hash-table iterators

hash-table iterators

[Roland Orre]

Dear guile developers,
I wrote to the guile user list about this earlier, but I would like
to rise the issue also here. 

I consider that the fundamental iterators for hash tables should
iterate over the handles not key and value as now.

That is, I see
hash-for-each-handle
hash-map-handles

as more fundamental than the currently included hash-for-each and
hash-map as the latter can be implemented from the former but not
vice versa.

/Roland




_______________________________________________
Guile-devel mailing list
Guile-devel@gnu.org
http://mail.gnu.org/mailman/listinfo/guile-devel

Re: About hash-table iterators

[Mikael Djurfeldt]

[Cross-posting to guile-devel]

Roland Orre <orre@nada.kth.se> writes:

> The elements of a hash table I consider the handles, not the key and
> the value as separate entities, therefore I don't consider the current
> iterators hash-map and hash-for-each in guile 1.7 very useful and the
> basic iterators I've used over the year, i.e iterating over the handles,
> can not be constructed from them. To be able to implement my previous
> functionality I had to copy and modify a few routines from hashtab.c to
> be able to implement e.g
>
> hash-for-each-handle
> hash-map-handles
> and from the latter can then e.g hash-table->list be implemented
>
> (define-public (hash-table->list htable)  (hash-map-handles id htable))
>
> and when you don't want to be able to modify the original table you can
> use hash-map instead:
> (define-public (hash-table-copy->list htable)  (hash-map cons htable))
>
> Is this a general interest? In that case these maybe they should be
> considered the basic iterators provided instead of hash-for-each and
> hash-map as the latter can be implemented from the other but not vice
> versa?

The design decision for hash-for-each and hash-map was to abstract
away the handle's, which are lower-level details of the representation
of the table, and, also, to promote a functional style of programming.

The decision is supported by the same choice made for Common Lisps's
`maphash' (although Common Lisp has setf so that the side-effects (the
*only* effects, btw) also include mutating the table).  Also, java
does not use the concept of handles either.

However, we *do* support handles in another set of functions.
(Unfortunately, the abstraction is not complete here.  If we pass out
a handle object, we should have accessor and mutator functions for
this object and not just presume that it is a pair!)

So, I'm inclined to support your idea of introducing
hash-for-each-handle.

I don't think we need to introduce hash-map-handles, though.

Also: Is it the case that hash-map has a really bad name?  It strikes
me that one would expect hash-map to return another hash table.
Should it instead be named hash-map-to-list or something better?

M


_______________________________________________
Guile-devel mailing list
Guile-devel@gnu.org
http://mail.gnu.org/mailman/listinfo/guile-devel

Re: About hash-table iterators

[Roland Orre]

On Tue, 2004-01-27 at 17:07, Mikael Djurfeldt wrote:
> [Cross-posting to guile-devel]
> 
> Roland Orre <orre@nada.kth.se> writes:
> 
> > The elements of a hash table I consider the handles, not the key and
> > the value as separate entities, therefore I don't consider the current
> > iterators hash-map and hash-for-each in guile 1.7 very useful and the
> > basic iterators I've used over the year, i.e iterating over the handles,
> > can not be constructed from them. To be able to implement my previous
> > functionality I had to copy and modify a few routines from hashtab.c to
> > be able to implement e.g
<snip>
> The design decision for hash-for-each and hash-map was to abstract
> away the handle's, which are lower-level details of the representation
> of the table, and, also, to promote a functional style of programming.

Yes, this is a good argument, although I like functional programming
style I'm far away from any purist here. Often I like the imperative
style of doing things because it's often more intuitive (at least for
me...). Although in the case with hash tables I consider the handle
keeping the value together as fundamental and is often using it this
way.

> However, we *do* support handles in another set of functions.
> (Unfortunately, the abstraction is not complete here.  If we pass out
> a handle object, we should have accessor and mutator functions for
> this object and not just presume that it is a pair!)

Yes you are right. This may save a lot of work in the long run...

> So, I'm inclined to support your idea of introducing
> hash-for-each-handle.

I'm very pleased by this (the less code to maintain in user space,
the better)

> I don't think we need to introduce hash-map-handles, though.
No, you are right, that is not necessary.

	Best regards
	Roland




_______________________________________________
Guile-devel mailing list
Guile-devel@gnu.org
http://mail.gnu.org/mailman/listinfo/guile-devel

Re: About hash-table iterators

[Mikael Djurfeldt]

Mikael Djurfeldt <mdj@mit.edu> writes:

> The design decision for hash-for-each and hash-map was to abstract
> away the handle's, which are lower-level details of the representation
> of the table, and, also, to promote a functional style of programming.
>
> The decision is supported by the same choice made for Common Lisps's
> `maphash' (although Common Lisp has setf so that the side-effects (the
> *only* effects, btw) also include mutating the table).  Also, java
> does not use the concept of handles either.

BTW, since then I've looked around, and virtually every Scheme
implementation I find have made the same choice as we have.

> However, we *do* support handles in another set of functions.
> (Unfortunately, the abstraction is not complete here.  If we pass out
> a handle object, we should have accessor and mutator functions for
> this object and not just presume that it is a pair!)
>
> So, I'm inclined to support your idea of introducing
> hash-for-each-handle.

Since I didn't get any feedback I went ahead and checked this into CVS
HEAD:

2004-02-13  Mikael Djurfeldt  <djurfeldt@nada.kth.se>

	(scm_internal_hash_for_each_handle, scm_hash_for_each_handle): New
	functions.


> I don't think we need to introduce hash-map-handles, though.
>
> Also: Is it the case that hash-map has a really bad name?  It strikes
> me that one would expect hash-map to return another hash table.
> Should it instead be named hash-map-to-list or something better?

2004-02-13  Mikael Djurfeldt  <djurfeldt@nada.kth.se>

	* hashtab.c, hashtab.h (scm_hash_map_to_list): Renamed from
	scm_hash_map.

2004-02-12  Mikael Djurfeldt  <djurfeldt@nada.kth.se>

	* boot-9.scm (module-map): Renamed hash-map -> hash-map->list.

M


_______________________________________________
Guile-devel mailing list
Guile-devel@gnu.org
http://mail.gnu.org/mailman/listinfo/guile-devel