Skip to content

ROADMAP.md

Latest commit

 

History

History
43 lines (36 loc) · 5.68 KB

ROADMAP.md

File metadata and controls

43 lines (36 loc) · 5.68 KB
Raw

Roadmap

Working list of where SvelteNet is heading. Items move to Shipped when they land — keep this file honest: if scope changes or something ships, update it in the same change.

Next

  • Publish packages — publish SvelteNet.Core, SvelteNet.AspNetCore, SvelteNet.Jint, and SvelteNet.FluentValidation to NuGet, plus sveltenet to npm. Packaging is complete (license metadata, Source Link, symbols, multi-targeted net8.0;net10.0, analyzer and build tool resolved from real project outputs, all four packed in CI); remaining work is versioning and the release flow.

Later

  • Blazor host — render islands from Blazor components. SvelteNet.Core stays host-agnostic specifically for this; new transports are minimal-API endpoints, not MVC filters.
  • Blazor SSR with enhanced navigation — the real prize. Enhanced nav does NOT re-execute inline <script> tags on navigation, so the current inline-mount emission won't hydrate islands after a nav; needs an island bootstrapper (external module / custom element that discovers and mounts islands on DOM changes).
  • Scaffolder page-directory mapping — currently assumes the *.Pages.* namespace convention; should fall back gracefully for custom layouts. Related: SveltePage.SvelteComponent derives the path from ViewEnginePath at runtime while the scaffolder derives it from the namespace, so the two can disagree.
  • Component-name safety — a SvelteComponents constants generator + analyzers so string component references (@await Html.SvelteAsync("Components/Card", ...)) are checked at compile time.
  • Persistent SSR worker for Node.js/Bun — those renderers spawn a process per render, which costs tens of milliseconds per island. A warm worker pool would make them competitive with Jint on latency without Jint's shared-module-state caveat.
  • Antiforgery for remote functions — commands and forms are defended by the X-SvelteNet header plus an Origin check. An opt-in token requirement would add defence in depth for apps that want it.
  • Cancellation for queries — RemoteQuery has no AbortSignal, so navigating away leaves requests in flight. Refresh races are already handled by generation counters.
  • No-JS form validation returns problem JSON — a [Form] post from a browser without JavaScript renders the raw application/problem+json document instead of redirecting back with the errors. Fixing it properly needs the errors carried across the redirect (TempData or equivalent), which is a feature rather than a patch. Enhanced submits are unaffected.
  • Metadata generator holds Roslyn symbols in the incremental pipeline — SvelteMetadataGenerator collects INamedTypeSymbols, which roots the Compilation and defeats incremental caching, so a large solution re-runs full semantic analysis on every keystroke in the IDE. Needs projecting to equatable value models inside the transform.

Shipped

  • Svelte 5 / ASP.NET rewrite: typed [SvelteProp] props and generated TypeScript
  • Opt-in SSR builder: client-only by default, pooled in-process Jint (SvelteNet.Jint), bundled Node.js/Bun CLI renderers, and custom renderer hooks
  • One-command Vite build (client + SSR via the Environments API), dev mode with HMR
  • SvelteKit-style forms with enhance(), including automatic antiforgery and double-submit protection
  • [SvelteComponent] typed MVC rendering
  • Remote functions ([Query]/[Command]/[Form]) with generated typed client and source-generated, descriptor-based dispatch and registration (no runtime reflection)
  • Experimental async: awaited queries in components, SSR via the in-process fetch bridge, hydratable head stash with zero-refetch hydration, reactive-thenable refresh semantics
  • Build-time TypeScript generation (SvelteNet.Build MSBuild target), incremental so repeat builds do no work
  • SvelteKit-style .svelte-net/types declaration tree and private .svelte-net/server output
  • Project-wide Svelte discovery with Razor/MVC/vertical-slice colocation and one generated remote class per C# service
  • PR CI covering .NET tests/format/pack on Linux and Windows, JS Node + browser tests, sample typechecks/builds, and docs
  • Samples split by concept (TodoApp, RemoteFunctions, MvcHello), all integration-tested
  • VitePress docs site deployed to GitHub Pages; Dependabot across actions/bun/NuGet
  • Validation errors as RFC 9457 problem details (application/problem+json, ASP.NET errors member) across remote functions, the SSR bridge, and enhanced page posts
  • BYOV validation pipeline: ISvelteRemoteValidators run between binding and invocation; DataAnnotations on parameters and complex argument types validate automatically
  • SvelteNet.FluentValidation: registered IValidator<T>s (FluentValidation, still free/Apache 2.0 — worth sponsoring) run automatically via the BYOV pipeline
  • Asynchronous SSR contract — ISvelteSsrEngine.RenderAsync and ISvelteSsrFetchHandler.HandleAsync. A server render can await remote queries; the previous synchronous contract blocked a request thread for the whole await chain. Jint, which cannot suspend a running script, renders on a worker capped by MaxConcurrentRenders
  • Descriptor-level authorization — [Authorize]/[AllowAnonymous] on remote services and methods are enforced identically over HTTP and through the in-process SSR bridge, which bypasses routing entirely
  • Lossless 64-bit numerics — long/ulong/decimal cross the wire as JSON strings and type as string, instead of silently rounding above 2^53
  • Collision-free generated type names — namespace qualification when short names clash or would shadow a DOM global, with [SvelteType] to override
  • MIT licensed, with license metadata on every package