www.cendio.com
Bug 6132 - Web Access is out of sync with noVNC
: Web Access is out of sync with noVNC
Status: CLOSED FIXED
: ThinLinc
Web Access
: trunk
: PC Unknown
: P2 Normal
: 4.8.0
Assigned To:
:
:
: 5480 5605 5610 5611 5983 6007 6126 6129 6135 6138 6175
:
  Show dependency treegraph
 
Reported: 2017-01-10 16:59 by
Modified: 2017-04-24 14:47 (History)
Acceptance Criteria:


Attachments
Screenshot of the old error message (54.13 KB, image/png)
2017-04-20 11:15, Samuel Mannehed
Details
Screenshot of the new error message (68.18 KB, image/png)
2017-04-20 11:16, Samuel Mannehed
Details


Note

You need to log in before you can comment on or make changes to this bug.


Description From cendio 2017-01-10 16:59:57
There has been a lot of development in upstream noVNC lately. ThinLinc's HTML5
client uses old noVNC code from 2016-09-05. Among the changes are fixes for the
following:

* Horizontal scroll (bug 5480)
* Double buffering (bug 5611)
* Incorrect position after panning (bug 5605)
* No update after panning (bug 5610)
* Can't use mouse and touch at the same time (bug 5983)
* On-screen keyboard closed when using tool bar (bug 6007)
* Can't move the control bar (bug 6126)
* Can't hear the bell/beep (bug 6129)
------- Comment #2 From cendio 2017-01-16 16:50:08 -------
Doing a vendordrop would also fix tearing issues: bug 6138
------- Comment #3 From cendio 2017-01-23 09:57:50 -------
As part of this bug, we also need to:

* disable the translation system in place in upstream code since it isn't up to
our standards yet.

* test keyboards after https://github.com/novnc/noVNC/issues/734 has been fixed
------- Comment #9 From cendio 2017-02-02 15:53:00 -------
(In reply to comment #3)
> As part of this bug, we also need to:
> 
> * disable the translation system in place in upstream code since it isn't up to
> our standards yet.

Disabled in the big vendor drop commit r32157

> * test keyboards after https://github.com/novnc/noVNC/issues/734 has been fixed

The change that broke this was reverted in commit r32158
------- Comment #11 From cendio 2017-02-06 11:15:47 -------
The "The HTML5 client is loading" message does not go away even after the
session started.
------- Comment #13 From cendio 2017-02-06 14:06:38 -------
Hopefully all regressions are now ironed out. Resolving.
------- Comment #14 From cendio 2017-02-10 15:07:40 -------
All done, closing this. The remaining testing is done on other bugs.
------- Comment #15 From cendio 2017-02-20 16:13:01 -------
I was having a look at the Chrome developer tools when making a login.

A successful login with 4.7.0 initiated about 577 KB worth of data transfer to
load noVNC and friends. Javascript was the biggest part of that, consuming
411KB.

With 4.7.0post_5377, the total request size has more than doubled to a whopping
1.3 MB. We now load almost 700KB of icons. 416K of those are a SINGLE icon!

> $ du -h www/images/tlclient_512.png 
> 416K    www/images/tlclient_512.png
------- Comment #16 From cendio 2017-02-21 11:11:11 -------
Apparently loading all icons is a known issue with Chrome:

https://bugs.chromium.org/p/chromium/issues/detail?id=324820

It apparently also was an issue with Firefox at one point:

https://bugzilla.mozilla.org/show_bug.cgi?id=751712

I checked the logs though and Firefox is no longer suffering from this issue
(although it still has the issue of not showing the requests in the debug tools
:/).

The bug reports also mention that none of the other major browsers suffer from
this, so it seems to be only Chrome left[1].

On the positive side, it loads the icons after everything else. IOW it doesn't
delay getting the session up and running.

[1] Possibly also Opera, since they're also using Blink now.
------- Comment #17 From cendio 2017-02-21 12:25:23 -------
More issues. The icons aren't loaded properly for the login page, and the 404
status code isn't sent properly to the browser.
------- Comment #20 From cendio 2017-02-23 17:30:34 -------
512x512 icon removed, and the others severely reduced in size via bug 6175.
Resolving again.
------- Comment #21 From cendio 2017-02-28 08:43:10 -------
A successful login does now correspond to 816 KB. 420 KB javascript and 182 KB
icons. Good enough.
------- Comment #22 From cendio 2017-04-20 11:13:18 -------
The vendordrop in commit 32157 brought in improved handling of uncaught errors.
This includes a traceback along with the error message. A side effect of this
is that bug 5651 now prevents users from going back to the login page.
------- Comment #23 From cendio 2017-04-20 11:15:24 -------
Created an attachment (id=800) [details]
Screenshot of the old error message

(In reply to comment #22)

This is how bug 5651 looked like before the changes brought in by this
vendordrop.
------- Comment #24 From cendio 2017-04-20 11:16:12 -------
Created an attachment (id=801) [details]
Screenshot of the new error message

(In reply to comment #22)

This is how bug 5651 looks like now.
------- Comment #25 From cendio 2017-04-20 11:22:32 -------
(In reply to comment #22)
> The vendordrop in commit 32157 brought in improved handling of uncaught errors.
> This includes a traceback along with the error message. A side effect of this
> is that bug 5651 now prevents users from going back to the login page.

One possible workaround could be to add an exception for this specific error
and ignore it.
------- Comment #26 From cendio 2017-04-20 13:33:02 -------
favicons no longer work on Firefox on Android. Seen on the Pixel tablet. They
work fine with 4.7.0.
------- Comment #27 From cendio 2017-04-20 14:14:47 -------
(In reply to comment #26)
> favicons no longer work on Firefox on Android. Seen on the Pixel tablet. They
> work fine with 4.7.0.

Scratch that. Not a regression. Apparently Firefox no longer shows favicons
when you've added a security exception for a site. Not sure it's something we
need to track, but I reported it upstream:

https://bugzilla.mozilla.org/show_bug.cgi?id=1358101
------- Comment #29 From cendio 2017-04-21 16:02:31 -------
The common NS_ERROR_NOT_CONNECTED error is hidden for now. Tester should verify
that following the instructions on bug 5651 does not cause an error message to
obscure the login-page-button. The error should still be printed to the browser
console however.
------- Comment #30 From cendio 2017-04-24 14:47:46 -------
(In reply to comment #29)
> The common NS_ERROR_NOT_CONNECTED error is hidden for now. Tester should verify
> that following the instructions on bug 5651 does not cause an error message to
> obscure the login-page-button. The error should still be printed to the browser
> console however.

A bit difficult to provoke for some reason, but I managed to get one instance
at least. And it did indeed only show up in the console, not on the screen.