#7748 returned in 1.3.0? dojo.clone fails in FF and not IE

[Phil Bowles]

dojo.clone fails in FF 3.0.8, works in IE 7.0.5730

Testcase is simple:

var newnode=new dijit.form.TextBox();
console.warn("newnode",newnode);
var clone=dojo.clone(newnode);
console.warn("clone",clone);

works 100% as per expected in IE7

in FF, fails thus:

e is undefined
fp()()/code/do.../event.js (line 33)
clone()(Object)/code/do...e/lang.js (line 257)
clone()(function())/code/do...e/lang.js (line 261)
clone()([input#dijit_form_TextBox_0.dijit, "onmouseenter", function(), 1 more... 0=input#dijit_form_TextBox_0.dijit 1=onmouseenter])/code/do...e/lang.js (line 243)
clone()([[input#dijit_form_TextBox_0.dijit, "onmouseenter", function(), 1 more... 0=input#dijit_form_TextBox_0.dijit 1=onmouseenter] 0=[4]])/code/do...e/lang.js (line 243)
clone()([[[input#dijit_form_TextBox_0.dijit, "onmouseenter", function(), 1 more... 0=input#dijit_form_TextBox_0.dijit 1=onmouseenter] 0=[4]], [[input#dijit_form_TextBox_0.dijit, "onmouseleave", function(), 1 more... 0=input#dijit_form_TextBox_0.dijit 1=onmouseleave] 0=[4]], [[input#dijit_form_TextBox_0.dijit, "oninput", function(), 1 more... 0=input#dijit_form_TextBox_0.dijit 1=oninput 3=1] 0=[4]] 0=[1] 1=[1] 2=[1]])/code/do...e/lang.js (line 243)
clone()([Widget dijit.form.TextBox, dijit_form_TextBox_0] _connects=[3] _deferredConnects=Object)

this looks to be the same code (without the webkit additions, which I don't understand) as trac #7748, i.e. its something to do with the special handling of mouseenter events in FF

..either way, I can't clone a node in FF without getting this error!

PS the line numbers in lang.js may be wrong - i put some console.debug output in my source installation to track this bug down

[dante]

you aren't cloning a node in your example, you are attempting to clone an instance of a dijit.form.TextBox? (complex object) ... I don't believe cloning widget instances is a supported behavior of dojo.clone ... you should be able to clone the .domNode no problem, but the events and whatnot aren't likely to be setup properly.

[Phil Bowles]

From the api docs: "Clones objects (including DOM nodes) and all children. Warning: do not clone cyclic structures." From the DojoCampus? docs: Part of Base Dojo (dojo.js) - Clones anything objects and/or nodes, returning a new anything.

I can't see how any of that states that widget cloning is not supported - quite the opposite in fact, as clone works exactly as expected (and documented) in IE...I get a clone of the textbox, which is what I wanted and what the docs say should happen.

There is definitely a problem with the code mentioned when running under FF.

Also not sure why the status had been changed...IMHO this is a blocker..code that works as per spec in IE fails completely in FF - I can't run the app

[dante]

Maybe I'm just missing something then ... but is this not the same situation as ​http://bugs.dojotoolkit.org/ticket/2456

clone() clones nodes, and value based objects. It does not do recursive/circular references at all (which a Widget instance is).

I suppose the documentation should be updated to reflect explicitly cloning of widget objects is not supported. One would expect cloning a widget reference to create a new instance of that reference, but that's simply not how it works (nor has it ever afaik). To do so would require Dojo knowing about Dijit, dojo.clone understanding dijit._Widget/dijit.registry and managing all that.

attached is a test case showing the cloning of a widget domNode reference. Even this introduces problems, as the id's assigned to the widgets by Dijit are cloned as well. When you focus the first [cloned] textbox node, the focus indication is set on the "original/real" instance.

Are you saying this is a regression, and dojo.clone used to clone widgets without problem? If that is the case, the severity should be reset, but I am under the impression this was never supported, and can be fixed by updating the documentation.

Perhaps a dijit.clone() API that would create new instances and handle the ID's/registry/etc would be a nice enhancement, but that kind of code doesn't belong in base.

[Adam Peller]

The culprit appears to be here near the bottom of dojo.clone:

	// Generic objects
	r = new o.constructor(); // specific to dojo.declare()'d classes!

it's cloning the _connect array which (on non IE browsers) has a connect handle with an argument in its constructor:

(function (e) {if (dojo.isFF <= 2) {try {e.relatedTarget.tagName;} catch (e2) {return;}}if (!dojo.isDescendant(e.relatedTarget, node)) {return ofp.call(this, e);}})

The clone code is designed to work for dojo.declare but won't work for objects which require arguments to the constructor. I'm not sure it's possible to clone such objects.

[Adam Peller]

the recursive problem, iirc, results in an infinite loop. We decided not to fix (#2486) Nodes are cloned by a special case so their recursive references aren't an issue, but if you did hit one, you'd know.

[James Burke]

The test page works for me now with latest 1.4 code, so closing. elazutkin did changes to clone in the 1.4 timeframe, so it may have been fixed with that.