[metroholografix]

I reworded my comment. My point remains in that a lot of possible issues disappear if I just choose to destroy the entire directory. There is a clear difference between one Chrome instance (and all its associated windows/tabs) segmented into one profile directory and in-process segmentation. The issues at [1] seem to invoke unexpected behavior in that you're asking for a new temporary container but you can't know ahead of time if that's what you will get. I'm writing "seem" because it's not clear from the description how that interaction works. They could be better worded.

With segmentation enforced at instance boundary (rather than in-instance), there is no unexpected behavior of this sort. All links open in the segmented instance that the browser window/tab you're using belongs to. If you want a new container, you start a new instance and you know that's exactly what you will get. There is no possible "fail open" result. Note that I'm not saying the Firefox behavior you described is a major issue, just that it proves you can have unexpected scenarios.

Moreover, jedberg is correct in that cross-profile data leaking is possible (partly what I meant by "implementing profiles" properly), except that it's very easy to see if that's happening without auditing Chrome. Use a tool that records all filesystem operations (e.g. dtrace on macOS).

At the end of the day, I choose one set of trade-offs over another.

[1] https://github.com/stoically/temporary-containers/wiki/Isola...

[brmgb]

You are being disingenuous.

The issue you are quoting is about opening new containers while following links. That's something your solution doesn't support at all.

Your solution is akin to manually opening a new container. That will always work in Firefox.