Re: [sv-ec] Proposal for Mantis 3001

Gord,

Point taken about the evaluation order concerns. I would be
disappointed to lose the generality, but I would prefer the restriction
you suggest, rather than not having typed constructors at all.

It was clear enough to me that, if a typed constructor were used in some
general expression, its "construction" activity should take place before
the expression saw its value. But I accept your concerns, and I'm
willing to back-pedal in the way you suggest.

I don't really see a problem with ref arguments. Today you surely
wouldn't expect to allow an arbitrary expression to appear as the actual
for a ref-formal argument; there is a plausible interpretation (the
reference is to a temporary variable that holds the expression's result)
but it's not really useful, and current tools generally don't allow it
even if the formal is "const ref". Similarly I wouldn't expect a
typed-constructor to be legal in such context.

Ternaries are more tricky, I agree. Evaluation order then becomes a
significant issue and I can see why you might not want to go there. We
really don't want objects popping into existence in unexpected and/or
indeterminate ways.

I'll upload a revised, restricted version before Monday's meeting.

Jonathan

On 11/09/2011 19:53, Vreugdenhil, Gordon wrote:
> Jonathan,
>
> I think that there are some fairly subtle issues that open up once you have object construction as a general expression. In particular, the issues of "referencing" might be a potential problem with, for example,
> ref arguments (is that even legal?). I'd have to think about issues related to "mid-expression" construction and reference in, say, ternaries and various other situations and what the implications on overall state might be (and whether they are always well defined in terms of order).
>
> I'd not be comfortable passing this as is right now without some deeper consideration which I won't have time to do until October due to travel commitments.
>
> If the context were restricted to the current forms (i.e. construction only as the immediate RHS of an assignment) I'd be more comfortable that implementations will end up doing the same thing.
>
> Gord.
>
> ________________________________________
> From: owner-sv-ec@eda.org [owner-sv-ec@eda.org] on behalf of Jonathan Bromley [jonathanbromley@ymail.com]
> Sent: Sunday, September 11, 2011 2:34 AM
> To: sv-ec@eda.org
> Subject: [sv-ec] Proposal for Mantis 3001
>
> hi EC,
>
> I've uploaded what I hope is a complete proposal for Mantis 3001.
> It's fairly simple and I think there was some consensus in an
> earlier discussion that "base=Derived::new();" makes sense.
> It would be nice to put this to bed if there's agreement.
>
> Jonathan Bromley
>
> --
> This message has been scanned for viruses and
> dangerous content by MailScanner, and is
> believed to be clean.
>
>

-- 
This message has been scanned for viruses and
dangerous content by MailScanner, and is
believed to be clean.