Re: [cedet-eieio] Cleaning up the EIEIO namespace

[Eric M. Ludlam]

On 02/13/2013 11:31 AM, David Engster wrote:
> Stefan Monnier writes:
> > I don't know CLOS either.  I also don't know EIEIO enough to know for
> > sure which functions are "internal" (and can hence move to "eieio-" or
> > even "eieio--" without any problem) and which are "exported", so that
> > renaming them has to be done more carefully (with obsolete aliases).
>
> Eric already posted links to the hyperspec where this can be looked
> up. From your bug report #10781, the following names are from CLOS:
>
> * Most of the slot-* names, like slot-boundp, slot-exists-p, etc.
> * make-instance, initialize-instance
> * with-slots
> * class-name, class-of
> * next-method-p, call-next-method
> * defgeneric, defclass, defmethod
>
> Now, whether to include them in the cl- or the eieio-namespace - I don't
> have a terribly strong opinion on that one. If it's deemed too hack-ish,
> then so be it, and we just prefix everything with eieio-. [Eric, if you
> feel more strongly about those names, then please speak up. :-) ]
>
> Actually, the most critical thing is 'oref' and 'oset', because this is
> used extensively (~1800 times in CEDET), and it is not from
> CLOS. Prefixing that with 'eieio-' would make code using EIEIO very
> verbose, but I guess there's just no way around that...?

I'm fine with renaming most EIEIO unique items with some eieio- flavor 
of prefix.  I know there is a debate of cl- vs eieio-.  Drew summed up 
well that EIEIO is an emulation of a subset of the OO parts of CLOS.  I 
think it is incomplete, particularly since it doesn't handle important 
aspects of method polymorphism, such as allowing method differentiation 
based on multiple input arguments, or against built-in types such as 
string or number.  Thus, my vote would be for an eieio- prefix since 
someday a better cl- or clos- variant might appear.

As for the names of CLOS methods, such as make-instance, etc. 
Personally, I think they should be left alone, but recognize that being 
a part of Emacs can come with a naming cost.  Unfortunately, that makes 
it hard to differentiate between CLOS parts and misc eieio internal parts.

I also think of oref/oset in the same bin as CLOS names such as 
make-instance, etc.  CLOS examples all use slot-value/setf instead of 
oref/oset.  setf is part of cl, and couldn't be used at runtime, so I 
mocked up the behavior with something matching aref/aset instead.  They 
are used all the time, and need a convenient name.  Having oref/oset use 
a different naming convention from the names from CLOS API would be 
inconvenient.

Eric