JSDoc type-cast parentheses applied to function parameters produce invalid output (`(row) = fallback`) — regression in 2.3.3

[thebjorn]

Since 2.3.3 (fix: preserve parantheses of jsdoc type casts, 3cbb677), the cast-parenthesization can fire on the left side of an AssignmentPattern in a function parameter list, where the parentheses are not valid syntax:

const row_template = ($$anchor, /** @type {any} */ (row) = $.noop) => {

(row) = $.noop is not a valid parameter binding, so the printed module fails to parse. 2.3.2 prints the same input as /** @type {any} */ row = $.noop, which is valid.

Related: #164 reported what looks like the same root regression for variable declarators (let /** @type {…} */ (computed);), fixed via #165/#167 — but the function-parameter case here still reproduces on 2.3.6.

Real-world impact

Svelte 5 resolves esrap ^2.x, so any Svelte 5 app with JSDoc-typed snippet parameters ({#snippet row(/** @type {any} */ row)}) now compiles to syntactically invalid JS with svelte ≥ 5.56.9 + esrap ≥ 2.3.3. vite build fails with parse errors that get attributed to whichever plugin re-parses the output first (in our case rolldown's builtin:vite-dynamic-import-vars, with the misleading message "Expected a semicolon or an implicit semicolon after a statement"). Dev mode and unit tests pass; only production builds break, so it surfaced as a Vercel deploy failure on a green CI.

Reproduction (via svelte 5, which drives esrap with loc-anchored comments)

package.json

{
    "type": "module",
    "dependencies": { "svelte": "5.56.9" },
    "overrides": { "esrap": "2.3.6" }
}

Repro.svelte (minimized from a real component; every remaining line is load-bearing — removing any of them makes the output valid again)

<script>
    const handle_delete = async (item) => {
        try {
        } catch (e) {
        }
    }
</script>
<div>
    <div>
    </div>
    {#if ok}
        <div>
            <Form enhance={enhance_create}>
            </Form>
        </div>
    {/if}
</div>
{#snippet row_template(/** @type {any} */ row, )}
    {#if x}
        <td >
            <Btn onclick={() => handle_delete(row)}></Btn>
        </td>
    {/if}
{/snippet}

repro.mjs

import { compile } from 'svelte/compiler'
import vm from 'node:vm'
import fs from 'node:fs'

const out = compile(fs.readFileSync('Repro.svelte', 'utf8'), {
    generate: 'client',
    filename: 'Repro.svelte',
    experimental: { async: true }
})
console.log(out.js.code.split('\n').slice(9, 12).join('\n'))
try {
    new vm.SourceTextModule(out.js.code)
    console.log('>>> output is valid JS')
} catch (e) {
    console.log('>>> output is INVALID JS:', e.message)
}

Run with node --experimental-vm-modules repro.mjs:

export default function Repro($$anchor) {
	const row_template = ($$anchor, /** @type {any} */ (row) = $.noop) => {
		var fragment = $.comment();
>>> output is INVALID JS: Invalid destructuring assignment target

Flipping only the esrap override:

esrap	result
2.3.2	valid
2.3.3	invalid
2.3.6 (latest)	invalid

Notes

• The comments in question are the user's /** @type {any} */ annotations on snippet parameters in the .svelte source; svelte re-emits them through esrap's comments option, and the cast-paren preservation from 2.3.3 wraps the re-anchored target even when it is a binding position.
• A secondary observation while minimizing: re-anchored comments can also land on unrelated nodes as nested casts, e.g. a {@render} call printed as /** @type {any} */ (/** @type {number} */ (row))($$anchor, …) — valid JS, but the comments have migrated from the snippet's parameter list to a call site.
• Happy to test a fix against the original (unminimized) component.