Dargent proposal

[gteege]

(This time I want to be more constructive than the last days 8^) When looking at the dargent layout specifications I noted that they are very specific for the layouted type and repeat much of the information from the type expression. Therefore I thought about integrating the layout specification into the type expressions, resulting in this proposal. For the details see the attached document. proposal-darg-synt.pdf Advantages of the proposal are:

• Simpler surface syntax, no redundancy for specifying the type structure in layout specs (such as for {f1:U32,f2:U32} layout record {f1:4B at 0B, f2:4B at 6B}, it now becomes {f1:U32 sized 4B at 0B, f2:U32 sized 4B at 6B})
• No need for layout declarations and layout names.
• No need for a predicate / test that a layout matches a type.
• Layout polymorphism can be supported with the existing polymorphism mechanism using type variables.
• More flexibility by supporting partial layout specifications.
• Layout information can automatically be specified for other types than records, as it was planned but not yet implemented for the current approach.
• More expressiveness than for the current approach (e.g., layout for unboxed abstract types, layout for type variables, empty layout for the unit type).
• The proposal includes support for endianness specification, including mixed endianness.

I realize that the proposal means substantial modifications for Dargent, but I think it could be worth the effort and I offer support for implementing it (e.g., change parser, extend surface syntax, provide examples and test cases, ...).