[patch] [cla] DataChart output is screwed/courrupted with htmlLabels: false in 1.4.X

[Jared Jurkiewicz]

DataChart? output is screwed/corrupted with htmlLabels: false in 1.4.X

The same problem does not occur in 1.3.1.

Using a modified version of the DataChart? test produces it. Will attach shortly along with screenshot of damaged output:

[Jared Jurkiewicz]

Testcase

[Jared Jurkiewicz]

Testcase support file.

[Jared Jurkiewicz]

The same mangling occurs on IE7, so it's not SVG specific.

[Jared Jurkiewicz]

Screenshot form FF2 showing mangling.

[Jared Jurkiewicz]

Raising severity. We should not ship 1.4 without this fixed.

[Mike Wilcox]

This happened after the performance refactor of the charting labels soon after the 1.3 release. If someone gives me some clues as to what's broken and where to look, I can try to help.

[Karl Tiedt]

This appears to be the problem I am running into... thanks Jared! Mike, I can add this tibit of data from my tests trying to track down this issue.

My charts appear extremely NARROW and ignore the CSS width's set on the chart's div however if I give my charts x axis a large min/max variance... the chart begins to render more and more normal... understanding of this is that its only taking into account the # of data items for each series * step sizes in determining the width of a chart... (speculation but seems pretty sound)

[Tom Trenka]

This is probably the changeset you want to look at: ​http://bugs.dojotoolkit.org/browser/dojox/trunk/charting/axis2d/common.js?rev=17224

[Karl Tiedt]

Missplaced blame!

​http://bugs.dojotoolkit.org/changeset/17597

Alex's code does *not* break this test :)

Liucougars however.... *sigh*

[liucougar]

if you revert [17597], does it work?

the original code fixed in [17597] actually returns nothing (because width/height is not defined at all)

[Karl Tiedt]

​http://ktiedt.dojotoolkit.org/dojo/test_DataChart.html is running trunk - [17597] and yes it works :)

[liucougar]

according to ​IE getBoundingClientRect MSDN doc, width/height is not in the returned object (TextRectangle? object)

according to ​FF getBoundingClientRect doc, width/height is added to FF 3.5 (not in 3.0 either)

could you try to modify _getTextBox in gfx._base to just do return { w: undefined, h: undefined };

does that fix your test?

[Karl Tiedt]

^{results in same output prior to your patch. (same url)}

[Jared Jurkiewicz]

Any status on this? It's still ugly as crap in 1.4 with htmlLabels: false.

[Jared Jurkiewicz]

I confirm if I revert [17597], the problem goes away. It's definitely related to that checkin.

Assigning over to Liu to look further at, since it is his change that is breaking it.

[liucougar]

[17597] is correct, afaict, something is broken in dojox.charting when htmlLabel=false, here is a patch to fix it, but I don't know why it fixes this bug, pass it to elazutkin

[Karl Tiedt]

Replying to liucougar:

[17597] is correct, afaict, something is broken in dojox.charting when htmlLabel=false, here is a patch to fix it, but I don't know why it fixes this bug, pass it to elazutkin

Correct me if I'm wrong, but if a bug is fixed, shouldn't the underlying code that uses it be checked for a) functionality and b) corrections needed because of the bug that was fixed?

If we keep passing this around, its never going to get fixed... [17597] should have included any changes for GFX based code (particularly for something like Charting that is one of our *big* dojox attractions)... just my 2cents...

[Jared Jurkiewicz]

Followup. Any word on this? The corruption is really, really bad/ugly.

[Jared Jurkiewicz]

We really need to get this fixed for 1.4. :-(

[Eugene Lazutkin]

(In [20320]) dojo.charting: fixing corrupted labels non-HTML labels, unrelated pie label positioning issue (not centered), and some minor performance corrections, !strict, refs #9357.

[Eugene Lazutkin]

For the record: liucougar's patch was right (there was one more place where it should be applied), all effects were related to the original speed-up patch (it looks like a simple oversight).

!getBoundingClientRect returns left, right, top, bottom values following the original Microsoft spec, but newer implementations (see ​http://www.w3.org/TR/cssom-view/#clientrect for details) return width and height as well. That's why that code worked on some browsers and didn't work on others.

I fixed another bug introduced by the speed-up patch: wrongly positioned pie chart labels.

Jared, please re-test with whatever code you have to confirm that it fixed all problems.

[Eugene Lazutkin]

The original ticket: #9011.

[Eugene Lazutkin]

The related ticket: #9065.

[Jared Jurkiewicz]

The fix works nicely now! No more wonky charts with htmlLabels: false.

Export SVG works very nicely with charting using such labels.

Closing.