Re: RFC PATCH PR64432 system_clock regression

[Jerry DeLisle <jvdelisle at charter dot net>]

On 03/01/2015 01:28 AM, Janne Blomqvist wrote:
> No idea about the PowerPC failure, but I wonder why is such a
> complicated approach needed, and why is there a need for the new
> library entry points? Wouldn't the issue reported in the PR (i.e.
> calling SYSTEM_CLOCK with mixed integer kinds, which is a bit
> nonsensical but standard-conforming) be fixed by simply something like

Of course I agree its nonsensical, but this is what it says.  I got into this 
out of curiosity about how one would achieve it.

> if (all arguments are integer(kind >= 8) or real)
>     system_clock_8
> else
>    system_clock_4
>
> which can be evaluated at compile time as types and kinds are fixed,
> and no new library symbols are required.

I actually started out as you describe.  I then was having problems getting the 
results of the calls to system_clock4 and 8 to transfer properly back into the 
callers variables of different kinds. Similarly to what I am now seeing on 
PowerPC.  I simply had better results with KIND=2 for those cases.

I am fairly certain I am missing a step in the conversion process.  I will keep 
exploring.

Jerry