371829 - stacks layers with style float:left, leading to text overlap

Status: NEW

(Core :: Layout: Floats, defect)

Description

[Georg Maaß]

User-Agent:       Mozilla/5.0 (Macintosh; U; PPC Mac OS X Mach-O; en-US; rv:1.9a3pre) Gecko/20070226 SeaMonkey/1.5a
Build Identifier: Mozilla/5.0 (Macintosh; U; PPC Mac OS X Mach-O; en-US; rv:1.9a3pre) Gecko/20070226 SeaMonkey/1.5a

The text layers are stacked instead of floating side by side. The css class defines them as:

.iconbar {
	padding-right: 15px;
    float: left;
    width: auto;
    height: 34px;
}

.iconbar img {
    float:left;
}


Reproducible: Always

Steps to Reproduce:
visit the page and scroll down to the Java, QuickTime, RealPlaye, WMP area. There the text is stacked.
Actual Results:  
Text is stacked

Expected Results:  
Text should float

This occurs with nightly build from trunk [Mozilla/5.0 (Macintosh; U; PPC Mac OS X Mach-O; en-US; rv:1.9a3pre) Gecko/20070222 SeaMonkey/1.5a] as well as with suiterunner build [Mozilla/5.0 (Macintosh; U; PPC Mac OS X Mach-O; en-US; rv:1.9a3pre) Gecko/2007022610 SeaMonkey/1.5a]. It looks identical.

No such problem with good old Mozilla [Mozilla/5.0 (Macintosh; U; PPC Mac OS X Mach-O; en-US; rv:1.7.13) Gecko/20060414]

Comment 1

[Georg Maaß]

Attachment: Screenshot of Problem

Comment 2

[Georg Maaß]

Attachment: Screenshot as it should be (and is in old Mozilla)

Comment 3

[Martijn Wargers (dead)]

This seems to have regressed between 2006-12-07 and 2006-12-08, so that means it's a regression of bug 300030.
A minimal testcase that shows the bug would be nice.

Comment 4

[philippe (part-time)]

Attachment: test case

essentially: a left floated block element that is floated and has auto width, and contains a left floated image shrink-wraps to much.

Comment 5

[Jeff Walden [:Waldo]]

(In reply to comment #4)
> Created an attachment (id=256546) [details]
> test case

Request: could you create a testcase which has the same structure (but doesn't use the faulty construct) and creates the rendering you expect?  (Note: unfortunately a screenshot won't work, because actual rendering varies depending on system fonts.)  This will help with adding the testcase to Mozilla's test suite.  Thanks!

Comment 6

[philippe (part-time)]

Attachment: test case that renders correctly

(In reply to comment #5)
I think this should do. The only difference with the faulty test case: I specified a width (non auto) on the floated block.

Comment 7

[Eli Friedman]

The issue here is that we throw out float data anytime we hit a newline.  This makes us underestimate the natural width of the float.  We really need to keep it around until we get to a line below the float.  However, we don't know how tall the float is, so we don't know when to do that.  If we just never clear out the floats, though, we'll overestimate in a lot of cases (although it might be better to overestimate than underestimate in general).  Also, the current model doesn't model the impact of floats on block lines at all.

Comment 8

[jklussman]

Is this the same issue as reported in http://forums.mozillazine.org/viewtopic.php?t=628938 regarding Bungie.net?