Re: [guppi-list] Stabilizing the DataSet interface?

[Jon Trowbridge <trow emccta com>]

On Tue, Feb 02, 1999 at 07:05:36PM +0100, Asger K. Alstrup Nielsen wrote:
> I've had a look at the new DataSets, and my conclusion is that
> I like it.  Getting rid of the Values was good, and things
> are simpler.  However, I did do a few changes to get a 
> DataConverter into it, because I feel it was a bit too
> crippled.

Yeah, after you made your original changes, some deficiencies in how I
had set things up occured to me.  In particular: as an arrogant
American:-), I hadn't given any thought to i18n issues.

> The DataConverter class hierarchy is a simplified 
> replacement of the old DataType hierarchy.
> <explanation of DataConverter omitted>

I haven't really sunk my teeth into this code yet, so I'm really
speaking from ignorance here, but just two quick questions:

(1) Is there any way we can use locales to make things like the
    RealConverter automatically do the right thing, i18n-wise?
    (I don't know the first things about locales or how they work...)

(2) Is there a nice way that we could make the use of the
    DataConverter framework optional for future DataSets.  (Primitive,
    half-formed ideas about some of the currently-unimplemented
    DataSets are rattling around in my brain, and I think that there
    may be some cases where the DataConverter is overkill.  On the
    other hand, I do see how some of the expanded functionality of the
    DataConverter could be useful for really smart and flexible ascii
    imports and other stuff... so only take this question half-seriously.)
 

> Besides that, I wondered about the Random instance
> in the DataSet is doing.  I thought that we had a global 
> Random instance, so do we need one in each DataSet?

This is all mildly screwed up, but harmlessly so.  Constructing new
Random objects is expensive, and at various points I've done funny
things to try to deal with that.  And Random isn't thread-safe, so
rather than make it thread-safe I've done other funny things to deal
with *that*.  And the result is a rather confused state of affairs
that (I think) can be easily recovered from later and doesn't really
touch all that much code...

> Finally, I hope we can soon declare some sort of freeze of the
> DataSet interface.

Yes, I think that we are getting close to that point.  The last time I
had a bit of free time to play with things, I got linear regressions
working in the guile bindings --- so I was *actually able to use goose
to calculate something*!  Nothing important or useful, mind you, but I
had a sense of accomplishment that you just don't get from mucking
around with low-level design and data-structure issues.  On the other
hand, I'm quite happy we have spent so much time mucking around with
said issues... I think that everything has been much improved because
of it, and I feel confident going forward.

-JT