What the heck? Contents of objects NOT alphabetical!??

Okay. I think I like your sorting algorithm. I'll take two please.

I thought it was pretty cool... but thats just me

Yes, it's slower, I'm sure. It might add a whole millisecond every time you sort a folder in inventory. Major lag source.

As someone observed, Windows does the same thing with filenames. In fact, I'll bet the same piece of code is responsible for both: SL probably uses the locale-specific collating function provided by Windows (or MacOS). And they do it because this sort order makes sense to non-programmers: 10 is supposed to come after 9, not before 2; just like uppercase 'B' comes after lowercase 'a', regardless of what the ASCII chart says.

Anyway, once you're dealing in larger-than-ASCII character sets, and especially Unicode, naive ASCII-betical sorting is no longer an option. (Should 'c-with-cedilla' sort after 'z'? Wouldn't German users expect 'a-with-umlaut' to sort to the beginning of the list, rather than the end?) It's complicated, hard to get right, and harder still to make efficient; and if Microsoft has already done the work, why re-invent it when you have better things to do?

If it means sacrificing a millisecond here or there to come up with something that makes more sense to non-technical users, I'm all for it. And if it means LL re-using existing code provided by the OS, rather than spending the time and effort to roll their own, I'm all for that, too.

Optimization effort is better spent on any number of other areas: physics, rendering, the database architecture, the LSL VM, LoD filtering, etc. The string collating algorithm is a very small drop in a very big bucket.

no, it means much more, not only according to my benchmarks and working out what the actual algorithm does, but because of the pain in the ass it becomes for developers who have to deal with this if sorting items is any sort of priority.

I don't think it's slower, it's faster, imho. It's the most basic binary sorting routine without any additional code to refine it.
[0-9]<A < B < E < Z < a < b < e < z.

Apple
apple
..because A<a, it lists first. No other checks required!

Say we have 2 words to sort, "boo39ger" and "boo7ger". Doesn't matter which we work with first, so let's just say we've pulled "boo39er" out of memory and tossed it into our list. Then we snag "boo7ger" -- we need to check the list and see where the new word goes: (italics means the char we're checking from "boo39ger", normal means char from boo7ger)
b==b, they're equal, check next...
o==o, equal, next...
o==o, equal, next...
7>3, boo7ger goes after "boo39ger". Sorting DONE = fast (isn't really alphanumeric, it's a close dirty but FAST comparions/sorting routine).
Toss in "boo2939239ger", you start at the top (probably anyway, maybe the bottom, a cheap sort could break it in half or quarters to speed things up). Anyway we'll go from the bottom cuz I'm a geek.
b,o,o==b,o,o all equal, keep checking
2<7 this goes before 7, but now we have to compare to the next item, "boo39ger"
b,o,o==b,o,o equal...
2<3, So boo2939239ger goes before boo39ger. Again, cheap but fast sorting. No screwing around with other functions/comparisons.

This is about as basic (fast, albeit sorta crappy) as sorting gets. Creating a more accurate sorting routine would entail more code and more work. Even a basic bubble sort is much slower than what the Lindens are using (apparently to me anyway).

I do support it though, because the sorting occurs on the Servers, not by the client. Running better sorting routines is a waste of resources, imho, given the amount of junk in inventories and the amount of inventory fetching that goes on constantly.

Bos

Personally I find this arguement a fascinating oddity and an example of peoples preference for out-dated "traditions" that are perhaps illogical but acceted over a more logical approach.

In point of fact...the number '11' is indeed a higher numerical value than the number '2', and as such...from a logical sorting perspective...should always come AFTER '2'. That we currently follow an illogical method of sorting that does not recognize the number '11' as eleven at all, but rather treats the numbers as letters in a word wherin all 'words' starting with '1' would come before all 'words' starting with '2' does not mean this practice is correct or logical...or even preferred.

Personally, both as a user, and as a programmer, the accepted illlogical practice of dealing with numbers as letters in a string has always bugged me for its counter-intuitiveness. Interesting to see how vigorously people will defend a 'tradition'.

The issues of performance that may or may not be caused by the requirement of additional processing is not addressed as it is not relevant to the particular phenomenon I was wishing to discuss. Indeed it is an important issue, merely not applicaple to the sociological persepective.

Just my own opinion and observation. Take it as such, please.

BURN HER!!!





I agree with you.

I find myself in the unusual position of having to disagree with just about everything you said, Bos.
Neither of the two techniques being discussed do pure ASCII sorts. For the purposes of this discussion A==a, B==b, A<b and a<B. What is being discussed is whether:
"11" < "2" because "aa" < "b"
OR
"2" < "11" because 2 < 11

Remember not to confuse the complexity of the sorting algorithm (insertion sort as you first describes, or the bubble sort you mention) with the complexity of the comparison function, which is where alphabetic vs. alphanumeric ordering differ.

Also any bubble sort is much slower than just about any other algorithm out their, regardless of the comparison function.

Change the sort order of your inventory from by-name to by-date. Notice that it is virtually instant. There is no way that's being done server-side. All the sorting load (for inventory at least) is ditributed to the clients.


llGetUNIXTime (sorry, couldn't resist that one )

Does she float?
You know, like a duck?

disclaimer:
My rebuttal is in fun, and because I'm bored. Plus I think it's worth a little rebuttal anyway cuz this thread is a sorta silly topic anyway. I know some feed on drama, I don't mean to be the cook.


What argument?
I wouldn't call it illogical "out-dated 'traditions'". Lindens appear to be using the most simple and very common (and most limited, other than no sorting at all) solution to sorting inventories.


First, it's dealing with strings, not numerals. Regardless if it's "12" or "Kitty", or "kitty12" it is a string, aka the item's name.
Second, is "Joe34blow" a word, or 2 strings and a numeral. (or maybe a VB variant!). I call it a string because, well it is a string. Numerals don't consist of w,o,l,B,e,o, nor J.
I do concur that it's illogical that 1,11,2 is the given sort order, for "numerals", but I understand why it's sorting like this -- that's perfectly logical to me. Annoying, yes, but logical.


How else can you define numbers in strings -- when looking at the entire string. Sure you can pick it apart and call the alphabetical characters "strings" and the numerical characters "numbers" but all stuffed together as an item's name? I'd call that a string, or at worst a "variant".


I personally think performance is probably a valid reason we have crappy sorting. Maybe they plan to improve it, but even if so I bet it's WAY down on the ToDo list. 1. Works. 2. Ain't Broke. 3. Can't optimize it any further, except to remove all sorting == no fix/change coming for this for a long ass time (maybe after Havok2008...).
Not sure what you were wishing to discuss, I scrolled back and didn't see you discussing anything before. This confused me. Dunno how sociology came about either. I'm confused (which is probably obvious by any of my drivel).

Good stuff previous to this quote. :)


Topic is Object contents though, that's server side far as I'm aware. That's why updating anything in an objects contents is about as painful as pulling teeth.



haha yeah, about freaking time. Been waiting for this since '03 :p Think I'll keep my sig around for 3 more years just to match the time-frame of actually getting this function. :)

Wont this make it impossible to do a binary search through object inventory? (Or make it *really* hard and slow?)

Right now we can use llGetInventoryKey for an exact match on an object's name, but if we want to indroduce any variation in the results we get (searching for all items starting with "foo" for example)....
==Chris

Ah, yes, out-dated "traditions" like the ASCII table sorting, which is a global standard of computing?

This is a display for regular old every day humans who are more likely to want "human logic" sorts rather than straight up binary sorts. I really fail to see how this is any sort of problem.

As I said, unless it effects how we're sorting in our scripts (i.e., doing something other than the binary sort we expect it to do) why is this any brand of problem?

I think its going to destroy SL - kick my dog in the nuts - and make the baby jesus cry.

Is this the 'culling' of 1.9?

I've never understood why windows orders numbers as they do, i.e. 1, 10, 11, 2, 3, 30, 31, 4 and so on. For this reason I always name my files using zeros, i.e. 01, 02, 03, 04...10, 11...20, 21, 22 and so on.

As far as I've always known, the number 9 has always been followed by the numbers 10 and 11. The number 1 has always been followed by the numbers 2 and 3. I've never know the number 1 to be directly followed by numbers 10 or 11, so it doesn't make sense to me that any list would list numbers in this order.

starax, i don't know who fed you the happy juice for the forums, but i like it

I've noticed that too Starax's star turn in the forums has been quite amusing, and a refreshing change.

* can I get a rebate on my wand now please? *