|Semantic (Type) analysis phase question firstname.lastname@example.org (Nicolás) (2000-03-23)|
|Re: Semantic (Type) analysis phase question email@example.com (Tom Payne) (2000-03-23)|
|Re: Semantic (Type) analysis phase question firstname.lastname@example.org (Pablo Moisset) (2000-03-25)|
|Re: Semantic (Type) analysis phase question email@example.com (2000-04-01)|
|Re: Semantic (Type) analysis phase question firstname.lastname@example.org (Tom Moog) (2000-04-03)|
|From:||Tom Moog <email@example.com>|
|Date:||3 Apr 2000 04:05:40 -0400|
I agree that one should use the quality of error messages in mind when
deciding where to issue error messages.
However, in many cases even such simple things as type checking cannot
be performed easily at parse time. There is the case of Java in which
classes may forward reference members. There is the case of VHDL
which can overload functions based on the return type as well as the
type of operands. In the case of VHDL one sometime must compare the
possible interpretations of two overloaded identifiers in order to
determine the type of a third operator:
f(i) := g(j);
Another oddity of VHDL is that f might have two perfectly valid
interpretations: (a) a parameterless function returning a vector or
(b) a function of one parameter returning a pointer to a scalar.
Return to the
Search the comp.compilers archives again.