[boost] Re: Variant implementation change (wasre:unions)finalsummary

From: Eric Friedman (ebf_at_[hidden])
Date: 2003-10-09 21:10:02

brock wrote:
[snip]
> There are a lot of disadvantages in allowing variants to be singular at any
> time.  We've been all over this.  But if we say they can be singular only
> after a failed assignment, then you can pretty safely ignore the possibility
> in most of your code...  I think.  I wonder what Eric and Dave think of this
> or if they finally got tired of all the noise and tuned out :)  Maybe this
> is noise too :)

I've suggested it before.

See http://aspn.activestate.com/ASPN/Mail/Message/boost/1831995.

 From an implementation standpoint, there is no problem at all with 
singular variants; I can code variant so that it destroys itself 
properly even when in a singular state. As well, since assignment to a 
singular variant is a meaningful operation, it is possible to "restore" 
a singular variant to full-status. However, visitation of a singular 
variant is *not* a meaningful operation in anyway and should be left 
undefined.

Given that singular variants will *only occur in the wake of failed 
assignment*, this may not be a problem. Allowing the possibility of 
singular variants at all, however, does represent a departure from where 
variant has been in the past. Before I make a change like this, I would 
appreciate some sort of consent from the community. So far, this has 
been significantly lacking.

Eric