Rebase on top of D9629

This breaks the -uacomment flag.

clear the container explicitely

Fix linter issue

Back on your queue, CI is broken.

Your strategy seems scary. How are you going to track down all the calls to make_shared<Proof> ?

• Reduce lock scope
• Avoid unecessary copies of datastructures
• Various simplifications

I don't get what is the intended behavior here. avaproof doesn't seems to be a permission that make sense.

Please go on with this.

Reviewing patches based on top of this, it is abundantly clear that that hasn't been thought through at all.

I think it's time to start using smart pointers instead of chasing proof copies all around.

Looks good overall.

Why isn't it part of D9562 ?

I'm not super convinced this is really a plus, as it adds an indirection to something that just does one thing.

We probably shouldn't ignore unsolicited messages en general.

This is doing a number of refactoring in addition to what it is actually doing. You need to split this up.

This is too big. You are touching a ton of thing everywhere here. I'm sure you can extract many refactoring from there. It doesn't matter if they are trivial, in fact that's the point, they can be landed right away and non trivial things aren't drowned into noise.

I think you'll find this doesn't working once you keep the private implementation - which you should.

So, actually, I think this is wrong.

Not really requesting changes, but back on your queue.

Can you explain why so many things need to be fully recomputed?