Add isCopyable etc

[mppf]

Motivated by work in PR #14835, I realized that list needs to avoid
having an init= or = function when the element type is a non-nilable
owned, e.g. owned C. The problem is that in that event, the source list
elements would be transferred out of and then store nil. Additionally
list should not include an init= function if the elements do not
include one.

In order to solve that problem, list will need to be able to query
whether or not the elements are copyable. In addition it makes sense to
be able to query if a type is only copyable from a ref, which is the
case for owned C? (say).

This PR adds the following queries to the Types standard module:

• isConstCopyable returns true if the type can be initialized from a
  const value (var x = myConstValue)
• isCopyable returns true if the type can be copy-initialized at all
  (but it might only be initializable from a mutable other variable, as
  is the case with owned)
• isConstAssignable returns true if the type can be assigned from a
  const value (x = myConstValue)
• isAssignable returns true if the type can be assigned at all (but it
  might only be assignable from a mutable rhs variable, as is the case
  with owned).
• isDefaultInitializable returns true if the type has a default value
  (e.g. supports var z: MyType;)

isCopyable and isConstCopyable are chosen because some types
(including map after PR #14793) use the ref-if-modified intent in
init=. For these types, it is not known at resolution time whether the
intent will be ref or const ref.

This PR adds the new functions to Types.chpl and the related primitives
and support in preFold. Flags are computed and applied to a type so that
multiple calls to isCopyable will not result in trying to resolve
init= each time. Additionally, this PR adjusts the expiring value
analysis to avoid visiting task functions twice and tidies some
whitespace in preFold.cpp.

Reviewed by @vasslitvinov and @lydia-duncan - thanks!

• full local testing

Future Work:

• isDefaultInitializable errors with record containing non-nilable owned #14854 - error with isDefaultInitializable applied to records
  containing non-nilable owned

[vasslitvinov]

The implementation looks good.

Note that your PR message would lead to closing #13195 upon merge. I suggest fixing that unless that's the intention.

[vasslitvinov]

The interface looks good to me, given that we need to have this information.

The only thing that gave me pause is hasDefaultValue, whereas I wanted to have isDefaultInitializable. In retrospect, hasDefaultValue is pretty good.

I suggest to coordinate the choice of the name for the above with the error that the compiler issues when default-initializing a non-nilable variable or a record without init(). Possibly by adjusting the error. Michael reports that currently the former reads Cannot default-initialize x: owned MyClass.

[dlongnecke-cray]

These routine names look good to me. It looks like hasDefaultValue would provide me with a more general solution for #14367?

[lydia-duncan]

The only thing that gave me pause is hasDefaultValue, whereas I wanted to have isDefaultInitializable. In retrospect, hasDefaultValue is pretty good.

I think either is fine, but prefer hasDefaultValue for succinctness.

[mppf]

I think if we were to use isDefaultInitializable we would perhaps want e.g. isCopyInitializable.

[mppf]

I have some concern that hasDefaultValue will be confusing when applied to a value.
For example, someone might misunderstand and expect hasDefaultValue(1) to return false because the default integer is 0. I do not thing that isCopyable has this problem.

Therefore I'd like to use isDefaultInitializable and leave the rest as-is.

[lydia-duncan]

I assumed hasDefaultValue would only take a type argument

[mppf]

I assumed hasDefaultValue would only take a type argument

All of these functions in Types, like isNilableClass or isInt take either a type or a value. I'd like to keep these new functions that way as well.

[dlongnecke-cray]

I think if we were to use isDefaultInitializable we would perhaps want e.g. isCopyInitializable.

While it is a bit more wordy I think that isDefaultInitializable captures the intended meaning a bit better. My gut reaction was the same as Vass's in that regard.

[lydia-duncan]

Couple of comments related to the future, but otherwise I think this is good