User-Agent: Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:1.7.5) Gecko/20041106 Firefox/1.0 Build Identifier: Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:1.7.5) Gecko/20041106 Firefox/1.0 When loading a page with a position: fixed element on it (see the URL), the element is positioned against the page itself, rather than the viewport - even though it is anchored to the viewport when you scroll. Reproducible: Always Steps to Reproduce: 1. Load the URL. 2. Scroll down. 3. Reload (not shift-reload, as that resets the scorll-position). You can also just resize the window for this step. Actual Results: The position: fixed header vanishes. Expected Results: The position: fixed header should still be anchored to the top-left of the viewport. It appears to be positioning the fixed element against the page, so if you only scroll down 1 line and then resize/reload the position: fixed element is then locked partially off the top of the viewport. Each time a reflow occurs (reload or resize window) the element is re-anchored based on the scroll position of the page. This bug is currently causing ChatZilla to not have any header boxes, as they use position: fixed (that's how I found it :) ).
This regression is pretty recent, I believe, as I have only noticed it the past couple of days - it happens on my 2004110603 build, and CTho's 20041104 build, but (for reference) not Firefox 1.0 branch. I'm going to try to get some nightly builds now, but my initial guess is that it's regressed in the past week or two at most.
2004-11-01-06 is OK. 2004-11-03-07 is OK. 2004-11-04-07 is broken. (Windows zip builds from ftp.mozilla.org) http://bonsai.mozilla.org/cvsquery.cgi?treeid=default&module=all&branch=HEAD&branchtype=match&dir=&file=&filetype=match&who=&whotype=match&sortby=Date&hours=2&date=explicit&mindate=2004-11-03+05%3A00&maxdate=2004-11-04+07%3A00&cvsroot=%2Fcvsroot The check-in for bug 266968, at 2004-11-03 08:16, would seem a likely place to start looking.
CCing bz since his bug looks related.
The problem is that the CSS specification is somewhat self-contradictory here. 1) The fixed-position div should not move when we scroll. 2) Since it has auto top and bottom, its position depends on where it would show up if it were statically positioned. 3) Where it would show up if statically positioned depends on the scroll position. Note that before the checkin for bug 266968 we mishandled scrolling when calculating static positions. So the testcase in this bug rendered consistently, but absolutely positioned divs wouldn't render right if things got scrolled. Now we account for the scroll position correctly, and this bug results. I've sent mail to email@example.com regarding this inconsistancy... let's see what they say. Taking, I guess. As for Chatzilla, I recommend specifying some offsets. You might want to file a separate bug on that.
The mail I sent is at http://lists.w3.org/Archives/Public/www-style/2004Nov/0031.html Note that my proposed changes to the spec to remove the contradiction will probably leave chatzilla broken if implemented.
Ah, it turns out it was only my custom output window that lacked a top property for the header in ChatZilla - the built-in one does specify top: 0px and is thus not affected by this change in behaviour.
*** Bug 273177 has been marked as a duplicate of this bug. ***
*** Bug 276154 has been marked as a duplicate of this bug. ***
Created attachment 170755 [details] [diff] [review] Fix This restores the old behavior of using the static position with all scrollable things scrolled to default position.... I suspect we do want to keep backwards compat here, unfortunately. :(
Comment on attachment 170755 [details] [diff] [review] Fix r+sr=dbaron (assuming you're restoring old code, didn't really look)
Yes, this is restoring the old code. Checked in for 1.8b.
This bug is back again on trunk (not in FF 1.5). Reopen or new bug? New testcase: http://silver.warwickcompsoc.co.uk/temp/pos_fixed_jump.html
New bug, and please cc me. But that URL gives me a 503; please attach a testcase to the bug you file?
Filed bug 331729 and broke the knee-caps of the guy admining the server. ;)