (a 304 when unchanged).
- A fingerprint of the manifest (every bundle's integrity) is kept per tree. When it differs from the reader's
last visit, every bundle the browser already holds gets one conditional request before any tab mounts:
a 304 refreshes it, and a 200 is cancelled at its headers, which drops the stale copy. The plain <script>
that follows, the gallery's own or a composite's for its children, then runs the new file at the same URL.
- A version query is not used: loadScriptOnce keys its registry by the full href, so X.js?v= next to a
composite's plain X.js would run the bundle twice.
- A first visit with an empty cache costs nothing extra (only-if-cached finds nothing held), and so does an
ordinary visit with an unchanged manifest. The mount waits at most 8 s for the check.
A preset page loads only poster-sections.js, so a preset of a poster type
registered by another plugin's script (upg100-list) could only say No
posters. The generator now gives no presets to a type whose file is not on
the tree, and check:poster-sections fails on any such preset: 9 presets
remain (7 regions, all regions, configure).
The tree generator writes a page per preset but never deletes one, so the
dropped presets' pages (the 10 upg100 ones and countries-other) are removed,
and a new arm holds the built preset pages to exactly the presets.
A review gallery for the whole-India poster challenge — fitting every
India people group on one printable sheet. It gathers each design
attempt with what was tried and the measured result, and lets a
reviewer label it, score it 1-10, and tag it a best practice or one to
avoid (saved locally, exportable as JSON).
Withheld from every public build: the attempts, their descriptions,
and the poster documents load only under the dev server, in a dev-only
gallery module that compiles away. A production build shows only the
"solution in progress" notice and emits zero whole-country-poster
bytes. Element <india-posters>; stage experimental, visibility team.
The pilot's country data now places every country in a real region, so the
country type has 7 regions. gen:poster-sections-presets regenerated the list
(19 presets); check:poster-sections had failed on the stale countries-other
preset and passes now. Bundle, tree and SRI rebuilt.
The measure-once design missed a host restyled from sized to height:auto after mount WITH in-flow content: it
stayed sized at its content height instead of getting the floor back, as the old probe did.
- After mount, follow() re-probes only when the reading says the host may have changed kind: floor on and the
host shorter than the floor (styled after mount, or capped), or floor off and the host changed size (it may now
follow its content). The re-probe saves every scrolled descendant (through shadow roots) and puts it back
before the frame paints, so no reader state is lost. A window-height change on a content-sized host still
writes nothing.
- A host capped below the floor (max-height) is latched by the probe that finds it, until it reaches the floor
again, so it is never re-probed in a loop. The 1 s post-flip window is gone.
- check:host-floor gains arm H (sized → auto with 420 px of content: the floor returns, at most one flip, the
scrolled child keeps its scrollTop) and counts re-probes in D/E. Poisons: the old per-resize probe, no capped
latch, no re-check of a sized host, no scroll restore.
- library/elements/hostFloor.js holds the decision ProfileLoader uses: probeOnce() at mount, follow() on every
host resize (reads only; at most one flip of the floor).
- A host CAPPED below the floor (max-height) follows its content up to the cap. It is found at mount, or, if
the cap arrives later, by the host shrinking right after a flip; either way it is latched as content-sized
and never flipped again, so it cannot loop. The post-flip check runs before the collapse rule, so a capped
host with empty content is latched rather than flipped back and forth.
- check:host-floor (gates:pre; self-test in gates:post): a stub DOM with a scrolled child that clamps when the
loader collapses. After mount, host resizes never write the loader nor move the child; a sized host fills;
late sizing flips once; a max-height cap (at mount: 0 flips; after mount: at most one flip and one revert)
never loops; a collapsed host gets the floor back; an inline host is left alone. Poisons: the old per-resize
probe, and a missing latch.
Merge studio's poster type registry + pins-in-state + preview-flag pilot into
the region data fix. Built-output conflicts resolved by taking the pilot's
public/ then rebuilding the three bundles that embed country-posters.json
(country-posters, poster-design, regions-dashboard) on the merged source, so
they carry both the registry changes and the corrected regions. Tree + SRI
regenerated. check-poster-section's dropped 'other' scope and the 18 region-only
records are preserved through the merge.
A regression from the host-sized floor: ProfileLoader re-measured its host on EVERY host resize by briefly setting
the loader's min-height to 0 and then to a huge value. On a content-sized host (100svh tall) a window-height
change, as a phone's URL bar makes, resized the host and ran that probe; collapsing and expanding the loader
clamped a scrolling child's scrollTop, and the embedded UPG-100 app, whose view follows its column's scroll,
flipped from Map back to its first view.
- The two-value probe now runs ONCE, at mount, before the profile renders.
- After that the loader only READS sizes and flips one CSS value: a block host shorter than the floor (only a
host sized from outside can be) turns the floor off; a block host that collapsed to nothing turns it back on;
a window-height change on a content-sized host changes nothing.
gen:poster-sections-presets writes poster-sections.instances.json from each
type's scopes() (all regions, then one preset per region titled by the type's
own section title in the site's English, then configure), the same list
check:poster-sections holds the file to. Today it reproduces the 20 presets
exactly, in the data's region order; when the data's regions change, running
it is the whole fix.
Measured headless (390, real touch, a host on a page that scrolls): in the detail view, a drag up from the people
group's floating picture above the sheet scrolled the PAGE (scrollY 0 to 135, en and ar), while the same drag from
the sheet's strip did not. The picture sits outside the sheet and its grab lip, so it had none of their handlers.
It now takes the sheet's capture-phase touch handlers and touch-action:none: a drag from it resizes the sheet (the
grab-zone test holds above the sheet, as for the lip), and a tap is handed back as a click that reaches its
existing @click (open the sheet fully), as before.
scrollTo(paneId) called target.scrollIntoView(), which scrolls every scrollable
ancestor, the embedding page included. On another site, "Map" / "Show on map"
moved the host's own page so that the map's top met the window's top, and a host
shorter than the window then showed its white space under the map. The column
is the scroller (position:absolute; overflow-y:auto) and each pane's
offsetParent, so col.scrollTo({ top: target.offsetTop }) lands the pane exactly
and leaves the page alone.
