This logs the entire file every single time. Probably forgot to remove it?

Any idea why it was checking for existence before and you're not anymore? Is the env supposed to change and needs to be rewritten? Or will the write fail on the second go around?

Would be nice to explain in the commit why rollup is necessary and why the other bump comes with it so reviewers wouldn't have to go digging

I think this is the wrong polyfill here and you want https://www.npmjs.com/package/url instead?

For context, this is used in the docs when clicking on any link. With the current url polyfill it throws Uncaught TypeError: url.parse is not a function

Would've been nice to know why we need those because most of them are actually used by transitive dependencies

• crypto-browserify -> crypto -> used directly in src/utils/memoize.js
• stream-browerify -> stream -> used by @taskcluster/client-web
• stream-http -> http, used by ws
• https-browserify -> https, used by ws
• browserify-zlib -> zlib, used by ws (for permessage-deflate)
• url -> url, used by catch-links/index.js (see message above since it's wrong right now)
• querystring-es3 -> querystring, docker-exec-websocket-client/src/client.js
• buffer -> buffer, src/components/Shell/index.jsx (Buffer.alloc)
• vm-browserify -> vm, used by asn1.js, which comes from crypto-browserify
• path-browserify -> path, jsonlint from src/components/CodeEditor/json-lint.js
• util used in browserify-zlib

AFAICT process isn't used. I checked the output with process dropped and it seemed to not be used. Probably can remove it

Worth noting here that we should remove docker-exec-websocket-client in a followup since docker worker is gone now.

This makes no sense. We're building a static bundle. There's no difference between dev and production dependencies. If anything, everything is a dev dependency and should be moved there. This commit makes it sound like it fixes something when it's a complete no-op.

And if we want to use both deps/dev-deps to mark "this is a tool" and "this is something getting bundled" then it missed @mdx-js/rollup.

Should be squashed in the right commit

Should be squashed in the right commit

In classic claude fashion, the bot gave up without even trying and decided that the worst possible way of fixing this was the only way.

There's no mention of the error it triggers which rejects bare JSON imports without an 'assert { type: "json" }' attribute is as pretty much as vague as one can get "it didn't work, something about json on the json file".

Worse, this "fixes" a previous commit in the stack that was also supposed to fix it. Overall this is not good, we should find why it's failing if it even is, document what got tried to fix it, give context and proper errors instead of just "let's add 250 lines of random JSON to the codebase"

Finally something the bot got "right". Although the threat here is basically non existent, it's still fixing something the correct way.

I'd rather write it like this though and avoid the double try/catch:

try {
  const code = fs.readFileSync(onScrollFile, "utf-8");
  if (!code.includes(WRONG_CODE)) return;
  fs.writeFileSync(onScrollFile, code.replace(WRONG_CODE, ""));
  console.log(`Fixed react-virtualized in ${onScrollFile}`);
} catch (e) {
  // Ignore write failures; vite will surface the underlying problem.
}