dcb210
commented
8 years ago I’m not sure why the varargout was created. I just know that the original implementation does not work if a 'typemap(argout) Type “”;’ exists. I think it’s needed for the case where there are no outputs at all, not when an output exists but is not assigned by the user. varargout{1:nargout} handles the no output at all case correctly.
On Sep 23, 2015, at 4:02 PM, Joel Andersson notifications@github.com wrote:
About varargout - I'm not quite sure I understand the logic above. It looks very complicated. I also don't understand when the current design won't work.
The MATLAB user expects to get an "ans" output, so it's quite central.
That being said, the current rule is a bit unclear. A cleaner solution would maybe be to have:
No varargout at all when there is no output for any overload That is, varargout{1} when there is exactly one output for all overloads varargout{:} or varargout{1:max(1,nargout)} depending on whether some feature has been set. A completely other approach is move the whole logic to the MEX. If I'm not mistaken, if you have a mex signature such as:
void fcn(int resc, mxArray resv[], int argc, const mxArray argv[]) you can assign to resv[0] even if resc is 0. That is, the ans output. This way, we can get an 'ans' output only for the specific overloads which have one output. No need to have all overloads behave the same way.
— Reply to this email directly or view it on GitHub https://github.com/jaeandersson/swig/pull/63#issuecomment-142712840.
jaeandersson
About varargout - I'm not quite sure I understand the logic above. It looks very complicated. I also don't understand when the current design won't work.
The MATLAB user expects to get an "ans" output, so it's quite central.
That being said, the current rule is a bit unclear. A cleaner solution would maybe be to have:
No varargoutat all when there is no output for any overloadvarargout{1}when there is exactly one output for all overloadsvarargout{:}orvarargout{1:max(1,nargout)}depending on whether some feature has been set.A completely other approach is move the whole logic to the MEX. If I'm not mistaken, if you have a mex signature such as:
you can assign to
resv[0]even if resc is 0. That is, theansoutput. This way, we can get an 'ans' output only for the specific overloads which have one output. No need to have all overloads behave the same way.