OpenEuphoria

1. length() of an atom

• Posted by DerekParnell (admin) Jul 26, 2010
• 1793 views

I'm proposing that getting the length() of an atom should return 1. Currently it crashes the program.

This would simplify some algorithms because it avoids having to test for atom/sequence before taking the length. In those cases, we just use 1 anyway for atoms.

My thinking is that if appending an atom to a sequence increases the sequence length by 1, then it must follow that the length of an atom is 1.

Can anyone give me some convincing reasons to not make this change for Eu4?

3. Re: length() of an atom

• Posted by DerekParnell (admin) Jul 26, 2010
• 1770 views

? equal(length(25),length({25})) 
 

Yes. The 'it' being that they have the same length.

I have no idea what you mean by that statement.

5. Re: length() of an atom

• Posted by _tom (admin) Jul 26, 2010
• 1779 views

This could "shake up" the language since it goes to a fundamental design from the origins of Euphoria. (Whether this is good or bad, is another question.)

Some initial ideas that came to mind. (But I need more coffee before I fully understand how changing length() will alter the language.)

a fundamental Euphoria building block

 length( 23 )      ---> undefined 
 length( { 23 } ) ---> 1 

fundamental to the definition of atom vs sequence

will patterns be always true

must follow "patterns"

length( { 5, 6 } ) ---> 2 
length( { 5 } )    ---> 1 
length( {} )       ---> 0 
 
length( 5 )        ----> undefined 

length( 5 ) ==> 1 does not fit any "pattern", and is therefore out of place

a strength of Euphoria is that these kinds of patterns are always followed

'true length' is used in many routines

many Euphoria operations depend on objects being "the same length"

simple meaning of $

seq( 1 .. $ )   vs.  seq( 1 .. length(seq) ) 

the "$" has a simple meaning

• where can this be used?

is this of value in programming?

seq( 1 .. length(5) ) 

automagic dangers

automagic conversions are common in hard to use languages

• automagic not a Euphoria "feature"
• Euphoria simplicity is due in part because there are no coersions, automatic type changes, overloaded operators and routines (we like simple !)
• introducing automagic features would change the entire character of the language
  

"behaves as if length"

from the documentation: "atoms are of length 1 for this purpose"

• we do have idea that atoms have a length dimension "of sorts"

the "behaves" may seem special to someone new to Euphoria, but the design patterns are consistent

{1, 2, 3 } & 9999   ---- > { 1, 2, 3, 9999  } 

the 9999 "behaves" as if it "should" have a length of 1 but, that is a that appears in the future and does not exist in the present

question about 'internals'

• the description of length() in the documentation has comment that the length is stored internally for fast access

• will an atom need to store data that it has a length of 1 ?
  
  

  ---

is the solution a new routine ?

• lets call it "size()"

length( 2 ) -- > undefined 
size( 2 ) ---> 1 

this way length() and size() are closely related, but sensitive to the difference between atom and sequence

even is not adopted as a routine, would the documentation be improved if "size" was used as a descriptor for some Euphoria operations instead of the way "length" is used now?

7. Re: length() of an atom

• Posted by mattlewis (admin) Jul 26, 2010
• 1738 views

This is not strictly true. The implementation of atoms has a bit of automagic in it. However, it's fairly tame compared to what some languages do.

While I haven't seen Derek's solution, I'm sure the answer is to replace the current call to error handling with a hardcoded return 1; or similar. That point in the documentation is to make clear that length() is always O(1), and doesn't do something like count the number of elements every time you call it.

length( 2 ) -- > undefined 
size( 2 ) ---> 1 

This strikes me as an interesting alternative, since it leaves existing code unaffected, but allows the optimization that lead to this proposal.

Matt

9. Re: length() of an atom

• Posted by DerekParnell (admin) Jul 28, 2010
• 1665 views

Really? "shake up"? Isn't that a bit over dramatic? It's effect is microscopic. The only people affected by this idea are those who rely on crashing an application when getting the length of an atom - and that has got to be a tiny, tiny, number of Euphoria developers.

And I'd argue if one is doing that then they really have a bad approach to the language and code design in general.

 length( 23 )      ---> undefined 
 length( { 23 } ) ---> 1 

How is it a fundamental difference between atoms and sequences when the language already deems that atoms have a length of one when it comes to some operations?

The "length()" function is only "entrenched" with its current semantics because that's what Robert defined for it. It is no more "entrenched" than other changes we have made from v3 to v4.

It's as useful for finding mistakes in your code as a divide by zero is. Both situations should be handled more gracefully than relying on these to crash the application.

length( { 5, 6 } ) ---> 2 
length( { 5 } )    ---> 1 
length( {} )       ---> 0 
 
length( 5 )        ----> undefined 

Why is length( 5 ) - part of a pattern? What pattern?

upper( { 65, 65 } ) ---> {97, 97} 
upper( { 65 } )     ---> {97} 
upper( {} )         ---> {} 
upper( 65 )         ---> 97 

Hmmm... not much of pattern there.

Huh? "true length" as opposed to what? "false length"?

Sure, there are Euphoria operations that depend on sequences being the same length, but these operations are by their nature SEQUENCE operations and atoms have nothing to do with it.

eg.    {1,2,3} + {4,5,6} --> {5,7,9} 

Or are you suggesting that if someone writes a routine that takes 'object' parameters and uses the length() function on those parameters, and relies on it crashing if the parameter is an atom? This is a bad design. If the parameters must be sequences, then either use 'sequence' argument signature or test for sequence() at run time.

seq( 1 .. $ )   vs.  seq( 1 .. length(seq) ) 

seq( 1 .. length(5) ) 

I don't get what you are saying. The '$' token only means 'length(X)' when it appears inside a subscript of the sequence X. Nothing about that changes if length(atom) is defined as 1.

The example ...

seq( 1 .. length(5) ) 

Why would someone do that? It would be the same as saying ...

seq( 1 .. 1 ) 

Actually, Euphoria currently "automagically" converts from integer to atom and visa versa all the time now without you having to do anything.

And how on Earth is defining length(atom) as 1 making things not simple? In fact, it actually simplifies some algorithms.

Euphoria already has overloaded operators. The '&' (concatenation) operator accepts different data types. The arithmetic operators accept different data types, etc ... The puts() function accepts different data types. I can go on....

We already have overloading in Euphoria. "automagic" already happens in Euphoria - so this change will not be introducing it.

{1, 2, 3 } & 9999   ---- > { 1, 2, 3, 9999  } 

So ...? That actually sounds like a very good reason to formalize the 'apparent' length of an atom.

Not at all. I've have implemented this idea in my local copy of Euphoria about a week ago and everything still works fine. Currently, Euphoria already tests the operand on length() to see if its a sequence and if so, fetch the internal length data but if not it crashes. My changes just replaces the 'crash' with a 'return 1' instead. And with the translator, some 'length()' expressions can get translated as a literal 1 without any runtime overhead.

length( 2 ) -- > undefined 
size( 2 ) ---> 1 

I also thought this but it seems contrived and adds another predefined (built-in) function when an existing function can do the job without breaking code or really causing anyone grief.

In summary, I'm not convinced that length(atom) = 1 is a bad idea yet.

11. Re: length() of an atom

• Posted by jimcbrown (admin) Jul 28, 2010
• 1664 views

I guess doing abs(length(x)) would be a compromise, but it'd be a pain to have to include std/math.e for what should be a builtin operation...

If you really needed old code to crash, you could always do this:

function length(object x) 
	if not sequence(x) then 
		crash("TODO: add eu-lint") 
	end if 
	return eu:length(x) 
end function 

13. Re: length() of an atom

• Posted by jimcbrown (admin) Jul 28, 2010
• 1625 views

   if sequence( x ) then 

   if length( x ) then 

   y = length( x ) 

But as this is already the case anyways, to avoid crashing, we lose nothing here.

function length(object x) 
	if not sequence(x) then 
		crash("TODO: add eu-lint") 
	end if 
	return eu:length(x) 
end function 

Why do you need the code to _CRASH_ ?

for t=1 to length( ftxt ) do 
   if ftxt[t] = '#' then 
   ... 

It will still _CRASH_ ...

I fully agree with you here. From a theoretical viewpoint, a length should only apply to sequences (and similiar objects such as linked lists, arrays, etc).

But you put it best below...

Agreed, and I feel the use as a programmer's helper here significantly outweighs the desire to adhere to theoretical dogma.

To borrow Jeremy's example, a function that takes an Euphoria object and serializes it into a memory block for use by a C function.

15. Re: length() of an atom

• Posted by ArthurCrump Jul 28, 2010
• 1630 views

If you want a function which returns the length of a sequence or 1 if the parameter is an atom it is possible to use:

result = length( {} & X )   -- or: 
result = length( X & {} ) 
 

Probably inefficient, but if an atom is concatenated to a sequence, it extends the sequence by 1 element, so an atom could be said to have a natural length of 1.
Also, if an empty sequence is concatenated to an object X, it leaves it unaltered if X is a sequence but would convert an atom into a sequence with one element.
Several of the standard routines which accept an atom or a sequence begin by converting the atom to a sequence with one element.
If length(Atom) did return an integer, the only existing programs which would be affected are those which fail.

By now you will have probably gathered that I am in favour of the proposal.

I suppose it would be possible to define a second parameter for the length function to specify the value to be returned for an atom, but how could the default be specified to cause a failure if it is used?

17. Re: length() of an atom

• Posted by euphoric (admin) Jul 28, 2010
• 1625 views

{x,y} has 2 elements involved 
{x}   has 1 element involved 
{}    has zero elements involved 
x     has 1 element involved 

Elements only exist in the context of a sequence. No sequence, no elements. That last 'x' is not an element.

Regardless, I don't think this change will hurt much, if at all. So have at it. :)

But I'm curious, how does returning 1 for length(atom) make coding simpler and more user-friendly? What case did you discover works better with length(atom) returning 1?

19. Re: length() of an atom

• Posted by mattlewis (admin) Jul 28, 2010
• 1653 views

   --- 
   r = x[1 .. i] & q & [i+1 .. $] 
   n += length(q) 
   --- 

You should be using splice():

  r = splice( x, q, i + 1 ) 

Especially in a loop, this makes a huge difference in performance (likewise for replace(), insert() and remove()).

Matt

21. Re: length() of an atom

• Posted by petelomax Sep 23, 2010
• 1642 views

That really didn't make a whole lot of sense to me!

Anyway, as soon as this thread reappeared, I had two thoughts:

To keep things in order, length(atom) should return {}.

Well, it made me chuckle anyway, of course I'm joking. However, the other solution is to code (the equivalent of)

function length(object x, object default_result={}) 
  if sequence(x) then 
    return eu:length(x) 
  elsif atom(default_result) then 
    return default_result 
  else 
    crash_and_burn() 
  end if 
end function 

Then of course you can simply code length(5,1) to get your 1 rather than crash, and no legacy code whatsoever will be affected at all (and I hope not even performancewise), nor will it affect anyone that disagrees with changing length().

Regards, Pete

23. Re: length() of an atom

• Posted by jimcbrown (admin) Sep 23, 2010
• 1532 views

I have quite a few of those ... but I can easily rewrite almost all of them to use crash() in std/error.e instead.

I have a really obscure application that actually relies on this behavior (just one), but it's really old and won't run under Euphoria 2.3 .. I'd hate to see a new feature held back because of just one single application.

25. Re: length() of an atom

• Posted by DerekParnell (admin) Sep 23, 2010
• 1452 views

It has been in there since rev 3326, July 29th. About 8 weeks ago.