Re: [Swig-devel] Return by value optimisation

[Olly B. <ol...@su...>]

On 2008-04-28, William S Fulton <ws...@fu...> wrote:
> The way I propose this gets implemented is to add a new attribute to the 
> out typemap, which when set flags SWIG to generate the optimal code. In 
> the typemap, $1 will be replaced by the call to the wrapped function 
> instead of the 'result' variable. Some checks will also be inserted into 
> SWIG such that if $1 is used more than once in the typemap, or the 
> 'newfree' or 'ret' typemap contain code, then a warning/error will be 
> generated as this will result in bad code being generated.

If we can reliably detect when this is/isn't going to work, why not just
automatically generate the optimal code when we know it will work?  That
way all user typemaps which could benefit will automatically.

It seems this is safe to use (from what you say) when $1 is used once
and there's no non-trivial newfree or ret typemap.

Cheers,
    Olly