Command

build

Description

#27581 limited modulepreload hints to a shallow set of 10 initial chunks, as a safe default for all applications (#27490). For an application with a deep initial import graph that default costs one network round trip per depth level: the browser only discovers the chunks at depth N after the chunks at depth N-1 have arrived and been parsed.

Our main has a static closure of 45 chunks across six depth levels. We post-process index.html after ng build to preload all of them. Measured in production RUM over three weeks (roughly a million initial loads per week), weekdays before versus after:

The slowest connections gained the most, which is the opposite of what the HTTP Archive correlation cited in #27490 would predict. That correlation covers preloads of any resource. Chunks in the entry point's static closure are all required before the application can execute at all, so preloading them does not compete with anything more critical than the stylesheet.

Describe the solution you'd like

Let index.preloadInitial accept a number or "all" in addition to the current boolean, keeping true as today's shallow-10 behaviour. That would let us delete a post-build script that has to track the CLI's index.html output, and which broke on the integrity attributes added in 22.x.

Describe alternatives you've considered

Post-processing index.html (what we do now); setting preloadInitial: false and writing every hint ourselves (same fragility); eager-loading the landing route into the main bundle (adds 330 KB gzipped to every first load for the 70% of visits that land elsewhere).