Re: Pre-RFC: namespace literal as first class citizen

On 2022-06-29 11:34 p.m., Branislav Zahradník wrote:
> Idea:
> Introduce namespace literal into language with it's own ref type and so.
> 
> Current state:
> - any string when used as a reference in method invocation is treated as namespace
> - language already supports trailing :: syntax, producing string without ::, and 
> warning, when namespace is not available yet
> 
> Other languages examples:
> - Java and their java.lang.Class (or .class attribute)

As it happens, I already invented a corresponding portable data type or possible 
representation called "Nesting" in my Muldis Object Notation / Muldis Data Language.

A "Nesting" is isomorphic to an array of 0..N character strings and has a 
literal syntax that looks like a Perl multi-level fully-qualified package name, 
such as "::Foo" or "Foo::Bar" or "Foo::Bar::Baz" etc, and it is intended to map 
to the same, or to the analogy in other languages.

A "Nesting" is also classified as a non-collective or terminal type, like 
booleans/numbers/character strings/blobs/etc, and unlike arrays/associative 
arrays/etc.

A "Nesting" is also used to represent a lookup or address into a nested 
collection type such as a Perl hash of Perl hashes of Perl hashes etc 
arbitrarily deep, and could also be used in an extension to a Perl hash literal 
to declare such, for example, these 2 tentative Perl expressions are equivalent 
/ produce the same hash ref with a hash ref element:

   {
     x1 => 3,
     x2::y1 => 4,
     x2::y2 => 5,
     x3 => 6,
   }

   {
     x1 => 3,
     x2 => {
       y1 => 4,
       y2 => 5,
     },
     x3 => 6,
   }

So what has this taught me about what may be good as a namespace literal in Perl?

First, while I could take it or leave it, I would lean towards the namespace 
literal NOT being a Perl ref type, but rather a Perl scalar type.  That's less 
likely to be surprising.

Syntactically, existing Perl programs could get this for free under newer Perls 
because the literal syntax is identical to how package names are now, at least 
if they had the "::" in them, or even if they don't, if we have a bareword 
followed by a "->" then we know it is this too.

Perhaps to maximize backwards compatibility and minimize changes, this could be 
implemented as a regular Perl character string but have some extra internal flag 
similar to the "created as boolean" (name?) or such flags.

For serializer/deserializer purposes, similar to the internal flags to treat 
booleans and numbers and strings as disjoint so they round trip through JSON or 
other formats, the marker can result in package names being mutually disjoint 
with all of those, and then supporting serialization formats such as mine can 
exploit it.

Anyway, this is just ideas I'm putting out there, I'm not really trying to push 
it at this time.

-- Darren Duncan

• Pre-RFC: namespace literal as first class citizen by Branislav Zahradník
• Re: Pre-RFC: namespace literal as first class citizen by Darren Duncan