>> Both 1 and 3 would provide full compile-time type safety, at the expense
>> of being a bit more horrible than 2.  I think I'm leaning towards 1, but
>> I don't have strong feelings about any of them.  Having thought about
>> this now though, in response to your question (thanks!), I definitely
>> think that we should do something here better than what I presented in
>> these patches.
>>
>> I'd be very curious to hear your thoughts.
>
> I think that options 1 or 3 are the best choices. My only qualm with
> option 1 is that it uses a callback. I can't even really explain *why* I
> don't like callbacks (maybe because I had a bad time trying to use them
> within Rust before, when I was lacking experience), but it just feels
> somehow wrong to me. (Maybe it's also because I stereotype callbacks to
> C -- which isn't necessarily a bad thing -- but it would certainly make
> me happier to find a more Rust-y way to approach this.)
>
> Thinking about it for a little bit longer, I'm leaning a bit more
> towards option 3, just because I think it would be a better idea to
> associate a trait and a socket type. My gut feeling (keeping in mind my
> very brief knowledge of traits, and even less in that of callbacks) is
> that this would end up being nicer to implement, though maybe not as
> simple.

In the time since I last posted, I've been leaning towards option 1.
But I don't really like callbacks either.  I think the main reason for
me is that then it's not really obvious what the function _does_.  It
some cases (if the callback returned false) it wouldn't actually do
anything at all, which feels a bit weird.  What would a good name for
such a function be?  It becomes very difficult.

On the other hand, making every enum variant a struct of its own feels
like a hack around a limitation of Rust, and it's a lot of boilerplate.
The size of the change puts me off it.

I think I'll probably have to try both out and see what feels best.

>> You might also be interested in implementing one of them.  Would be a
>> good starter crosvm exercise.  It's probably the type of thing where you
>> could be pretty confident you'd got it right if it compiled, but I could
>> offer some guidance on testing it as well, if you wanted.  If not,
>> though, please give me your thoughts on approach anyway, and I'll
>> implement whichever one we think is best myself at some point. :)
>
> Although I would be interested in implementing one of them (#3,
> specifically), time is not on my side at the moment. I've recently
> started my first 40-hour-per-week job, so I'd like to adjust to that
> schedule before I really start doing things again.

Oh, congratulations!  Definitely let yourself adjust to the job as a
priority!

>> I'll definitely wait longer to push crosvm patches in future, now that I
>> know somebody might be looking at them!
>
> Don't worry about it; it's because I took so long to respond. :P

Didn't you respond, like, the next day? :P

> P.P.S. Thanks for going into such detail in all your responses (and
> apologies for not doing the same)!

No worries!  It was really useful to have somebody on the other end to
go into detail to, because otherwise I wouldn't have thought about this
so much and come to the decision I did.