In Bug 5672, it was clarified that treat as can be evaluated lazily.

I'm fairly sure that the input of a typeswitch expression _can't_ be evaluated lazily, yet treat as is specified in FS to be normalized to a typeswitch expression:

[Expr treat as SequenceType]Expr
==
typeswitch ([ Expr ]Expr)
  case $fs:new as SequenceType return $fs:new
  default $fs:new return fn:error()

Is this normalization rule correct?

With the function:

declare function local:items()
{
  (1, 2, 3, "four")
};

consider:

(local:items() treat as xs:integer*)[position() lt 4]

and:

(typeswitch ( local:items() )
 case $new as xs:integer* return $new
 default return fn:error())[position() lt 4]