Path: utzoo!attcan!uunet!mcvax!enea!sommar
From: sommar@enea.se (Erland Sommarskog)
Newsgroups: comp.lang.eiffel
Subject: Re: Multi-branch instructions and unique values (Version 2.2 preview)
Message-ID: <4377@enea.se>
Date: 19 Mar 89 11:36:04 GMT
Organization: ENEA DATA AB, Sweden
Lines: 68

Bertrand Meyer (bertrand@eiffel.UUCP) writes:
>   It is important to note that the test  instruction  has
>no ``else'' clause or equivalent, and that its effect if none of
>the test values matches that of the  expression  is  NOT  to
>execute  an empty operation but to trigger an exception. The
>test instruction (as its Pascal  and  Ada  counterparts)  is
>potentially  dangerous  because it reflects the programmer's
>expectation that the value of e will be one  of  the  stated
>test values.  The least we can expect from the programmer is
>that he precisely specify what these expected values are.
>If  at  run-time  the expression does not evaluate to any of
>these, then this is a serious abnormal case that  should  be
>detected immediately, not quietly ignored.

I fully share the view that if the case selector is not matched,
this should be considered as an error and an exception should be
raised. However, I object to the omission of an OTHERWISE clause
and to to the comparison to Ada and Pascal.

Ada and Pascal first. I don't have any Pascal standard reference
at hand, but the behaviour I expect is to abandon execution if the
CASE selector is out of range, unless an OTHERWISE clause have
been given. (OTHERWISE is not standard Pascal.) A problem may
be if the compiler allows you to turn these checks off. (VAX-
Pascal does.)
  Ada have a different approach. The fragment below is illegal
Ada:
   ch : character;
   ...
   Case ch of   
      when 'A'..'Z' => Text_io.Put_line('Capital letter');
      when 'a'..'z' => Text_io.Put_line('Common letter');
   End case;
Ada requires you to either mention all elements of the selector's
subtype or give an OTHERS clause. The only possibility to get into
a CASE statement in Ada without a match is if the selector is outside 
its subtype due to defaulting from initiating it. This certainly is a 
problem, but hardly not one with CASE statements as such.

Why have an OTHERWISE clause? Say we have a simple interactive 
program where we read one character from the user and then chose
a command. Without OTHERWISE I would have to write: (Pascal syntax,
since I'm lazy.)
    CASE 'ch' OF
      'A', 'a' : Do_a_command;
      'F', 'f' : Do_f_command;
      'R', 'r' : Do_r_command;
      chr(0)..'@',
      'B'..'E',
      'G'..'Q',
      'S'..'`',
      'b'..'e',
      'g'..'q',
      's'..chr(255) : writeln('Illgeal command');
    END;
With an OTHERWISE it would be very cleaner. Not talking on how much
easier it would be to add one more command.
  There are other situations where an OTHERWISE nicely exhibits the
problem behind the logic. At least in Pascal and Ada programs. I
understand that Eiffel will take away some of these situations,
but I doubt that it will remove all of them.
  My suggestion is that the TEST statement is given an optoinal
OTHERWISE clause. If the clause is not present, an exception is
raised if there is no choice matching the selector.
-- 
Erland Sommarskog
ENEA Data, Stockholm              This signature is not to be quoted.
sommar@enea.se