[Proposal] readonly local functions

[tge97]

Summary
Local functions currently support the async, unsafe, static and extern modifiers. This is a proposal to include the readonly modifier.

Motivation
Local functions as they are implemented, even when they are used in moderation, can obscure updates to captured variables and thereby make it harder to read and debug code which uses them. Often, developers want to write local functions which do not update captured variables but there it currently no way to restrict local functions in this way whilst at the same time benefiting from the convenience of captured variables.

Proposed behaviour

1. All captured variables in a readonly local function would be read-only (i.e. they would be treated as if they were in arguments).
2. Like ordinary local functions, readonly local functions in structs would have no access to the instance or instance members.
3. In classes, readonly local functions would have access to readonly fields and static members only.
4. Like ordinary local functions, modifiers applied to the enclosing method take precedence.
5. Readonly local functions would not have access to local functions in the same scope unless they were also readonly.

Concerns

1. The proposal would allow readonly local functions in classes, where readonly methods do not otherwise exist.
2. It seems necessary to very strictly limit access to instance members in classes to avoid ambiguity about the effect of the modifier and maintain a degree of consistency between its usage within classes and structs.

[jaredpar]

Like ordinary local functions, readonly local functions in structs would have no access to the instance or instance members.

Did you mean to say

Like ordinary static local functions, ...

Ordinary local functions have full access to the instance and instance members.

In classes, readonly local functions would have access to readonly fields and static members only.

At the moment readonly is not a valid modifier on class types but it is a topic we've discussed several times over the year. I'm fairly against committing to the behavior of a readonly local functions impact on class fields without us first deciding how it impacts them as a normal method / type modifier.

[tge97]

Thanks for getting back on this.

Ordinary local functions have full access to the instance and instance members.

Does this not violate CS1673?

At the moment readonly is not a valid modifier on class types but it is a topic we've discussed several times over the year. I'm fairly against committing to the behavior of a readonly local functions impact on class fields without us first deciding how it impacts them as a normal method / type modifier.

Yes, I completely agree.

[jaredpar]

Ordinary local functions have full access to the instance and instance members.

Does this not violate CS1673?

For struct yes but your statement wasn't limited to struct, it is a broad statement. Given that you later referred to the behavior for class specifically have to assume you mean the broader context here.

[tge97]

For clarity, 'proposed behaviour 2' refers explicitly to structs.