Re: [lmi] wxmsw-2.9.0 regression: messagebox doubling

[Greg Chicares]

On 2009-02-26 12:55Z, Vadim Zeitlin wrote:
> On Thu, 26 Feb 2009 06:04:18 +0000 Greg Chicares <address@hidden> wrote:
> 
> GC> My fear is that we might have discovered some area where lmi
> GC> unintentionally relies on implementation-defined behavior of gcc.
> 
>  I don't see how could 5-argument transform() work with input sequences of
> different sizes with any standard library implementation. Or do you mean
> some difference upstream which is responsible for CurrentCoiMultiplier_
> being empty? This is possible, of course.

Yes, I suspect it's "some difference upstream".

>  BTW just to be totally sure: neither CCoiMultiplier_ nor
> CurrentCoiMultiplier_ are supposed to be empty, aren't they?

Correct: they are not supposed to be empty.

> GC> It is interesting that similar failures are observed in classes
> GC> InterestRates and MortalityRates, under similar circumstances.
> GC> Below is an experimental patch that tests the data at various
> GC> points for the MortalityRates case, which fails first.
> 
>  Just in case it can help, here is the sequence of messages I get after
> applying this patch:
> 
>       ---------------------------
>       55 is CurrentCoiMultiplierRealized_.size()
>       55 is years_to_maturity()
>       45 is issue_age        ()
>       65 is retirement_age   ()
>       0 is inforce_year     ()
>       2009 is effective_year   ()

That seems uninteresting, as I expected it to be.

>       [file .\input_realization.cpp, line 440]
>       ---------------------------
> 
>       0 is z.CurrentCoiMultiplierRealized_.size()
>       0 is CurrentCoiMultiplier.size()

Here I have 55 for both sizes. I believe this is where it goes wrong.
I will scrutinize this now.