Files
iptvnator/libs
4grayandClaude Opus 5.5 d57da4dfec perf(web): take the Material snack-bar off the startup path (#1939)
* perf(web): take the Material snack-bar off the startup path

The seven services and the app shell created with the app injected
MatSnackBar, and snack-bar.mjs statically imports Material's button, the
icon button chunk, ripple and the CDK a11y module, so about 110 KB sat in
the initial bundle although a toast shows only after an event.

A root LazySnackBar (@iptvnator/services/lazy-snack-bar) resolves the
snack-bar from a dynamic import the way LazyMatDialog does, retrying after
a failed chunk load, and opens text toasts with MatSnackBar.open's
arguments; a failed load is logged without the message. The eager
importers open their toasts through it with unchanged messages, actions,
durations and panel classes; lazy code keeps injecting MatSnackBar. The
startup status defers its Retry button, the other eager importer of the
button chunk, with a native button as the error fallback.

Local production build: renderer.initialBytes 1,349,048 -> 1,236,918
(-112,130, -8.3%); no Material module is in the initial bundle any more.
The CDK overlay, scrolling and portal chunks (about 78 KB) stay through
the FullscreenOverlayContainer provider in app.config.ts.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

* fix(web): keep a native Retry while the startup button chunk loads

Greptile (local round 1): the failure screen had no Retry button until
the deferred Material button chunk arrived. One native button, drawn in
the flat button's shape with theme colours, is now the placeholder and
the error view of the defer block, so a retry is possible while the
chunk downloads and after a failed load. Baseline and docs follow the
re-measured value, 1,237,440 bytes.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

* docs(perf): point the initialBytes evidence at the opened pull request

The baseline and the performance doc named the predicted number; the
pull request opened as #1939.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

* test(web): cover the toast actions behind the lazy snack-bar

Greptile (PR round 1): the Refresh and Trust-host actions now run after
LazySnackBar.open resolves, and no test exercised them. The PWA update
toast test holds the toast promise, then resolves it and picks Refresh;
the page reload is a field the spec substitutes, as jsdom's location
cannot be spied on. The Electron test does the same for the TLS trust
toast: the confirmation opens only once the toast is up, approving it
records the host and retries the fetch.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

* test(web): type the stubbed toast ref as Observable<undefined>

CI's spec typecheck rejected `typeof EMPTY` (Observable<never>) for a
stub that returns `of(undefined)`.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

* perf(web): lower the initialBytes baseline to the merged measurement

Merged with #1938 the branch measures 1,231,455 bytes on a local
production build (the radio play control's fab tokens left the
stylesheet); the baseline follows it.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

* perf(web): lower the initialBytes baseline to the zoneless merged measurement

Merged with #1929, #1936 and #1938 the branch measures 1,178,109 bytes
on a local production build; the baseline follows it.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

* perf(web): re-measure initialBytes after merging #1903 and #1942

Merged with #1903 (self-hosted server storage, which raised master's
baseline by 7,245 bytes on purpose) and #1942 the branch measures
1,187,983 bytes on a local production build, below master's 1,285,523;
no Material module is in the initial bundle. The new favorites effect
spec provides the LazySnackBar stub the merged effect injects.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* perf(xtream): open favorites toasts through LazySnackBar

#1934 gave the Xtream store's favorites feature a static MatSnackBar
injection. The store exists from startup, so that one import put
Material's snack-bar, button, icon-button and ripple chunks (about
95 KB) back on the renderer's initial path. The feature now opens its
two toasts through LazySnackBar with the same messages and durations.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* fix(perf): count statically imported chunks in renderer.initialBytes

Angular's builder emits at most ten modulepreload hints. Once the
startup graph passed ten chunks, the rest were fetched through
main.js's own imports but never counted, so the counter stopped growing
and an eager MatSnackBar import on master went unnoticed. The script now
follows the static import graph of every counted script (dynamic
import() targets stay lazy; a missing imported file is an error), which
matches the bundler's metafile closure exactly: 1,196,205 bytes for this
branch, where the old counter read 1,098,535.

The baseline follows the new measurement and its note now explains the
current limit (Greptile).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* fix(perf): parse startup imports and count them per request URL

Codex (PR round): the regexes also matched import-shaped text in strings,
templates and comments, which could count a lazy chunk or fail on a file
that does not exist. Static declarations are now read from TypeScript's
syntax tree (top-level import/export-from statements only). Imports are
also deduplicated by request URL, like the index.html references, so a
preload of chunk.js?v=1 no longer hides an import of ./chunk.js. The
real build still measures 1,196,205 bytes.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
2026-10-15 22:59:31 +02:00
..