I guess this is more temporary patches since we haven't fixed the asymmetry between setting open and calling show().

Yeah idk what the end result we want is, perhaps we can change this to an assert in future and make sure close watcher is created when open is added but for now this seems like the minimal fix to prevent a crash.

I think it's potentially worth a discussion at whatnot about what we envisage the end state to be for the open attribute because it causes many an issue so it might be worth spending time fully fixing it up.

So I just put up a patch in Chromium to fix (I think all?) of these crasher/assert cases. Instead of relaxing the asserts, it adds attribute change steps for open that construct the close watcher and add the dialog to the open dialogs list when the open attribute is added to the <dialog>. WDYT about that approach, instead of the one in this PR?

Provided there's no unexpected behaviour (such as the dialog being added to the list (set in chrome so might not be noticed) twice when calling show() and showModal()) I think that's preferable. Would solve the initially open case too.

For the sake of completeness I think it should match :open too. I know that's not a spec concern but it would be good to add a WPT to ensure it's synced to the attribute correctly. (Likewise when removing the open attribute)