(j3.2006) (SC22WG5.5440) [ukfortran] RE: Proposal of a new requirement : CMDMSG optional argument in 3 subroutines

[N.M. Maclaren]

On Jan 29 2015, Rafik Zurob wrote:
>
>I also checked the POSIX spec page for getenv() 
>to see if it defines any errors -- it does not.  (I'm not saying everyone 
>on UNIX/Linux/BSD calls getenv().  I was just thinking that if getenv() 
>defined errors, some vendors might have defined status > 2 values to 
>correspond to those errors.)  The Windows GetEnvironmentVariable function 
>also does not appear to define errors not covered by status values -1 to 
>1.

POSIX is, as we all know, a crock.  Implementations can check that the
string is valid and unambiguous and diagnose that separately.  Ones
with significant trailing spaces are ambiguous in Fortran terms.  setenv()
forbids ones with '=', but POSIX says that any characters other than
alphanumerics and underscore are system-dependent (see 8.1).

And, of course, Unix-derived systems are not the universe.

Allowing an implementation to do better than POSIX seems a good idea.
Requiring it to do so would be a mistake.

Regards,
Nick Maclaren.