Re: Warning from libgomp during make check

[Rik]

On 09/30/2014 01:08 PM, John W. Eaton wrote:
> On 09/26/2014 08:52 PM, Rik wrote:
>> On 09/26/2014 12:43 PM, Mike Miller wrote:
>
>>> I also think it might be useful to allow setenv (name, "") as it is
>>> now. There might be reasons to want to set an empty value rather than
>>> delete a variable. When I added that cleanup, I wasn't thinking that
>>> it would delete the variable, but that it would set it to an empty
>>> string (which it does) and that OpenMP wouldn't care (but it does).
>
> If I understand correctly, setenv (name, "") does remove the variable
> from the environment on Windows systems.
>
>> My suggestion would still preserve this.  The two-input form of setenv
>> would remain exactly the same.  It would only be the one argument form,
>> setenv (name), which today expands to setenv (name, "") that would change.
>> I would have it expand to unsetenv (name) instead.
>
> If you do that, you'll make Octave's setenv incompatible with Matlab
> since it is documented that setenv (name) is exactly equivlent to setenv
> (name, '').
>
> Is the gnulib unsetenv function portable to Windows?

I guess we'll see.  I left setenv the same and added unsetenv via gnulib in
this cset (http://hg.savannah.gnu.org/hgweb/octave/rev/6f0290863d50). 
Since this function isn't supported by Matlab there isn't a direct
compatibility concern.  I then used the new unsetenv function to take care
of the warning in cset
(http://hg.savannah.gnu.org/hgweb/octave/rev/9163a6e9b096).

--Rik