1504444 - Sticky elements disappear after changing tab

Status: RESOLVED DUPLICATE of bug 1528957

(Core :: Layout: Positioned, defect)

Description

[Pochang Chen]

Attachment: sticky-bug.html

Steps to Reproduce
------------------
1. Open sticky-bug.html (as attached).
2. Scroll to bottom.
   The header should be visible (sometimes it's not; this may be the same bug).
3. Switch to another tab.
4. Switch back to the tab with sticky-bug.html.
   The header should be visible, but it's not (100% reproducible).

Tested versions
---------------
Arch Linux
firefox 63.0.1-1 + sway-git r5050.a918844e-1: Bugged
firefox 63.0.1-1 + plasma-desktop 5.14.2-1: Bugged
firefox 65.0a1.20181103-1 + sway-git r5050.a918844e-1: Bugged
firefox 65.0a1.20181103-1 + plasma-desktop 5.14.2-1: Bugged

Comment 1

[Stuart Philp :sphilp]

After scrolling down a ways, the header disappears once the scroll bar disappears. May be related to bug 1455142, moving to p&z  for :kats review

Comment 2

[Kartikaya Gupta (email:[e-mail])]

Emilio, any thoughts on what's going on here? It doesn't smell like an APZ bug per se, since the buggy behaviour appears (albeit not very consistently) after a main-thread repaint. Also it happens with APZ disabled, and with WR enabled/disabled.

Comment 3

[Emilio Cobos Álvarez (:emilio)]

We have a couple other similar sticky bugs... I guess this is specific to tables?

The table painting code is not great, but probably can dig what's up with this... I also wanted to look into bug 1498772, so who knows, maybe I get to look into both.

But I'd bet this is not an APZ but, yeah.

Comment 4

[Emilio Cobos Álvarez (:emilio)]

I took a look today.

We're hitting this code path on the bogus paint:

  https://searchfox.org/mozilla-central/rev/b096dcf0ea226af628fe03f7e7acb56a25853533/layout/tables/nsTableRowGroupFrame.cpp#217

This is the cursor code not dealing with sticky positioning at all. It also probably doesn't deal with transforms properly? I'd need to check.

In any case, I wonder how useful is this code. I can fix it so it tries to deal with sticky positioning, but  This was added in bug 241796, but it looks like APZ and RDL should kind of supersede this code... Matt, does that sound right?

Should I just remove this code, or should I make it understand sticky positioning? It's only used for DL building so shouldn't have other perf impact...

Comment 5

[Matt Woodrow (:mattwoodrow)]

This will be fixed by bug 1528957.