svelte-scripts.ts

Find a component's script tags: the <script> tags at the top level of the source, in source order. A <script> nested in an element, a block, or <svelte:head> isn't one, matching Svelte, which reads only a top-level tag as the instance or module script.

scanScriptTags finds the candidates. When it proves every one of them top-level, which covers the usual component with its scripts ahead of the template, that's the answer. Otherwise Svelte's parser decides: the source is parsed with script and style content blanked, and the tags at the offsets of the root's instance and module are kept. svelte2tsx parses the same blanked source, but demotes only the tags the parser reads as nested elements, so a tag spelled inside an expression string or an attribute value is still a script to it and isn't one here.

Limits: a </script> or </style> inside a string or comment ends that tag's blanked content early, and what's left over is parsed as template. And svelte2tsx's own tag matcher is narrower in one spelling: it doesn't take a closing tag with whitespace before the > (</script >), so there its virtual doesn't have the scripts as written while this returns them. A legacy component written that way documents no props, with a svelte_prop_failed warning for each export let; a runes one documents none, with no warning.

Options for findScriptTags.

What findScriptTags returns.

A script tag is a module script when it carries a module attribute (Svelte 5) or context="module" (Svelte 4) — Svelte's read_script test, minus its error paths. module inside an attribute *value* doesn't count.

Parses a Svelte source whose script and style content is blanked, and says where its top-level scripts are. Throws when the source doesn't parse.

A script tag the scan matched, before anything has decided whether it's top-level.

Scan Svelte source for script tags with content, in source order, and mark the ones the scan itself proves top-level.

The scan is syntactic and knows nothing about nesting, so it returns nested tags too; findScriptTags is the function that answers which are the component's. Limits: whitespace around an attribute = makes a tag unmatchable (Svelte rejects one of those spellings, and svelte2tsx matches neither), and </script> in a string literal ends the content early.

provenTopLevel holds for a tag when everything from the start of the file to it is whitespace, complete HTML comments, and earlier script tags that are themselves proven. Svelte's parser reads each of those at the root and comes back to the root: whitespace is text, a comment runs to the first -->, and a top-level script runs from its opening tag to the first </script>, the same extents the regex matched. So it meets this tag at the root, where a <script> is the instance or module script.

Anything that would need more than that argument leaves the tag unproven, along with every tag after it:

• any other text or markup before the tag: a <style>, <svelte:options>, an unterminated comment, what's left over when a </script> in a string ended an earlier match early, and tags the regex doesn't match (<SCRIPT>, a spelling with whitespace around =)
• a self-closing <script />, which Svelte's parser rejects at the top level
• attributes outside UNAMBIGUOUS_ATTRIBUTES_MATCHER

How findScriptTags decided which tags are top-level:

• 'scan' — the scan proved every tag top-level (provenTopLevel), so nothing was parsed
• 'parser' — Svelte's parser said which tags are the instance and module scripts
• 'unparsed' — the parser threw, and every matched tag is returned

A located script tag: parsed attributes, content, and where the content starts in the source.

The part of Svelte's parsed root that findScriptTags reads.