Article 129610 of comp.os.vms:
Path: nntpd.lkg.dec.com!denton.zko.dec.com!amartin
From: amartin@denton.zko.dec.com (Alan H. Martin)
Newsgroups: comp.os.vms,vmsnet.misc
Subject: Re: C++ Question: why isn't label() an lvalue?
Date: 19 Sep 1995 15:26:15 GMT
Organization: DEC
Lines: 30
Message-ID: <43mnen$us4@nntpd.lkg.dec.com>
References: <1995Sep14.091737.299@thehulk>
NNTP-Posting-Host: denton.zko.dec.com
Xref: nntpd.lkg.dec.com comp.os.vms:129610 vmsnet.misc:4367

In article <1995Sep14.091737.299@thehulk> munroe@dmc.com writes:
>DEC C++ question here. GCC allows this inside a class definition:
[Applying ``&'' to the results of a struct-valued function call]
>But DEC C++ 1.3 complains as follows:
...
>            return (str$compare(&label(), &v) == 0 ? SS$_NORMAL :
>SQS$_NOTINSLOT) ;
>................................^
>%CXX-E-NEEDLVALUE, In this statement, "label()" is not an lvalue, but occurs in
>a context that requires one.

The program does not conform to the ARM (or the 28-Apr-95 ANSI C++ draft, for
that matter).

ARM §5.3 (Unary Operators) states on p. 55, "The result of the unary &
operator is a pointer to its operand.  The operand must be a function, an
lvalue, or a qualified-name.  ..."

Furthermore, ARM §5.2.2 (Function Call) states on p. 51, "A function call
is an lvalue only if the result type is a reference".  ``label()'' is a struct
value - not a function, qualified-name or reference.


You could store the results of calling label into a temporary variable.  Or,
you could define a jacket for str$compare (e.g. ``str_compare'') which takes
arguments of type const ``DSC$DESCRIPTOR &'', and then call the jacket with
``label()''.  (You could even overload str$compare).
				/AHM
-- 
Alan Howard Martin				AMartin@TLE.ENet.DEC.Com