fortran program fails -- help...

[Toon Moene]

>  Still, I'm wondering if it wouldn't be better for
>  compilers to do what's right for existing programs by
>  default, and do ANSI compliance only if asked for.  At
>  least that would make sense where there is a vast body
>  of older code.  My guess is that this is the case for
>  Fortran; it may also apply to C++...

There's no question that we *could* do that; however, it would  
severly impede optimisation.  The reason fo this is that the  
backend's optimisation passes try to optimise the use of pseudo  
registers; unfortunately, once you say -fno-automatic *all* simple  
variables are memory based instead of allocated to pseudo registers,  
which grinds optimisation to a screeching halt.

Cheers,
Toon.