Re: van Emde Boas hash.

Re: van Emde Boas hash.

[A. Soare]

What algorithm uses emacs for hashing the obarrays of symbols ?





____________________________________________________

Gagnez un séjour au Maroc en découvrant les sketchs désopilants des ReVoila sur http://www.lesrevoila.fr/

Re: van Emde Boas hash.

[Stefan Monnier]

> What algorithm uses emacs for hashing the obarrays of symbols ?

Read the source: C-h f intern RET then click on top right to get to the
source code, ...

It's a hash table with simple haqshing into buckets that contain linked
lists of symbols.


        Stefan

Re: van Emde Boas hash.

[A. Soare]

> > I did look at the code, and I cannot understand the algorithm from
> > `make-hash-table'.
> 
> It's one of the standard hashing schemes, where the hash-table is
> resized as it grows, which should hopefully keep the access time
> more-or-less constant.
> 
> 

Seems like red black trees.

This is not better than van Emde B.




Alin




____________________________________________________

Derniers jours pour remporter le séjour au Maroc sur http://www.lesrevoila.fr/ 

Re: van Emde Boas hash.

[A. Soare]

> > What algorithm uses emacs for hashing the obarrays of symbols ?
> 
> Read the source: C-h f intern RET then click on top right to get to the
> source code, ...
> 
> It's a hash table with simple haqshing into buckets that contain linked
> lists of symbols.

As far as I understand, the algorthm is so:

oblookup ( obarray, sym )

 ;; hash is an integer, the index of a bucket that may contain sym
 hash = hash_string (sym)
 ;; bucket is a vector. obarray is a vector of many buckets
 bucket = obarray [hash]
 ;; search sym in its bucket using a naive search
 FOR every symbol S in bucket, check whether S has the same name
 as sym. If so, return S. If no symbol matches, returns the hash.


Even if it is very unlikely that we can find many symbols into a bucket, the algorithm does not require constant time for every search.

Hashing using van Emde Boas requires in the worst case a time of log_2(N), where N is the length of the largest symbol name in obarray.




Alin.



____________________________________________________

Derniers jours pour remporter le séjour au Maroc sur http://www.lesrevoila.fr/ 

Re: van Emde Boas hash.

[Stefan Monnier]

> As far as I understand, the algorthm is so:

> oblookup ( obarray, sym )

>  ;; hash is an integer, the index of a bucket that may contain sym
>  hash = hash_string (sym)
>  ;; bucket is a vector. obarray is a vector of many buckets
>  bucket = obarray [hash]
>  ;; search sym in its bucket using a naive search
>  FOR every symbol S in bucket, check whether S has the same name
>  as sym. If so, return S. If no symbol matches, returns the hash.


> Even if it is very unlikely that we can find many symbols into
> a bucket, the algorithm does not require constant time for
> every search.

Yes, it's a very simple hashing algorithm.  I'm sure we can come up with
something more efficient.  Then again, I haven't seen any indication
that this is a performance problem.


        Stefan

Re: van Emde Boas hash.

[David Kastrup]

Stefan Monnier <monnier@iro.umontreal.ca> writes:

>> As far as I understand, the algorthm is so:
>
>> oblookup ( obarray, sym )
>
>>  ;; hash is an integer, the index of a bucket that may contain sym
>>  hash = hash_string (sym)
>>  ;; bucket is a vector. obarray is a vector of many buckets
>>  bucket = obarray [hash]
>>  ;; search sym in its bucket using a naive search
>>  FOR every symbol S in bucket, check whether S has the same name
>>  as sym. If so, return S. If no symbol matches, returns the hash.
>
>
>> Even if it is very unlikely that we can find many symbols into
>> a bucket, the algorithm does not require constant time for
>> every search.
>
> Yes, it's a very simple hashing algorithm.  I'm sure we can come up
> with something more efficient.  Then again, I haven't seen any
> indication that this is a performance problem.

I should think that it makes up a sizeable portion of byte-recompiling
and autoloading, both of which are not often benchmarked.  But it isn't
something which you commonly do on data sets in a loop.  Scanning files
is done just once usually.

-- 
David Kastrup

Re: van Emde Boas hash.

[A Soare]

> Yes, it's a very simple hashing algorithm.  I'm sure we can come up with
> something more efficient.  Then again, I haven't seen any indication
> that this is a performance problem.

I think that the lists can be hashed using van Emde Boas trees, and this would
give a really great improvement to emacs.




Alin

Re: van Emde Boas hash.

[A. Soare]

Van Emde Boas arrays are hashed arrays (of numbers for example), which have the property that all operations are realized into a constant number of steps.

For example, for an array of length 2^32, the complexity is O(0), and the number of steps is less than log_2 (32) = "maximum 5 steps" for every operation.

The operations are:

* insert a new number
* delete a number
* seek for a number
* find the successor of a number
* find the predecessor of a number



____________________________________________________

Gagnez un séjour au Maroc en découvrant les sketchs désopilants des ReVoila sur http://www.lesrevoila.fr/

Re: van Emde Boas hash.

[tomas]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On Fri, Nov 20, 2009 at 07:41:15AM +0100, A. Soare wrote:
> 
> Van Emde Boas arrays are hashed arrays [...]
> 
> For example, for an array of length 2^32, the complexity is O(0), and
> the number of steps is less than log_2 (32) = "maximum 5 steps" for every operation.
> 
> The operations are:
> 
> * insert a new number
> * delete a number
> * seek for a number
> * find the successor of a number
> * find the predecessor of a number

If  I understood the Wikipedia page correctly (thnks, Deniz), you
can find "next" and "previous" to a number not in the hash? Then
they look like a good candidate to represent markers and boundaries of
overlays, no?

Regards
- -- tomás
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.6 (GNU/Linux)

iD8DBQFLCgpZBcgs9XrR2kYRAvh6AJ9TxH4v/4jFSN90gK/X/QHgJQuf/wCfUYYU
mZjOWuTH3spYQJWvZ3HtWzg=
=7Ccl
-----END PGP SIGNATURE-----

Re: van Emde Boas hash.

[A Soare]

> If  I understood the Wikipedia page correctly (thnks, Deniz), you
> can find "next" and "previous" to a number not in the hash?

When looking for the next of a number N, it will return the next/prev number
from the array greater than N. Does not matter whether N belongs or not to the
array.

> Then
> they look like a good candidate to represent markers and boundaries of
> overlays, no?

I do not understand. How are the markers and boundaries represented ?

Alin.


PS:



I post again this message, because I discovered http://post.gmane.org/, which is
very nice for reading the news groups.

Emacs can be used convenient to read/post to emacs devel? Can somebody help me
to confugire gnus or something else for?

van Emde Boas hash.

[A. Soare]

I am going to implement van Emde Boas. Do you consider that this hash method could find some useful applications in emacs?

If so, I will try to implement it with Emacs Lisp data structures...




____________________________________________________

Gagnez un séjour au Maroc en découvrant les sketchs désopilants des ReVoila sur http://www.lesrevoila.fr/

Re: van Emde Boas hash.

[Stefan Monnier]

> I am going to implement van Emde Boas. Do you consider that this hash
> method could find some useful applications in Emacs?

I have no idea what it is,


        Stefan

Re: van Emde Boas hash.

[Deniz Dogan]

2009/11/19 Stefan Monnier <monnier@iro.umontreal.ca>:
>> I am going to implement van Emde Boas. Do you consider that this hash
>> method could find some useful applications in Emacs?
>
> I have no idea what it is,
>
>
>        Stefan
>
>
>

http://en.wikipedia.org/wiki/Van_Emde_Boas_tree
Summary: Associative array at O(log m) for "all operations", where "m"
is the number of bits used for the integer keys.

-- 
Deniz Dogan

Re: van Emde Boas hash.

[A Soare]

> I have no idea what it is,
> 
>         Stefan
> 
> 

I see in elisp manual so:

   * Lookup in a hash table is extremely fast for large tables--in
     fact, the time required is essentially _independent_ of how many
     elements are stored in the table.  For smaller tables (a few tens
     of elements) alists may still be faster because hash tables have a
     more-or-less constant overhead.

What algorithm uses the hash of elisp ? It is faster than van Emde?



I did look at the code, and I cannot understand the algorithm from
`make-hash-table'.

Re: van Emde Boas hash.

[Stefan Monnier]

> I did look at the code, and I cannot understand the algorithm from
> `make-hash-table'.

It's one of the standard hashing schemes, where the hash-table is
resized as it grows, which should hopefully keep the access time
more-or-less constant.


        Stefan