Re: TIFF Image Support Project GSOC

Am 29. März 2021 um 01:27 Uhr schrieb "Carnë Draug":
> On Sun, 28 Mar 2021 at 12:00, Markus Mützel <[email protected]> wrote:
> > Am 25. März 2021 um 19:01 Uhr schrieb "Carnë Draug":
> > > On Wed, Mar 24, 2021 at 11:30 PM Nicholas Jankowski <[email protected]> wrote:
> > > > [...]
> > > > @David - are there any similar existing wrappers/class
> > > > implementations worth pointing out in particular as a reference?
> > >
> > > Not really.  The closest thing is the @ftp class which uses the old
> > > style classes.  Its methods are pretty much single lines that simply
> > > call the corresponding __dunder__ method.
> >
> > A few months back I started implementing a classdef wrapper for
> > RapidXML that also follows a similar approach as the old-style @ftp
> > class.
> > See: https://savannah.gnu.org/bugs/index.php?59245
> >
> > I don't remember in which state it currently is. But I believe it
> > was working more or less.  That implementation also aimed to achieve
> > some compatibility with Matlab's XML interface. IIUC, that isn't an
> > issue for this project because the libtiff wrapper would be internal
> > to Octave.
> 
> Actually, the libtiff wrapper is supposed to implement the Tiff class
> which exists in Matlab and is missing in Octave.  So yeah, Matlab
> compatibility is required.  Having imread/imwrite/imfinfo make use
> that Tiff class is something to do after.

Thanks for clarifying. I didn't read that correctly before.

> This whole having multiple __dunder__ functions which are then called
> from almost single line m files seems a bit silly.  I think it would
> be cleaner to do it all in C++.  Is there a way to add a class to
> Octave from an oct file only?

If you want to have something that looks a lot like a class from a user perspective, you could overload the "subsref" method of a octave_base_value class. You could use that subsref function to dispatch to the "actual" methods (see e.g. the octave_java class in libinterp/octave-value/ov-java.h and .cc). 
But IIUC, you won't have all of the classdef features with that approach (like access restrictions, inheritance(?), ...).
I don't know if it is possible to implement a full-featured classdef class just in C++ code. (I would be interested as well if someone could show how to do that.)

Markus