Open core-ai-bot opened 3 years ago
Comment by iwehrman Thursday Dec 19, 2013 at 19:15 GMT
It still reproduces with Chrome 32.0.1700.68.
Comment by larz0 Thursday Dec 19, 2013 at 19:30 GMT
I can't reproduce it on Chrome Version 32.0.1700.68 beta (OS X Mavericks).
Running the HTML file within the source code.
Comment by iwehrman Thursday Dec 19, 2013 at 19:34 GMT
Sometimes it takes a few tries. It also might be sensitive to the width of the window? I've seen it on both 10.9.0 and 10.9.1.
CC@
peterflynn
Comment by redmunds Friday Dec 20, 2013 at 16:26 GMT
I just hit this on Mac 10.8.5 with Chrome 32.0.1700.68 beta.
Comment by peterflynn Friday Dec 20, 2013 at 17:49 GMT
Tagging Sprint 36 -- we should make sure a Chromium bug exists for this issue so it doesn't get into stable builds without warning.
@
iwehrman Should we remove "with Live Development" from the title? I'm pretty sure we reproduced it on your machine from a plain old file://
URL too...
Comment by njx Friday Dec 20, 2013 at 22:48 GMT
I did some searching in the Chromium bugbase. There have been a number of issues filed regarding webfonts, but they seem to involve them falling back to a non-webfont rather than failing to render entirely. All of the recent webfont bugs have been duped to https://code.google.com/p/chromium/issues/detail?id=236298, which was filed in April, whereas our bug only started happening with the latest beta, so I don't think it's the same issue.
I'm trying to create a reproducible case that we can file against Chromium, but am having difficulty. I can easily reproduce this from Live Preview in Brackets, but have been unable to reproduce it by just launching Chrome from the command line. Even if I hit the built-in Brackets server, and launch using:
open -g -n -a "Google Chrome" --args
--no-first-run
--no-default-browser-check
--remote-debugging-port=9222
--user-data-dir=/Users/nj/Library/Application\ Support/Brackets/live-dev-profile/
http://127.0.0.1:50331/index.html
which I think simulates exactly what brackets-shell does in OpenLiveBrowser()
, I can't get the problem to happen.
Comment by njx Friday Dec 20, 2013 at 22:54 GMT
One possible theory: it requires the interstitial page to be loaded first in order for the problem to occur. That wouldn't explain why@
iwehrman saw it with an ordinary file URL, but perhaps it's more likely to happen if another page was in the middle of loading in the same tab, or some such.
Comment by njx Friday Dec 20, 2013 at 22:55 GMT
Or, slightly differently, perhaps it has to do with the fact that the target page is loaded via remote debugging instead of from the command line.
Comment by njx Saturday Dec 21, 2013 at 00:08 GMT
I did an experiment where I made it so instead of Live Development agents initializing on the page load event, they initialize on the "frameStoppedLoading" event. This seemed to get rid of the problem, probably because it avoids touching anything until the web font loading script has completed. Not sure this really gets us closer to figuring out the underlying problem (and it doesn't at all affect the point that we were at one time able to reproduce it by manually navigating to a file URL), but it's a data point.
Comment by njx Friday Jan 10, 2014 at 01:16 GMT
Another data point: moving the font script to the bottom of <body>
fixes the problem, again presumably because it changes the timing of font loading so it doesn't interfere with whatever connection we're making in Live Development.
Comment by njx Saturday Jan 11, 2014 at 00:08 GMT
Moving to sprint 37. We have a workaround for the Getting Started material in case the bug hits Chrome stable soon.
Comment by MarcelGerber Tuesday Jan 14, 2014 at 21:07 GMT
FYI: Chrome Stable was updated today. http://googlechromereleases.blogspot.de/2014/01/stable-channel-update.html
Comment by njx Tuesday Jan 14, 2014 at 23:23 GMT
Good thing we added the workaround :) I confirmed that, without the workaround, the problem occurs in the new stable build.
Comment by njx Thursday Jan 30, 2014 at 00:13 GMT
Updated title and description to reflect the current state of things.
Comment by njx Thursday Jan 30, 2014 at 00:42 GMT
BTW, I just verified that it happens with Typekit's loading script as well.
I'm thinking it has something to do with the fact that the scripts deliberately block rendering in order to avoid the FOUT. The timing of when they allow rendering to continue is probably tickling some new bug/behavior in Chrome.
I'm going to try to see if someone on the EWF/Typekit team can help figure this out.
Comment by njx Tuesday Feb 04, 2014 at 01:30 GMT
@
rogervila ran into this in #6701 as well. He ran into it with the Google Fonts loader, so it isn't something specific to the EWF/Typekit loader (although it might be similar or the same).
Comment by njx Wednesday Feb 19, 2014 at 19:35 GMT
The MissionBit students were running into this with (I think) Google Fonts as well.
Comment by redmunds Tuesday Feb 25, 2014 at 18:44 GMT
Looks like this was fixed in Chrome 33: http://blog.typekit.com/2014/02/04/chrome-bug-affecting-web-fonts/
Comment by jasonsanjose Tuesday Feb 25, 2014 at 18:48 GMT
I saw it today with PR #6997 and Chrome 33 on windows. :cry:
Comment by njx Tuesday Feb 25, 2014 at 23:29 GMT
Interesting. I've been suspecting that the bug they fixed was different from ours - the bug they fixed was one where web fonts fall back to a system font. The last comment on that Typekit post sounds more like the problem that we're seeing (where the web font disappears completely). It's good to know that it's reproducible outside of our live development case, although it doesn't really help us figure out how to reproduce it...
Comment by njx Friday Mar 14, 2014 at 23:28 GMT
This is already in Ready but I'm nominating it for 1.0 since we definitely want it fixed by then.
Comment by JeffryBooher Tuesday Mar 25, 2014 at 01:25 GMT
Seems pretty stable with Chrome 34.0.1847.76 beta on Mac.
Comment by njx Tuesday Mar 25, 2014 at 03:25 GMT
@
JeffryBooher What files were you testing with? Did you sync back to an earlier version of Getting Started that used webfonts?
Comment by JeffryBooher Tuesday Mar 25, 2014 at 03:56 GMT
@
NJX I was testing my own code from scratch which helped me uncover some other problems but I had a conversation with@
peterflynn who suggested a few other test scenarios to look at. I should have amended my comments to reflect that. I will have a look at the old getting started page before we ripped out EWF and see how it does later tonight.
Comment by JeffryBooher Tuesday Mar 25, 2014 at 20:18 GMT
Verified that all scenarios are fixed in Chrome 34. Moving to Tracking.
Comment by njx Tuesday Mar 25, 2014 at 20:49 GMT
Does it work in Chrome 33 as well? (That's what I would expect if the changes we made to Live Preview startup timing were what fixed it.) If so, maybe we should just close it as fixed.
Comment by JeffryBooher Monday Mar 31, 2014 at 17:14 GMT
Verified that it does work with Chrome 33. FBNC@
njx
Issue by iwehrman Thursday Dec 19, 2013 at 19:13 GMT Originally opened as https://github.com/adobe/brackets/issues/6275
Update:
Originally, the bug was that the web fonts on the Getting Started page would often render as blank when using live development in Chrome 32 (which was in beta at the time).
Chrome 32 is now in the stable channel, and we removed the web fonts from Getting Started as a workaround so people wouldn't experience the issue with the Getting Started page. However, this could still occur with other pages that use web fonts in live preview.
It's not clear whether this only happens with Edge Web Fonts (which we were using on the Getting Started page) or whether it would happen with other web fonts as well. It's possible that there's something specific about the EWF loading script that was triggering some new bug in Chrome.
Original screenshot:
This may be a Chrome beta issue; I'm using 32.0.1700.55.