dojo.widget.incrementalFinderDataProvider with "backoff" (prevents many requests)

[ialpert@…]

dojo.widget.incrementalComboBoxDataProvider = function(/*String*/ url, /*Number*/ limit, /*Number*/ timeout){

this.searchUrl = url; this.inFlight = false; this.activeRequest = null; this.allowCache = false; this.timeout = null; this.cache = {};

this.init = function(/*Widget*/ cbox){

this.searchUrl = cbox.dataUrl;

};

this.addToCache = function(/*String*/ keyword, /*Array*/ data){

if(this.allowCache){

this.cache[keyword] = data;

}

};

this.startSearch = function(/*String*/ searchStr, /*String*/ type, /*Boolean*/ ignoreLimit){

if(this.inFlight){

this.timeout = dojo.lang.setTimeout(this,this.startSearch,20,searchStr,type,ignoreLimit); return;

} if(this.timeout) {

dojo.lang.clearTimeout(this.timeout); this.timeout=null;

}

this.performSearch(searchStr, type, ignoreLimit);

};

this.performSearch = function(/*String*/ searchStr, /*String*/ type, /*Boolean*/ ignoreLimit){

var tss = encodeURIComponent(searchStr); var realUrl = dojo.string.substituteParams(this.searchUrl, {"searchString": tss}); var _this = this; var request = dojo.io.bind({

url: realUrl, method: "get", mimetype: "text/json", load: function(type, data, evt){

_this.inFlight = false; if(!dojo.lang.isArray(data)){

var arrData = []; for(var key in data){

arrData.push([data[key], key]);

} data = arrData;

} _this.addToCache(searchStr, data); _this.provideSearchResults(data);

}

}); this.inFlight = true;

};

};

[bill]

1. Have you filed a CLA (www.dojotoolkit.org/icla.txt)?
2. Please attach a diff file against the latest dojo 0.4.1RC code

[guest]

1. Yes i have -- faxed this morning (is there anything else i need to do WRT this?)
2. couldn't find tagged version (probably my svn ignorance) so i diffed against trunk (hope this is ok)

[alex]

a good chunk of this patch is whitespace noise. Is that intended? I'll merge it, but I'd prefer a cleaner patch.

[bill]

This isn't a patch against the latest. (Well technically it is, but...) He took an old version of the incrementalDataProvider and modified that, instead of modifying the newest version. So it basically lost all the recent changes. We need a patch against the newest version.

[Izaak Alpert]

Err true, i didn't write tests for it as well, doing so now.

[Izaak Alpert]

Allright I've rewritten the code against latest (i apologize -- no excuse for that last diff).

Couple of problems

1. there's still lots of whitespace in the diff file after the incrementalComboBoxDataProvider class which is the only thing i changed. Any suggestions as to how to prevent this (bit of an SVN newbie)?
2. I've tried writing a test for this however since there's no latency in file requests it's difficult to exhibit the request limiting code, any suggestions as to how to fake out latency?

[bill]

Hi Izaak,

Looks like the whitespace changes were actually things you fixed (IE, the original file had spurious whitespace). I checked in the whitespace changes as a separate checkin.

For testing, I'd suggest writing a php file. Although dojo is purely client side we can stick a few PHP files in the test directory as necessary.

As far as the algorithm goes, I'm not sure this makes sense. If I type the letter "a" and then type the letter "l", but it takes a long time to search for all words starting w/the letter "a", then eventually a drop down will be displayed with all the words that start with "a", even though I've already typed in "al". The current code will fire off request for "a" and "al" (as I type "a", and then "l"), and then just ignore the return value from the "a" request.

[Izaak Alpert]

There may be cases where if the servers request is fast enough that 20msec value should probably be changed (I originally was going to tie this into the server response time). This number seemed to work reasonably well where the server response time was around 0.5 seconds.

What I've been trying to achieve was if somebody types "abc" quickly the b request basically doesn't get sent. It sounds like the 20msec setting is too long? I'd like to show the user results if they hesitate, but prevent unnecessary requests...

Perhaps i should build a buffer and always only send after ever X msec (where X is sufficiently small that they don't feel like it's slow)?

WRT Testing Is there any documentation/examples as to set that up I can look at (I presume there's some kind of server that gets instantiated to serve up these php files).

[Izaak Alpert]

FYI i have had slightlyoff (from #dojo) have a look at it, seemed to think it was allright.

Also behavior where typing "a" then "l" and the "a" request getting ignored exists prior to patch.

[Izaak Alpert]

Ok here's the latest with a test... Using no timeout

[Izaak Alpert]

PHP file for testing request limiting functionality

[Izaak Alpert]

diff of test_ComboBox.html from dojo/tests/widgets

[Izaak Alpert]

ok please ignore last diff -- i've tested it out with different latiencies, i'm not convinced it works...

[Izaak Alpert]

Ok here's a timeout less and test that actually works...

[Izaak Alpert]

combo box with request prevention (no timeouts)

[Izaak Alpert]

PHP file for testing request limiting functionality

[Izaak Alpert]

test for combBox with request limiting behaviour

[Izaak Alpert]

Ok this one seems to avoid race conditions and the tests is reasonably configurable

[Izaak Alpert]

I have done some loose tests on this with the ComboBox? searchDelay set to 0 to see if it had an effect; it did not (beyond speeding up initial requests)

[bill]

Combobox/Select? have been completely rewritten; not sure if this is even an issue anymore. But we should have some kind of backoff to prevent too many requests going out. Assigning to Mark to consider.

[bill]

This ticket is really long and the patch is no longer valid because Combobox has been rewritten. Replaced with #4102.