Collation API not complete for search

Shawn Steele Shawn.Steele at
Mon Mar 28 14:31:39 PDT 2011

My initial thinking was that “type” would indicate a more restrictive or a fuzzy matching, with the idea that when known strings are sorted (like records from a database being sorted for display), a detailed sort is appropriate, however when querying the database to see if there’s a record for “nebojsa”, it makes sense in many languages to also find “Nebojša”.

That’s not “string search”, it’s searching complete strings within the database record to see if it matches another complete string, and that’s what my original idea was.

Searching within a string is a more complicated problem, but, likely, fuzzy and exact ideas are still useful.

- Shawn

From: Nebojša Ćirić [mailto:cira at]
Sent: Monday, March 28, 2011 2:26 PM
To: Phillips, Addison
Cc: Mark Davis ☕; es-discuss at; Shawn Steele
Subject: Re: Collation API not complete for search

How do you solve search like this in offline mode?

- Text is in Serbian and contains word "šnala"
- User searches for "snala" since he can't input š easily (android phone or nook or kindle keyboards).

Without StringSearch you won't get a match...

I do agree this may complicate things for now and if we decide to postpone it we should also remove collationType option from the collator since it's pretty useless on its own.

28. март 2011. 14.04, Phillips, Addison <addison at<mailto:addison at>> је написао/ла:
This discussion has had me pretty confused. I never understood why you would *want* string search inside collator: the APIs and usage models are completely different. While there is some underlying relation, it’s just confusing to try to jam them into the same API.

StringSearch is modestly useful, but really I don’t see it as a particularly high priority for us.


From: Nebojša Ćirić [mailto:cira at<mailto:cira at>]
Sent: Monday, March 28, 2011 1:36 PM
To: Mark Davis ☕
Cc: es-discuss at<mailto:es-discuss at>; Shawn Steele; Phillips, Addison
Subject: Re: Collation API not complete for search

Shawn, would you be ok with adding this new API to the list for 0.5 so we can support collation search?

I'll edit the strawman in case nobody objects to this addition.
25. март 2011. 16.34, Nebojša Ćirić <cira at<mailto:cira at>> је написао/ла:
In that case I wouldn't put this new functionality in the Collator object. A new StringSearch or StringIterator object would make more sense:

options = {
  collator[optional - default, collatorType=search],
LocaleInfo.StringIterator = function(options) {}
LocaleInfo.StringIterator.prototype.first = function() { find first occurrence} = function() { get me next occurrence of pattern in source}
LocaleInfo.StringIterator.prototype.matchLength = function() { length of the match }
... (reset, setPosition...)
25. март 2011. 15.14, Mark Davis ☕ <mark at<mailto:mark at>> је написао/ла:

I think an iterator is a cleaner interface; we were just trying to minimize new API.

In general, collation is context sensitive, so searching on substrings isn't a good idea. You want to search from a location, but have the rest of the text available to you.

For the iterator, you would need to be able to reset to a location, but the context beforehand could affect what happens.


— Il meglio è l’inimico del bene —

On Fri, Mar 25, 2011 at 14:22, Mike Samuel <mikesamuel at<mailto:mikesamuel at>> wrote:
2011/3/25 Mike Samuel <mikesamuel at<mailto:mikesamuel at>>:
> 2011/3/25 Nebojša Ćirić <cira at<mailto:cira at>>:
>> find method wouldn't return boolean but an array of two values:
> Sorry if I wasn't clear.  The !! at the beginning of the call to find
> is important.
> The undefined value you mentioned below as possible no match result is
> falsey because !!undefined === false.
>> myCollator.find('gaard', 'ard', 2) -> [2, 5]  // 4 or 5 as a bound
>> myCollator.find('ard', 'ard', 0) -> [0, 3]  // 2 or 3 as a bound
>> I guess [2, 5] !== [0, 3]
> True, but also [2, 5] !== [2, 5].
>> We could return [-1, undefined] for not found state, or just undefined.
>> I agree that returning a boolean makes for easier tests in loops.
>> 25. март 2011. 14.00, Mike Samuel <mikesamuel at<mailto:mikesamuel at>> је написао/ла:
>>> 2011/3/25 Nebojša Ćirić <cira at<mailto:cira at>>:
>>> > Looking through the notes from the meeting I also found some problems
>>> > with
>>> > the collator. We did specify the collatorType: search, but we didn't
>>> > offer a
>>> > function that would make use of it. Mark and I are thinking about:
>>> > /**
>>> >  * string - string to search over.
>>> >  * substring - string to look for in "string"
>>> >  * index - start search from index
>>> >  * @return {Array} [first, last] - first is index of the match or -1,
>>> > last
>>> > is end of the match or undefined.
>>> >  */
>>> > LocaleInfo.Collator.prototype.find(string, substring, index)
>>> > We could also opt for iterator solution where we keep the state.
>>> Assuming find returns a falsey value when nothing is found, is it the
>>> case that for all (string, index) pairs,
>>> !!myCollator.find(string, substring, index) ===
>>> !!myCollator.find(string.substring(index), substring, 0)
Maybe a better way to phrase this relation is

will any collator ever look at a code-unit to the left of index when
trying to determine whether there is a match at or after index?

E.g. if the code-unit at index might be a strict suffix of a substring
that could be represented as a one codepoint ligature.

>>> This would be false if the substring 'ard' should be found in 'gard',
>>> but not 'gaard' because then
>>>     !!myCollator.find('gaard', 'ard', 2) !== !!myCollator.find('ard',
>>> 'ard', 0)
>>> If that relation does not hold, then exposing find as an iterator
>>> might help prevent a profusion of subtly wrong loops.
>>> > The reason we need to return both begin and end part of the found string
>>> > is:
>>> > Look for gaard and we find gård - which may be equivalent in Danish, but
>>> > substring lengths don't match (5 vs. 4) so we need to tell user the next
>>> > index position.
>>> > The other problem Jungshik found is that there is a combinatorial
>>> > explosion
>>> > with all ignoreXXX options we defined. My proposal is to define only N
>>> > that
>>> > make sense (and can be supported by all implementors) and fall back the
>>> > rest
>>> > to some predefined default.
>>> > --
>>> > Nebojša Ćirić
>>> >
>>> > _______________________________________________
>>> > es-discuss mailing list
>>> > es-discuss at<mailto:es-discuss at>
>>> >
>>> >
>>> >
>> --
>> Nebojša Ćirić
es-discuss mailing list
es-discuss at<mailto:es-discuss at>

Nebojša Ćirić

Nebojša Ćirić

Nebojša Ćirić
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <>

More information about the es-discuss mailing list