NICE Language Committee: MODEL
To: VIP
Reply-to: bertrand@eiffel.com
Subject: ST-SYNONYMS
Copy to: eiffelgroup
From: Bertrand Meyer
Mailer: BOOM
-----------------------------------------------------------
Key : 94-lang-synonyms
Message from: Steve Tynor (late March 1994)
ETL is not clear (or not clear enough, for me!) on what the following
means:
a, b is
external "C"
end;
[Note: there is no explicit `alias' clause.] Specifically, do `a' and
`b' both call the same external function, "a" or do they each call a
different one ("a" and "b" respectively)?
The multiple declaration rule in section 5.16 says that this is
equivalent to:
a is
external "C"
end;
b is
external "C"
end;
which would imply that `a' calls "a" and `b' calls "b".
However, the section is liberally scattered with the term `synonym'
which would imply that `a' and `b' ought to do the same thing (hence
both `a' and `b' should both call the same external ("a"?)). On
reflection, however, even when talking about non-external features,
synonym is an inaccurate term. For attributes:
c, d : INTEGER;
`c' and `d' are _not_ synonyms: they introduce two separate attributes.
[On the other hand, as I reread section 5.16, I think I could construct
an argument that it requires `c' and `d' to indeed refer to the same
attribute and be synonyms, despite the fact that this is inconsistent
with the semantics of entity_declarations for formal arguments and
locals.].
The current Tower implementation was (probably incorrectly) influenced
by the synonym-interpretation and causes `a' and `b' to both call
"a". We've had reports from customers that the ISE compiler calls "a"
and "b". I've not had a chance to test Eiffel/S. What do we as a group
think is correct? I'm leaning to the ISE interpretation due to
orthogonality w/ attribute semantics (or at least
entity_declaration-consistent attribute semantics).
If we all agree, I think the use of the term `synonyms' is misleading
and ought to be stricken from the next version of ETL.
---------------------------------------------------------------
Answer (Bertrand Meyer)
ETL states explicitly that synonym declarations are
equivalent to duplicate declarations. I think this is indeed
the semantics that is appropriate in practical cases.
So it is indeed correct that a should call a and b should call
b in the example.
I agree that the term "synonym" may be misunderstood but
can't think of a better one for the moment. I see this
as a problem of expression and will look for a better
way to describe and explain the notion. But as for the
actual semantics there is no doubt (in my mind
at least) that the correct interpretation
is the one that works as if the declarations had been
duplicated.
|
