On Thu, January 4, 2007 13:16, Sven Schreiber wrote:
> First, let's define the concept of a "binary
bundle". A binary bundle is a
> binary file which contains one or more objects. An object can be
> 1) a scalar
> 2) a matrix
> 3) a dataset
> 4) a model
I would also add: 5) some text, comment, etc.
Not difficult, ok.
Why does it have to be binary? Human-readable is an advantag IMO.
you have tons of data, but I doubt you would be using gretl then (no
Point taken. If I knew more than I do about this, the best thing would
actually be xml. Trouble is, I know how to write functions for reading/writing
binary stuff, with xml I'm lost. But I could learn.
About data size: no offense taken, sure, but I'd like you to try what happens
with big datasets. In my experience, gretl is actually very efficient. I've
handled datasets with hundreds of thousands of observations much better than
in other packages.
> I think this would provide a very neat way to do what I assume
> However, the coding effort is definitely not trivial. Could we schedule this
> for 1.7.0?
What if you start out just with the ability to read in "bundles" with
genr/matrix/print commands (for scalars/matrices/text). Essentially just
opening the file and executing the literal contents as if the user typed
them in directly. like a limited version of 'run'. That wouldn't be so
difficult, would it?
Mmmhhh... I don't like this very much. Setting up a specialised version of
"run" that we know in advance is bound to be scrapped soon looks like a waste
of resources to me...
(Should we take this off the user list? but keep me cc-ed please, or
maybe I should subscribe to gretl-devel...)
Indeed. I've cc-ed this message to gretl-devel.
Riccardo (Jack) Lucchetti
Dipartimento di Economia
Facoltà di Economia "G. Fuà"