OpenEuphoria

1. Standard library praise and challenge

• Posted by jeremy (admin) Sep 11, 2011
• 3471 views

Just some standard library praise. I was recently working with a library retrieved from the archive that was written pre 4.x. I'm not going to mention the library or author. Anyway, I was working on taking advantage of some of the standard library routines instead of the ones written in the library that accomplish the same task. I ran across ltrim, rtrim and trim. Those methods trim whitespace from the left, right or both sides of a sequence. In the standard library we have trim_head, trim_tail and trim which accomplish the same task, except that they will trim anything you want by using an optional second parameter trim(content, what = " \t\r\n") basically. Anyway... the ones written in the library I was working on were pretty simple and looked fine to me at first glance. However, one benchmark I had setup was parsing a 25mb file using the library.

Stock from the archive the library took 21 seconds to complete. Simply removing the local implementations of ltrim, rtrim and trim and using trim_head, trim_tail and trim instead from the standard library parsing time dropped to 16 seconds. That's a 24% increase in performance by a simple search/replace operation using the new standard library.

This is what the new standard libary is all about. A large collection of group maintained/peer audited/performance tested routines. Not only do we get a large number of common functions to use instead of having each programmer reinvent the wheel with each new project, we also get everyone's head wrapped around a single problem and the very best solution possible.

So... this was just a bit of praise about the standard library but also a challenge to Euphoria users. Look at the source for your favorite standard library routines. Do you see a way they can be improved? Do you see a way to make them faster? Do you see a way to make them more robust? If so, create a ticket and let the development team know! 2 eyes are better than 1, 100 eyes are better than 50.

Jeremy

3. Re: Standard library praise and challenge

• Posted by DerekParnell (admin) Sep 11, 2011
• 3393 views

They do!? Doesn't sound like a common requirement to me. Now that we have optional arguments, it might be better to have recursive behavior controlled by an argument with the default being NOT to recurse.

5. Re: Standard library praise and challenge

• Posted by useless Sep 13, 2011
• 3297 views

I'd say max({"Head","Tail"}) should give you a word, max("HeadTail") should give you a letter.

Likewise, max( { {"2,Tail"} , {"1,Head" } } ) should return a sequence.

I wouldn't bet on that.

useless

7. Re: Standard library praise and challenge

• Posted by mattlewis (admin) Sep 13, 2011
• 3284 views

I'm not sure, although that's not really what people are asking about. Let's look at max:

public function max(object a) 
	atom b, c 
	if atom(a) then 
		return a 
	end if 
	b = mathcons:MINF 
	for i = 1 to length(a) do 
		c = max(a[i]) 
		if c > b then 
			b = c 
		end if 
	end for 
	return b 
end function 

I can see what this is getting at. We're in math.e, and the intent is to deal with atoms only. From that perspective it makes sense, although I wonder how often it would really be used for that purpose.

An alternate is to add a more simple, but intuitive, routine, possibly in sequence.e. This other routine would simply do a compare() against each element, starting with the first. I guess the default return would probably be an empty sequence for that routine.

Matt

9. Re: Standard library praise and challenge

• Posted by DerekParnell (admin) Sep 13, 2011
• 3286 views

I give you an empty bag and ask you to pull out the largest object. You can't pull out any object so therefore this is an error. The results are defined - you pull nothing out, which is of course neither the biggest nor the smallest, thus we have a problem - also known as an error situation. One way to signal this error is to just give back the bag.

It's clear that math:max() is intended to only work with numbers, but I think we need max/min functions that work with objects too.

public function max(object a) 
    object b,c 
 
    if atom(a) then 
        return a 
    end if 
 
    if length(a) = 0 then 
        return a 
    end if 
 
    b = a[1] 
    for i = 2 to length(a) do 
        c = a[i] 
        if compare(c, b) > 0 then 
            b = c 
        end if 
    end for 
 
    return b 
end function 

11. Re: Standard library praise and challenge

• Posted by bill Sep 14, 2011
• 3254 views

I don't see max or min of null ({}) being an error.

The max is as the min, {} (null).

The max and min of { {}, {'a'} } should be {'a'} as null cannot be less than or greater than another value.

- as in SQL.

13. Re: Standard library praise and challenge

• Posted by DerekParnell (admin) Sep 14, 2011
• 3211 views

For maths library, maybe this make sense (I'm not fully convinced of that yet, by the way), but it not always the intuitive result. For example, given a list that contains a mixture of numbers and strings, return the sum of that list. Most spreadsheet programs ignore non-numbers in the summing range, so maybe that's what people expect to happen.

15. Re: Standard library praise and challenge

• Posted by jimcbrown (admin) Sep 14, 2011
• 3199 views

Same here.

Yep, except for the small hitch that Euphoria has no object or singleton that can unambiguously represent 'nothing' or the NULL concept in general.

max({}) -- {} 
min({}) -- {} 

Like Derek said, the empty sequence is not nothing. But he also said

And I, while wishing to clarify that the empty sequence is not nothing, agree with both Jeremy and Derek on this.

Of course, this is not perfect.

max({{}, {}}) -- returns {} also 

But that's close enough for me. If it really matters, I'd just do an "is_empty" check on the original sequence after calling max()/min().

17. Re: Standard library praise and challenge

• Posted by jimcbrown (admin) Sep 14, 2011
• 3220 views

If no one else beats me to it, I'll review it and probably apply the patch sometime today.

19. Re: Standard library praise and challenge

• Posted by jimcbrown (admin) Sep 14, 2011
• 3186 views

Which is not the case for the (proposed) max() and min() in std/sequence.e

This is not inconsistent. The current implementation of max() and min() in std/math.e do not return a value representing NULL - they always return a valid atom.

Should they return NULL? Maybe. Currently, they don't.

21. Re: Standard library praise and challenge

• Posted by mattlewis (admin) Sep 14, 2011
• 3193 views

Actually, it can be tested! Use the object type. It might be good to allow euphoria code to explicitly use this. Also, I noticed that we're missing NaN in our math constants.

Matt

23. Re: Standard library praise and challenge

• Posted by jimcbrown (admin) Sep 14, 2011
• 3188 views

We don't have "NULL sequences", whatever those are. We only have empty sequences.

This works for me. I can live with math.e's max()/min() returning NaN when given {} as input.

25. Re: Standard library praise and challenge

• Posted by jimcbrown (admin) Sep 14, 2011
• 3171 views

I'm a bit confused. We might not have a constant in std/math.e or elsewhere, but we can do it:

constant NaN=-(2*power(10,308)/(3*power(10,308))) 

27. Re: Standard library praise and challenge

• Posted by mattlewis (admin) Sep 14, 2011
• 3157 views

-- nan.ex 
constant NaN=-(2*power(10,308)/(3*power(10,308))) 
? NaN 

It's not working!

$ eui nan.ex 
-0.6666666667 

Matt

29. Re: Standard library praise and challenge

• Posted by mattlewis (admin) Sep 14, 2011
• 3162 views

-- nan.ex 
constant NaN=-(2*power(10,308)/(3*power(10,308))) 
? NaN 

$ eui nan.ex 
-0.6666666667 

Sorry, that was my snarky way of saying that it's not a portable way to get a NaN. We use bigger floating point numbers for 64-bit euphoria, and 10e308 is a lot smaller than 1.19e4932.

Matt

31. Re: Standard library praise and challenge

• Posted by mattlewis (admin) Sep 14, 2011
• 3184 views

-- nan.ex 
constant NaN=(2*power(10,9008)/(3*power(10,9008))) 
? NaN 

That gives me -nan.

Matt

33. Re: Standard library praise and challenge

• Posted by jimcbrown (admin) Sep 14, 2011
• 3191 views

-- nan.ex 
constant NaN=(2*power(10,9008)/(3*power(10,9008))) 
? NaN 

Ok, I'm stumped.

35. Re: Standard library praise and challenge

• Posted by jimcbrown (admin) Sep 14, 2011
• 3169 views

-- nan.ex 
constant NaN=(2*power(10,9008)/(3*power(10,9008))) 
? NaN 

{0,0,0,0,0,0,0,192,255,255} 

{0,0,0,0,0,0,0,192,255,127} 

I understand the difference, but I'm still clueless on how to create positive NaN. Surely it's not as simple as multiplying by -1 ?

37. Re: Standard library praise and challenge

• Posted by jimcbrown (admin) Sep 14, 2011
• 3145 views

I was looking for a quick series of mathematical functions that'd work as-in in 4.0 and 4.1 that would do the job, to prove a point.

(sqrt(-1) does not work in 4.0, btw).

Agreed. Above all, I was trying to avoid this:

Anyways, back to my original point: NaN and Nothing are conceptually separate values. NaN is a floating point number with a specific set of representations that are defined as the value of certain floating point operations (such as sqrt(-1) or fdiv(inf, inf)).

While it may make sense for NaN to be the output of std/math.e's min() and max() and sum() and perhaps even or_all() when they are given an empty sequence as input, NaN does not work the same way as our NOVALUE (or even Python's Nothing object).

39. Re: Standard library praise and challenge

• Posted by DerekParnell (admin) Sep 14, 2011
• 3129 views

Does this work for you guys?

include std/convert.e 
constant NaN= float64_to_atom({0,0,0,0,0,192,255,127}) 
? NaN 
? 3 + NaN 
 
function tn(object a) 
  return a * 2 
end function 
? tn(NaN) 

41. Re: Standard library praise and challenge

• Posted by jeremy (admin) Sep 14, 2011
• 3107 views

include std/convert.e 
constant NaN= float64_to_atom({0,0,0,0,0,192,255,127}) 
? NaN 
? 3 + NaN 
 
function tn(object a) 
  return a * 2 
end function 
? tn(NaN) 

I get nan's.

C:\Development\Projects>eui test.ex 
nan 
nan 
nan 
 
C:\Development\Projects>eui -v 
Euphoria Interpreter v4.1.0 development 
   32-bit Windows, Using System Memory 
   Revision Date: 2011-09-09 18:29:16, Id: 5145:0da3aad80586 

43. Re: Standard library praise and challenge

• Posted by SDPringle Sep 15, 2011
• 3059 views

It would surprise me that a programmer without enough forethought to check to see if a sequence has no length (or ensure it always has some length) before calling min() or max() will have forethought to check for an error code. In the case where a programmer is not diligent, whatever that error code is for passing {}, be it, the bag({}), NaN, PINF, NINF, or "WTF", will be used in the next calculation as if everything is till okay.

Most recently, I was interested in what the maximum tension on a joint inside of a truss system for choosing the kind of joint one should use. The error value would be used to determine what joint type to use. Logical errors propagate through the code until the program says that the weakest joint is okay to use.

This following is an abbreviation of what I wrote for my private library and I use it myself. If you don't want to check for errors you turn type checking off. max and min assume that s[1] exists. There is no need to return nan, PINF, NINF, or "".

type nonemptysequence(sequence s) 
    return length(s) != 0 
end type 
 
function max(nonemptysequence s) 
... 
 
function min(nonemptysequence s) 
... 

Shawn Pringle

45. Re: Standard library praise and challenge

• Posted by petelomax Sep 16, 2011
• 2998 views

Of course I'm still smarting that max(sequence s) exists at all.

It should be max_sq(sequence s) and max(atom,atom).

Pete

47. Re: Standard library praise and challenge

• Posted by mattlewis (admin) Sep 16, 2011
• 2976 views

Or...

public function max( object a, object b = a ) 
    if atom(a) then 
        if a >= b then 
            return a 
        else 
            return b 
        end if 
    end if 
    b = mathcons:MINF 
    atom c 
    for i = 1 to length(a) do 
        c = max(a[i]) 
        if c > b then 
            b = c 
    end if 
    end for 
    return b 
end function 

Matt

49. Re: Standard library praise and challenge

• Posted by mattlewis (admin) Sep 16, 2011
• 2966 views

? max("First", "Second") 

Under that, you'd get the max of "First". I would have documented that the function was to be used in the ways that Pete was looking for, which make sense as a math function, to me. Your use appears to be looking for the general object comparison, which I think we were saying should be somewhere else.

Matt

51. Re: Standard library praise and challenge

• Posted by mattlewis (admin) Sep 16, 2011
• 2984 views

I thought so! But I couldn't remember those names. I don't spend enough time in the stdlib. I was looking for greatest/greater/bigger.

Matt

53. Re: Standard library praise and challenge

• Posted by SDPringle Sep 18, 2011
• 2936 views

There seems to be very divergent opinions on what should happen with max

Mine are:

max(atom) => typecheck error 
max({})   => typecheck error 
max({4,32,2}) => 32 
max({"Ax",'a'}) => "Ax" 
max({"Ursula","Susan","Barbara"}) => "Ursula" 

Similar results with min. sum and product should work as if you string the members with separators + and * respectively. or_all should work as if you run or_bits(..,or_bits(..., ... ) on the members. The members of a sequence s are defined as s[1], s[2], ... , s[$]. Fortunately, we have namespaces and we can roll our own function definitions.

Shawn Pringle

55. Re: Standard library praise and challenge

• Posted by bill Sep 18, 2011
• 2957 views

Re

type nonemptysequence(sequence s)  
    return length(s) != 0  
end type  
  
function max(nonemptysequence s)  
...  
  
function min(nonemptysequence s)  
... 

no good

nonemptysequence S = { {} } 
 
length(S) is 1 
 
max(S) is -inf 

57. Re: Standard library praise and challenge

• Posted by SDPringle Sep 18, 2011
• 2906 views

I want to clarify that the special cases for max and min that I described here were not meant to be descriptions of the standard library functions max and min. Rather, what functions with these names in my private library return and what they should return according to my philosophy.

The standard library returns some value for max{} and min{}. It flattens its argument. Meaning max({{24,4},{400}}) is the same as max({24,4,400}) in the standard library.

The ones I developed error out when passed an atom or an empty sequence. It doesn't flatten its argument. You can find a maximum or minimum object of a sequence of elements. The arguments may all be strings or matrices.

My max and min were developed separately before I joined this project and the existing max and mins in the standard library were already there when I joined. AFAIK.

Shawn Pringle

59. Re: Standard library praise and challenge

• Posted by SDPringle Sep 20, 2011
• 2790 views

I think Bill, Jim, Derek and I are in agreement of what max() and conversely min() should do when the list is not empty. When the list is empty is result is rather academic from my point of view.

Condier though:

1. 1. EUPHORIA's routines tend to die rather than give an error code when given bad parameters. Let's be consistent here. In general I prefer to see an error code but there is no way to tell an error code from a result here.
   2. Certainly, the idea of what a maximum of the empty set is undefined.
   3. It follows that the interpreter should die in this case.

Consider division by 0. There is no NaN returned. It dies, instead. This is the way the work is designed. Instead of going off into different directions we should try to keep with the same design principles.

Shawn Pringle

61. Re: Standard library praise and challenge

• Posted by jimcbrown (admin) Sep 20, 2011
• 2774 views

result = { [NOT] SUCCESS, 23 } -- one way to do it 

I am not in favor of the radical change in the type of return value.

Consider this:

The input to value() could come from anywhere. When value() returned an error code, that signalled a problem but the code could still continue. E.g. if you use value() to get a number but the user types "hello world" instead, you can prompt the user to reenter the number.

This sort of error is recoverable, and can not easily be validated by the calling code (at least, not without duplicating much of the functionality of value() itself).

The sole error with max()/min() can be checked for easily (a simple length check and possibly a type check to make sure it's a sequence) - and as Matt and Shawn and others have pointed out, it's an error that usually signifies a greater problem in the calling code - one that's probably irrecoverable beyond closing all open files, showing the appropriate prompt to the user, and then dying gracefully.

Also, this is a very minor point, but consider anyways:

value() has multiple error codes (well, 2 - failure and eof) whereas we only have the one for max()/min()

63. Re: Standard library praise and challenge

• Posted by bill Sep 21, 2011
• 2647 views

The only reason why I would object to dying or error on max({}) is:

One is returning the maximum of a list:
An empty list is a valid list,
{} is (potentially) a member of every list.
MAXREAL < {} < {MINREAL} i.e. it has an ordering whereas sqrt(-1) does not.

I think it would be advisable to avoid a situation where max({}) dies or raises an error but max({ {} }) and max({ {},{} }) do not.

Jim: remember I am not returning an atom.

65. Re: Standard library praise and challenge

• Posted by bill Sep 22, 2011
• 2628 views

Objects:

you said:

Agreed. Even sqrt(-1) dies instead of returning NaN, despite  
the latter behavior being explicitly allowed by the floating  
point standard. 
 
I think this is also keeping with the design principle of  
"Fail Early." 

Sqrt(-1) should not return NaN because i is not anything like a real or a sequence. There cannot be an order for complex numbers unless you define 2 numbers as being equal if the distance from (0,0) (or (0,0,0) ... etc is the same) (metric spaces).

If max({}) fails what of max({ {} }). Are you say max() can only return a number?

Bill

67. Re: Standard library praise and challenge

• Posted by jimcbrown (admin) Sep 22, 2011
• 2606 views

Agreed. Even sqrt(-1) dies instead of returning NaN, despite  
the latter behavior being explicitly allowed by the floating  
point standard. 
 
I think this is also keeping with the design principle of  
"Fail Early." 

You may disagree, but this is specified in the floating point standard. After all, NaN does not specify a real number - it literally stands for Not A Number.

I suppose we could just return {} from sqrt(-1). This is clearly not a number, although the return value could be confused with the result of sqrt({}).

max({ {} }) returns {} but max({}) fails. max({ {}, {} }) returns {}.

69. Re: Standard library praise and challenge

• Posted by SDPringle Sep 23, 2011
• 2472 views

Its posts like these that make me wish we had like buttons for the forum. :)

Shawn

71. Re: Standard library praise and challenge

• Posted by DerekParnell (admin) Mar 01, 2012
• 2057 views

stdsort:sort(append(haystack, needle))

I wouldn't like to replace the existing functionality but we could certainly add to it. Maybe a new code INSERT_SORTED that would insert the new item into an already sorted list.