transform.filter.code regexes with .+ and /s flag cause "Maximum call stack size exceeded" on large files

[JacobSobota]

Hi! 👋

Firstly, thanks for your work on this project! 🙂

Today I used patch-package to patch vite@7.3.1 for the project I'm working on.

The package which had Maximum call stack size exceeded had a JS file 5.7MB, I know it's a lot but It's a very complex thing, which can't be split unfortunately.

Here is the diff that solved my problem:

diff --git a/node_modules/@nuxt/vite-builder/node_modules/vite/dist/node/chunks/config.js b/node_modules/@nuxt/vite-builder/node_modules/vite/dist/node/chunks/config.js
index 2b39a20..2c31471 100644
--- a/node_modules/@nuxt/vite-builder/node_modules/vite/dist/node/chunks/config.js
+++ b/node_modules/@nuxt/vite-builder/node_modules/vite/dist/node/chunks/config.js
@@ -27587,7 +27587,7 @@ async function getWorkerType(raw, clean, i$1) {
 	if (workerOpts.type && (workerOpts.type === "module" || workerOpts.type === "classic")) return workerOpts.type;
 	return "classic";
 }
-const workerImportMetaUrlRE = /new\s+(?:Worker|SharedWorker)\s*\(\s*new\s+URL.+?import\.meta\.url/s;
+const workerImportMetaUrlRE = /import\.meta\.url/;
 function workerImportMetaUrlPlugin(config$2) {
 	const isBuild = config$2.command === "build";
 	let workerResolver;
@@ -27683,7 +27683,7 @@ function assetImportMetaUrlPlugin(config$2) {
 		transform: {
 			filter: {
 				id: { exclude: [exactRegex(preloadHelperId), exactRegex(CLIENT_ENTRY)] },
-				code: /new\s+URL.+import\.meta\.url/s
+				code: /import\.meta\.url/
 			},
 			async handler(code, id) {
 				let s;

This issue body was partially generated by patch-package.

[github-actions]

Hello @JacobSobota. Please provide a minimal reproduction using a GitHub repository or StackBlitz. Issues marked with needs reproduction will be closed if they have no activity within 3 days.

[JacobSobota]

As I didn't want to share the package of my project I've made a reproduction in a different mannef, you can check it out here:
https://stackblitz.com/edit/vitejs-vite-jnnb4dw6?file=regex-repro.cjs
just run node regex-repro.cjs it will create a 3.8MB minified JS file and will try to run a regex which is exactly the same as in @nuxt/vite-builder and as for me this is the result: assetImportMetaUrl (current): 81135ms it doesn't throw the Maiximum call stack size, due to the project differences, however I'm sure you'll see the problem with that Regex expression :)

[bluwy]

Thanks for the repro! It does seem like that regex specifically is causing problems and the rest are fine, and we may also want to handle this part:

vite/packages/vite/src/node/optimizer/rolldownDepPlugin.ts

Lines 309 to 311 in 9ed2b43

	filter: {
	code: /new\s+URL.+import\.meta\.url/s,
	},

@sapphi-red do you know why we're not using the regex used in the transform hook directly for the filter? e.g. /\bnew\s+(?:Worker|SharedWorker)\s*\(\s*(new\s+URL\s*\(\s*('[^']+'|"[^"]+"|`[^`]+`)\s*,\s*import\.meta\.url\s*(?:,\s*)?\))/dg. Is it because it's complex and could slow down the processing by rolldown?

[sapphi-red]

Before the hook filters were added, there was a quick check before the complex regex:

vite/packages/vite/src/node/plugins/workerImportMetaUrl.ts

Lines 184 to 193 in d3c2167

	function isIncludeWorkerImportMetaUrl(code: string): boolean {
	if (
	(code.includes('new Worker') || code.includes('new SharedWorker')) &&
	code.includes('new URL') &&
	code.includes(`import.meta.url`)
	) {
	return true
	}
	return false
	}

I first went with a regex that aligns with it (slightly different though):

vite/packages/vite/src/node/plugins/workerImportMetaUrl.ts

Lines 184 to 185 in 2436afe

	const workerImportMetaUrlRE =
	/new\s+(?:Worker|SharedWorker)\s*\(\s*new\s+URL.+?import\.meta\.url/s

Then I improved the regex a bit reducing unnecessary backtracks in vitejs#336.

[sapphi-red]

Similar to vitejs#336, maybe using /new\s+URL.+?import\.meta\.url/s instead of /new\s+URL.+import\.meta\.url/s can work? I guess .+ would cause more backtracks than .+?.

[bluwy]

.+? would still stall quite a while with the repro, I think it's because the repro has new URL without import.meta.url so the regex is still trying to continue matching. Should we try using the full regex? It should be able to fail faster if no match. Or we can pass an array to filter.code to match multiple parts like before.

[sapphi-red]

I think we can try the full regex. Passing an array would be an OR so it will match more than having only new URL.