Re: move constructors likely a requirement

[kingcrimson]

> Il giorno 5 set 2019, alle ore 20:20, Carlo de Falco <address@hidden> ha 
> scritto:
> 
> 
> 
> Il 05 set 2019 20:10, Carlo de Falco <address@hidden> ha scritto:
> 
> 
> Il gio 5 set 2019, 19:55 John W. Eaton <address@hidden> ha scritto:
> On 9/5/19 12:35 PM, Carlo De Falco wrote:
> 
> 
> 
> I'm surprised those are the only warnings, and that there is a warning 
> on line 122 but not also for line 107 in the same function.
> 
> My guess is that it is due to the  NoAlias wrapper which is applied to 
> variable r being returned on line 122 but not on line 107.
> 
> I suppose returning a derived type prevents the compiler from applying 
> optimization.
> 
> 
> jwe
> 
> c.
> 
> 
> Here is the definition of the NoAlias template class : 
> 
> https://hg.savannah.gnu.org/hgweb/octave/file/d171d356767b/liboctave/array/Array.h#l886
> 
> c.


For the case of the warning on line 122 of tril.cc,     
I don't really see the point in using NoAlias as the elements of the 
Array r are accessed through the fortran_vec () pointer anyway so 
the call operator for r is never actually used.


My suggestion would be to just remove NoAlias then ...
changeset attached.

c.

open_spTJFglc.txt
Description: Text document