Julia: lazy coalesce syntax proposal

[StefanKarpinski]

I propose the following syntaxes:

• val ?? default for val === nothing ? val : default
• val ?= default for val === nothing ? val : (val = default)
• d[key] ?? default for get(()->default, d, key)
• d[key] ?= default for get!(()->default, d, key)

[nalimilan]

+1, but I think there are two secondary questions to address:

• should missing be supported too (coalesce handles both nothing and missing)
• should Some values be unwrapped (coalesce does, and that's what languages with a Nullable type do)

While we're at it, I'll mention that some languages like Kotlin use val!! to assert that val !== nothing, equivalent to the unexported notnothing helper (can be used to unwrap Some too); Swift and TypeScript use a single ! for that, but obviously it's not possible in Julia. Some languages like Swift and Kotlin also allow val?.field as a shorthand for val === nothing ? nothing : val.field, and col?[i] for col === nothing ? nothing : col[i]. We might want to ensure this syntax is available in case we want to support it at some point.

[StefanKarpinski]

We could potentially make prefix !!x an operator. This is a useless thing to do currently since !x is an error unless x is one of true, false or missing and !!x is the identity on all of those values. In order to make that non-breaking, we might want to make it an error now (or at least document that it might change). Of course, if we make !!x mean x === nothing ? error() : nothing then that would actually not be a breaking change since any code that worked before would still work.

[rfourquet]

Isn't there a typo in the OP with === instead of !==, and also would it be x === nothing ? error() : x for x!! ?

[StefanKarpinski]

Yes, I got the order wrong in the ternary.

[bkamins]

I have just found this issue. My opinion is that if we implement ?? (which I support) we then can update coalesce then to avoid issues described in #26927.