376365 - update margin collapsing to slight rule change

Status: NEW

(Core :: Layout: Block and Inline, defect)

Description

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

There's a slight change to the margin collapsing rules in http://lists.w3.org/Archives/Member/w3c-css-wg/2007JanMar/0535.html that's been accepted for the next draft of CSS2.1.  We should implement this.

Steps to reproduce:
 load http://dbaron.org/css/test/2007/0329-blog-examples/1

Actual results:
 hello is the same distance below the blue block as A is above it

Expected results:
 there should be a gap between hello and the blue block so hello is 50px below A

Comment 1

[fantasai]

Attachment: suggested patch

I'm pretty sure that conditional fixes the bug. I'm less sure about whether it's the right place to put the conditional.

Comment 2

[fantasai]

My build still passes http://www.hixie.ch/tests/adhoc/css/box/block/margin-collapse/microsoft/all.html with flying colors (except 005.html, of course, since the test is now wrong). I also ran the reftests and nothing new failed.

Comment 3

[Robert O'Callahan (:roc) (email my personal email if necessary)]

I probably should be the reviewer...

Shouldn't the fix be after "clearance = aState.mY - (currentY + topMargin);"?

+      clearance -= clearance + incomingMargin.get();

This is equivalent to "clearance = -incomingMargin.get();" Can you explain why this is correct? I don't see it. I think it would be better if we explicitly did the calculation described in Hixie's email.

Comment 4

[fantasai]

> Shouldn't the fix be after "clearance = aState.mY - (currentY + topMargin);"?

That's where I thought it should go, but it didn't work there. I spent hours trying to figure out why, but my programming fu wasn't strong enough. Even setting clearance to random numbers there didn't break my testcases. -__-;; I don't understand at all.

> Can you explain why this is correct?

I can try to explain why I *think* it's correct...

The point of the spec fix is that clearance should never move an element *up*.
It only moves an element up if it's negative, hence the check for negativeness.
It also only moves the element up if it winds up overcompensating for the hypothetical margin collapse. The margin can't collapse more than the amount of the previous margin, so if the magnitude of negative clearance is more than that, then it's overcompensating, and it's overcompensating by
  |clearance| - incomingMargin
which, given that clearance is negative, is the same as
  clearance + incomingMargin
Subtract out the overcompensation, and the element no longer moves up.

That's my logic. Hyatt thinks its wrong, so maybe I'm missing something important here. Hence "suggested patch" rather than "proposed patch" and a request for dbaron's review since he's more of a margin collapsing guru than I am.

Comment 5

[fantasai]

Note that, if my logic serves me correctly, if a negative margin is one (or both) of two margins collapsing, then the amount of the collapse must be non-negative. A  negative margin collapsing with a positive one is the sum of the two (no difference if they collapse or don't collapse), and two negative margins collapsing are the max of the two, i.e. the difference between not collapsing and collapsing is positive. Therefore (correct me if I'm wrong) clearance is only negative if it's compensating for a positive-positive collapse.

Comment 6

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Comment on attachment 262698 [details] [diff] [review]
suggested patch

I should have transferred this review request ages ago.

Comment 7

[Robert O'Callahan (:roc) (email my personal email if necessary)]

Comment on attachment 262698 [details] [diff] [review]
suggested patch

I kinda forgot about this. I think it's right, though.

We should have a reftest.

Comment 8

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

And updated version of this patch is:
http://hg.mozilla.org/users/dbaron_mozilla.com/patches/raw-file/7c922ff64911/clearance-should-not-move-up
and it fixes our handling of the following tests in the CSS 2.1 test suite:
http://test.csswg.org/suites/css2.1/20101027/html4/margin-collapse-clear-005.htm
http://test.csswg.org/suites/css2.1/20101027/html4/margin-collapse-clear-011.htm

Comment 9

[Robert O'Callahan (:roc) (email my personal email if necessary)]

Do you want to land it for FF4?

Comment 10

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

I tend to think not; I'd rather land it afterwards.

Also see:
http://lists.w3.org/Archives/Public/public-css-testsuite/2010Nov/0124.html

Comment 11

[fantasai]

I'll note that the CSS2.1 spec could get blocked on whether this fix breaks web compat, and currently only Mozilla seems capable of figuring that out within the next year or two.

Comment 12

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

This seems to break Acid2.

Comment 13

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

This also fixes http://test.csswg.org/suites/css2.1/20110111/html4/margin-collapse-164.htm

Comment 14

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

So apparently:
 * the IE folks say fixing this breaks some sites
 * the tests are actually invalid, although only unintentionally so, since "hypothetical position" is defined **relative to the parent**

So it sounds like we might fix the tests.

Comment 15

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Though http://lists.w3.org/Archives/Member/w3c-css-wg/2007JanMar/0514.html shows that the explicit intent of the discussion was to fix the test that has now become http://test.csswg.org/suites/css2.1/20101027/html4/margin-collapse-clear-005.htm

Comment 16

[Boris Zbarsky [:bzbarsky]]

Sounds like this patch is not ok to go on m-c, right?  If I'm wrong, please let me know!

Comment 17

[BugBot [:suhaib / :marco/ :calixte]]

The bug assignee is inactive on Bugzilla, so the assignee is being reset.