STR: * Open http://plasmalang.org/about.html (make sure you shift-reload, since I suspect this has to do with web font loading). * Go to the "goals" section. Expected results: * An ordered list appears, with all its bullets. Actual results: * Only the first bullet shows until you reflow the page. This is reproducible with and without stylo.

Works fine on release fwiw.

Err, I didn't plant to take this just yet, since I'm kinda busy this week. Will take a look when I have the time though.

It seems Emilio reproduced this is v54 by shift-reloading, hence why the 'regression' keyword was removed. Probably P3 given that.

So this isn't a reflow bug. I checked that things are reflowed correctly, since that was the first suspicion. But it should have been more obvious than that: when the bug occurs, the list bullet will paint correctly just as a result of switching to a different tab and switching back to the tab that had the bug (which doesn't reflow, but does repaint). I suspect this is somehow related to downloadable fonts not working right with the way that the bullet frames draw text (which doesn't use text runs... though comments indicate that they ought to).

It looks like this is a regression after all, but an old one. I get this from mozregression: 1:42.77 INFO: Narrowed nightly regression window from [2016-04-18, 2016-04-20] (2 days) to [2016-04-18, 2016-04-19] (1 days) 1:42.77 INFO: Got as far as we can go bisecting nightlies... 1:42.77 INFO: Last good revision: 1f16d3da9280e40ada252acf8110b91ee1edbb08 (2016-04-18) 1:42.77 INFO: First bad revision: ae7413abfa4d3954a6a4ce7c1613a7100f367f9a (2016-04-19) 1:42.77 INFO: Pushlog: https://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=1f16d3da9280e40ada252acf8110b91ee1edbb08&tochange=ae7413abfa4d3954a6a4ce7c1613a7100f367f9a I'm on a slowish connection right now (which might make it reproduce super easily).

With hg bisect in the Nightly range from comment 8, I'm pretty sure this came from this cset: changeset: 293631:9116ef367e02 user: Markus Stange <mstange@themasta.com> date: Mon Apr 18 13:49:14 2016 -0400 summary: Bug 1236043 - Use TiledRegion for the invalid region of a layer. r=jrmuizel My STR were: 1. visit http://plasmalang.org/about.html 2. Do repeated reload / shift-reload for 1-2 minutes, and watch for times where the entire page gets painted at once *except* for the "2." in the numbered list In affected builds, this aggressive reloading would usually reproduce with one or two not-quite-complete repaints (missing the "2.") within a minute or so of reloading.

Regressions from bug 1236043 are usually cases of existing bugs, where we already computed an insufficient invalidation region, but then simplified that region to something larger that happened to include the pixels that we forgot. It's likely that you'll get a different regression range if you test with the testcase in this bug instead of with the full page, because on the testcase there's less that gets invalidated when the font loads (really only the word "Text"), so the region simplification behavior that we were using before bug 1236043 would have been less likely to hide the insufficient invalidation.