Replacing "might" with "can" obscures the distinction of possiblity and risk

[tkoeppe]

This is a tracking issue for a discussion that arose from the "could"/"might" wording changes.

We initially proposed replacing "might" with "can" in several cases, but this has a different nuance that might not be desirable.

As @zygoloid puts it:

I prefer the nuance of "might" here, as we're talking about a thing that might unfortunately happen to an unsuspecting programmer, not something they can choose to do.

I wonder if that's a good barometer in general: using "can" to discuss risks seems like it has the wrong nuance.

We have multiple instances where "might" is used to call out a risk, not a possiblity.

Additional feedback on "might" => "can" changes:

This note to me sounds like it's directed at the programmer, not the implementation, and the change from "might" to "can" reverses that. So this is at least a change in nuance.

[Re "f might throw" vs "f can throw" or "it is possible for f to throw":] This is stating something as fact that is not known to be true -- we don't know whether it's possible for f to throw because we haven't seen its definition. (There's a surprisingly subtle difference between "X might Y" and "it is possible for X to Y" here.)

See the discussions in #4384 for context.

I expect that we will want to rephrase the offending phrases more widely than by just replacing one word with another.