Re: constructors and multiple entry points

[Gabriel Dos Reis <gdr at integrable-solutions dot net>]

"Alisdair Meredith" <alisdair.meredith@uk.renaultf1.com> writes:

| Gabriel Dos Reis wrote:
| 
| > I know the proposals did not dig into all the corner cases -- and I
| > don't even know whether they considered the case.  But, at some point,
| > someone has to go through the sheer number of proposals and try to
| > paint a global picture and see how they interact with existing
| > implementation techniques; C++ is not known to be orthogonal :-)
| 
| Oh, we dug through that proposal quite thoroughly <g>

no doubt, but I claim the coverage wasn't complete and did not
consider other proposals that are targetting the same syntactically
crowded areas and semantically connected notions.

| I specifically brought up the case of diagnosing recursion and making
| it malformed.  The consensus was that while it was clearly a
| programming error to write recursive ctors, no-one wanted to diagnose
| recursive chains of calls, especially with all the other wonders we can
| throw in to make 'indirect' recursion hard to detect.

I don't consider it "nice".  I think it is an essential invariant.

| It would be nice if there was a sop in there so that well-meaning
| implementations could choose-but-not-be-required-to declare the easily
| diagnosed cases malformed, but I don't think that got any support
| either, as no-one was aware of anyone wanting to do this.
| 
| The point I forgot to make though, is that any program using recursive
| constructors is broken, so I don't know how much it is worth worrying
| about g++ implementation details for broken programs - it might be best
| to just ignore the issue (so long as it does not produce new
| compile/link time errors, and I don't think that is the case here)

By the same token, any program that contains an error is broken.
Consequently, we should remove the notion of diagnosable constructs.

-- Gaby