Re: Passed by reference for an octave_value type storing an

[Olaf Till]

On Mon, Nov 12, 2012 at 07:01:35AM +0100, c. wrote:
> Hi Olaf,
> 
> On 11 Nov 2012, at 19:00, address@hidden wrote:
> 
> > Message: 2
> > Date: Sun, 11 Nov 2012 14:58:05 +0100
> > From: Olaf Till <address@hidden>
> > To: Octave Maintainers <address@hidden>
> > Subject: Passed by reference for an octave_value type storing an
> >     external        resource?
> > Message-ID: <address@hidden>
> > Content-Type: text/plain; charset="us-ascii"
> > 
> > Hi,
> > 
> > by this post
> > 
> > https://mailman.cae.wisc.edu/pipermail/help-octave/2012-November/054991.html
> > 
> > I learned that there are Octave Forge packages storing external
> > resources (e.g. descriptors for open serial, parallel, or i2c devices)
> > in classes derived from octave_base_value hooked into
> > octave_value.rep, which are then passed to user space as Octave
> > variables of a newly created type. These Octave variables are then
> > used as arguments providing the descriptor to various user-called
> > functions of the package. Example:
> > 
> > octave-forge/main/i2c/serial/serial_class.h, /serial_class.cc,
> > /serial.c, /srl_close.cc.
> 
> I cannot find the path you mention above, I guess you refer to files in the 
> directory
> octave-forge/main/instrument-control/src/serial/
> right?

Yes.

> > To keep track whether the resource is open or closed, the returned
> > value of .get_rep() of the passed octave_value is casted to non-const
> > with a C-style cast and then a change is made within .rep (e.g. in
> > srl_close.cc).
> > JWE, when I suggested the latter for a previous attempt of yours to
> > treat Comedi-related resources conceptually similar to the above
> > package, you discouraged casting away const this way.
> > 
> > (thread: 
> > https://mailman.cae.wisc.edu/pipermail/octave-maintainers/2008-November/013305.html)
> 
> Recently I gave a short presentation about Octave in a seminar in my 
> department at Politecnico di Milano [1,2]
> Partly inspired by recent discussion in the thread mentioned above and partly 
> because of questions about this 
> topic by some colleagues, one of the examples in the slides (myobject.{cc,h}) 
> is about bypassing Octave's pass-by-value
> semantics.
> 
> There I'm using statements like 
> 
> const myobject mo  (static_cast<const myobject&> (args(0).get_rep ()));
> 
> which does not cast away const. 

Without casting away const you cannot change the referenced
value. Without changing the referenced value, this technique is not
applicable to the issue I described. The issue is that pass by
reference is used to change an octave_value to make it keeping track
of the state (open/closed) of a resource (descriptor).

Olaf

> > But what is the actual danger in it, as long as
> > 
> > - users can't change values of this type except using functions
> >  provided by the package,
> > 
> > - functions provided by the package never make copies, but make each
> >  change within the existing .rep?
> > 
> > (I'm trying to achieve an agreement on a strategical suggestion for
> > such problems.)
> > 
> > Olaf
> 
> c.
> 
> 
> [1] PDF:  http://jordi.platinum.linux.pl/octave/what-is-octave.pdf
> [2] LaTeX sources:  http://inversethought.com/hg/what-is-octave/
> 
> 

-- 
public key id EAFE0591, e.g. on x-hkp://pool.sks-keyservers.net

signature.asc
Description: Digital signature