max / quasi

0.42.0: the webview emitter writes htmx 4 `2246072d`. Max ruled it on 2026-08-18: the webview emitter targets htmx 4, not 2.x. quasi is pre-release, so writing the transport against a model that is being replaced means writing it twice, and 4.0 having no announced stable date is a bet taken deliberately rather than a thing to wait out. Wiki `quasi-overview`, "htmx 4, ruled 2026-08-18". Every change was checked against 4.0.0-beta6's own source rather than against the migration guide, which is also what the template now vendors. The swap is `outerMorph`, and it is unconditional. Under 2.x morphing came from idiomorph, which is why the shell carried a `morph_src`, registered `hx-ext="morph"` on the body and the emitter asked for a morph only when the extension was loaded: htmx fell back to a destructive `innerHTML` in silence when it was missing, and silently destructive is the one outcome decision 7 exists to avoid. In 4 there is nothing that can be missing, so `morph_src`, `without_morph()`, `morphs()` and the `morphs: bool` threaded through sixty call sites are all gone. `outerMorph` and not `innerMorph` because the extension's bare `morph` meant outerHTML. The lock is `hx-disable`, 4's name for 2.x's `hx-disabled-elt`. Worth saying out loud because the rename is a swap of meaning: what 2.x called `hx-disable` is `hx-ignore` in 4, so a find-and-replace in either direction is wrong. Nothing here emits `hx-ignore`. Inheritance is explicit, so the headers a document sends go on the body as `hx-headers:inherited`. A bare `hx-headers` there would be a header the body sends on its own requests and no control sends on any, which is the whole reason they are on the body. An emitter knows every attribute it writes, so this costs one attribute name and no design. `quasi_http::htmx::CONFIG_META` and `RESPONSE_HANDLING` are deleted rather than updated. 4 swaps every status but 204 and 304, and `responseHandling` is not a config key any more. That constant existed because 2.x dropped a 4xx, which made decision 9's classified errors (403 `Denied`, 404 `NotFound`) render nothing at all; the emitter gets it for free now, and a screen wanting per-status behaviour says so on the element with `hx-status:XXX`. The two scripts this crate ships listen on `htmx:after:settle`, 4's spelling of the settle event. The scaffolder vendors htmx 4.0.0-beta6 and no longer vendors idiomorph. A 2.x bundle behind `Shell::htmx_src` is a page whose controls fetch nothing and whose swaps destroy state, so the bundle and this crate move together. One behaviour difference is documented rather than designed around: on a GET, htmx 4 deletes an address parameter whose name an `hx-vals` value repeats before appending it, where 2.x appended both. Nothing in the tree names one thing in both bags on a read. Consumers forward-fixed to 0.42: MNW server (2282 lib tests green), goingson under `--features quasi` (677 green), audiofiles. MNW is now in the state the ruling sequences for: its described screens emit 4's markup while `static/htmx.min.js` is still 2.0.4, and the site's own migration is mnw-server `6b87e5ff`. Exposure is bounded by `QUASI_SCREENS`, which is off unless a host sets it. Workspace: 23 test binaries green, clippy clean but for the pre-existing `large_enum_variant` on `Outcome`.