[Fwd: Question on PR25289] Latest patch attached

[FX Coudert]

> That fallback was a conservatism in case something else screws up.  Not 
> really needed.  However, It is exactly what we have now without the patch.

Yes, but... there is no fallback in the front-end, which means it will 
never work. What about using a simple "int" (gfc_c_integer_kind, IIRC) 
in the fallback case? Admitedly, it shouldn't happen very often.

> OK, let me make sure I have this right.  You are saying leave gfc_offset 
> at off_t everywhere.  Then in io.h (st_parameter_dt) define:
> 
> - gfc_offset rec;
> + gfc_large_io_int_kind rec;
> 
> Is this what you wanted?  gfc_large_io_int_kind being kind=8 or kind=4 
> depending on the system.  I must admit this will be less intrusive and 
> yet easily understood.

That's what I'm proposing, except gfc_large_io_int_kind if the front-end 
name, and in the library you probably want to define it as 
GFC_LARGE_IO_INT or something like that (it's not a kind any more, but 
an integer type).

FX