asm198x-web

The asm198x assembler compiled to WebAssembly, with a wasm-bindgen surface a web page can call. Nothing is uploaded anywhere: the assembler runs in the visitor's tab.

Published to npm as @asm198x/web; scripts/build-npm.sh builds it. The crate itself stays off crates.io: a wasm-bindgen surface returning JavaScript values has no Rust caller.

npm install @asm198x/web

Surface

import init, { assemble, assemble_project, snapshot, listing, dialects } from '@asm198x/web';

await init();

const result = JSON.parse(assemble('sjasmplus', ' org 32768\n ld a,1\n ret\n'));
if (Array.isArray(result)) {
  // The diagnostics array: `result[0].message`, `result[0].span.line`.
} else {
  // The `AssemblyResult` object: `result.bytes`, `result.origin`, `result.symbols`.
}

listing('acme', '*=$c000\n lda #1\n rts\n');  // the `--listing` text, or null
JSON.parse(dialects());                        // [{ name, aliases, blurb }, …]

assemble returns the string asm198x --message-format=json prints, so the CLI reference documents the shape. It returns null for a dialect name the table does not list.

For a named in-memory project, including target selection and the browser-safe native output options:

const result = JSON.parse(assemble_project(JSON.stringify({
  dialect: 'sjasmplus',
  target: 'z80n',             // optional: z80, z80n/next
  root: 'src/main.asm',
  files: {
    'src/main.asm': ' include "part.asm"\n',
    'part.asm': ' nextreg $07,$02\n',
  },
  output: 'raw',              // optional; 'hunk' is available for vasm
})));

The file map is text: source includes are supported, while binary incbin payloads remain outside this surface. hunk is included because it is wholly represented by the returned bytes; filesystem-only CLI behaviour is not.

Build

wasm-pack build --target web crates/asm198x-web --out-dir pkg-web
wasm-pack test --node crates/asm198x-web

The crate is excluded from the workspace, so it carries its own Cargo.lock; cargo update -p asm198x --manifest-path crates/asm198x-web/Cargo.toml refreshes the path entry after the library's version moves.

Size

Every dialect ships in one module. Measured 2026-09-04, opt-level = "z", after wasm-opt: 1,505,582 bytes raw, 514,986 gzipped. A per-platform split (#495) is a decision for when a consumer needs one.

Running what you assembled

snapshot(dialect, source) returns a 48K .sna — the same bytes asm198x --dialect pasmonext --sna writes, verified byte for byte — so a page can assemble and then run the result:

const sna = snapshot('pasmonext', source);
if (sna) spectrum.loadSnapshot(sna, 'sna');   // @emu198x/zx-spectrum

Spectrum Z80 only, and the source needs end <addr> for its entry point, exactly as the command line demands. null means it did not assemble; assemble() says why.

Size: take one architecture

@asm198x/web carries every dialect — 1.4 MB raw, 480 KB gzipped. Almost none of that is useful to a page teaching one machine, so the package is also built per CPU architecture:

| package | raw | gzipped | |---|---|---| | @asm198x/web (all) | 1470 KB | 503 KB | | @asm198x/z80 | 486 KB | 185 KB |

entry is the only thing that names the library's assembler entry points, so a build that does not select an architecture never references it and the linker drops it. Nothing is stripped by hand.

scripts/build-npm.sh z80        # -> @asm198x/z80
scripts/build-npm.sh mos6502    # -> @asm198x/mos6502
scripts/build-npm.sh            # -> @asm198x/web, everything

dialects() reports what the build you have can actually assemble, not the whole table, so a picker made from it can never offer a name assemble then refuses.

Even at 185 KB this is bigger than an emulator embed, so a lesson page should still load it lazily — on the first keystroke, not on page load.