Fixing a missing defensive check for an absent window global variable that was breaking service worker environments

[rory-instil]

Reusing the root variable defined in the file above - and used in a similar way on line 12 - we can enable service worker environments to use this library.
Otherwise these environments will encounter a window is not defined error.

This will close #397.

[morganhvidt]

This fixes my issue 🥳 I'm looking forward to the merge!

[nyacg]

I think this fixed my issue too, would be great to get it merged since it's approved

[makutamoto]

It seems this PR has not been released yet.
Could you release this?

[TranquilMarmot]

Any chance this could get released soon? 😁

[poltak]

This was merged in a while back now but there still hasn't been a release. Any idea when this might happen?

Lots of web extensions are having to move to manifest V3 in the next few months and one of the big changes there on Chrome is that your extension's background script will now run as a Service Worker. Which lacks a window global and thus turndown stops working. This PR would fix turndown for MV3 web exts

[rory-instil]

I've been using this pnpm patch / patch-package in lieu of a release: https://gist.github.com/rory-instil/2091eeeaf087d6caaff7c91a7e16f384

[martincizek]

Released v7.1.3. Please try it out and let me know if the issue is fixed.

[poltak]

Just verified it fixes it for me in Chrome MV3 background service worker context. Thanks @martincizek !

[poltak]

I spoke too soon without giving it a good enough test. After properly testing it, it errors out trying to access document global, which is not available in service workers, here:
https://github.com/mixmark-io/turndown/blob/master/src/html-parser.js#L42

I see you're using domino to afford HTML parsing in non-browser environments. I'm guessing this would also work in service workers, though I'm not sure how you could easily differentiate standard browser envs from service workers in a similar way to what you're currently doing to differentiate node from browser.

It would also probably also significantly increase the SW script size. An ideal solution would be to afford the user passing in their own HTML parsing implementation, as that may allow users to avoid the script size increase if they're already using some other non-browser HTML parsing solution, like cheerio. Of course, that's more work.

[martincizek]

It would also probably also significantly increase the SW script size.

There are different builds for browser and non-browser. I think you can influence this with how you declare the dependencies. And the online browser-build does not contain domino indeed.

[martincizek]

I spoke too soon without giving it a good enough test. After properly testing it, it errors out trying to access document global, which is not available in service workers, here: https://github.com/mixmark-io/turndown/blob/master/src/html-parser.js#L42

This code should not be accessed if you provide turndown() a DOM node yourself.

[poltak]

Ah that's quite obvious in hindsight. Seems to work fine with a Document object produced via linkedom's worker mode. Thanks for the support @martincizek !