Readable "is" expressions

[Jason House]

Below is my ramblings about long-standind issues with is expressions.  I think it was inspired by a recent post about how to translate some D 2.x is expressions into D 1.x...  I have a few thoughts on how I might fix the syntax, but I'd first like to know if my objections to the current syntax is shared with others.

According to http://www.digitalmars.com/d/2.0/expression.html#IsExpression, there are currently 8 forms for is expressions:
  1. is ( Type )
  2. is ( Type : TypeSpecialization )
  3. is ( Type == TypeSepcialization )
  4. is ( Type Identifier )
  5. is ( Type Identifier : TypeSpecialization )
  6. is ( Type Identifier == typeSpecialization )
  7. is ( Type Identifier : TypeSpecialization , TemplateParameterList )
  8. is ( Type Identifier == TypeSpecialization , TemplateParameterList )

I see 3 issues with groking is expressions:
  1. "Type Identifier" does not always create something matching type "Type"
  2. Type Specialization has multple forms
  3. Use of ":" or "==" can be ambiguous, especially when reading code

Issue 1:
  In all other parts of D code, "Type Identifier" creates something matching "Type".  In an is expression, this is always true with form 4, sometimes true with forms 5 and 6, and never true with forms 7 and 8.  When it isn't true, the identifier is reused on the right side fo the ":" or "==" and corresponds to a derived type.  With enough thinking about an is expression, the meaning should become clear, but I don't think a casual reader of D code will get it on first glance.

Issue #2
  Type specialization has three forms:
    A. A pure type
    B. A type with deduction arguments
    C. A keyword representing a class of types
  Form C is not valid for all forms of is expressions (more on this in issue #3).  As discussed in issue #1, forms A and C have Identifier mean one thing while form B will have Identifier mean somethign else.  Form B can have up to one symbol match "identifier", and one or more symbols match inside TemplateParameterList.  I think there was a recent post to the newsgroup with an example like is(T U : U*, U) where the identifier U is both the identifier and part of the TemplateParameterList.

Issue #3
  When reading D code, the meaning of ":" can be ambiguous.  It's a form that does not match any languages that I know of.  Unlike ":", the meaning of "==" is clear when reading code.  ":" can be translated as "is implicitly castable to" and "==" can be translated "is equal to".  
  When writing code, an ambiguity pops up when a keyword TypeSpecialization (form C above) is used.  is(T:struct) and is(T==struct) can make sense.  In my mind, T is part of the larger group of structs, just like a short is part of the larger groups of integers (and is implicitly castable to int because of this).  That implies to me that the syntax should be is(T:struct), but this is not correct.  The D syntax is is(T==struct)