Re: [PATCH v4 0/8] MR4380: mshtml: Keep inner window's link to the outer window until it's destroyed.

[Gabriel Ivăncescu]

And of course the gitlab subscription reply doesn't work via email…


Well that's why I wanted it before code freeze, so any possible regressions are detected now. Isn't that the point?

For storage events, yeah I can easily get rid of the test and just keep the current "correct" behavior with a comment. I'd rather keep that than silently drop events (or even if it's logged, it's not that obvious), because that's 1) incorrect and 2) it's extremely hard to debug if something broken ever depends on it. Plus, it's not like sending the events is extra code than silently dropping them, most of the patch is in the tests.

As for the "hide problems in the future", I actually think the exact opposite. Right now, we use the document's outer_window in same way, except it's completely broken and can access invalid memory if the outer window happens to die, which would be very hard to debug (especially considering most apps using mshtml are bloated and slow with millions of frameworks).

What problem would this hide? The things that *are* different than before (e.g. parentWindow), are tested so they're fixed and correct, with the others being same as before, except not randomly crashing in some cases.

Maybe it's not by itself a strong argument, but outer_window on document is now just simply broken and I'd like to fix that before the code freeze.

Another reason is that such things would also be very hard to figure out if something even depends on (lack of events especially if it's corner case like this). Considering it also fixes a broken reference from outer_window that's completely arbitrary right now, I think that's reasonable enough.

-- 
https://gitlab.winehq.org/wine/wine/-/merge_requests/4380#note_53975