js agent r1635 strange behavior

[GoogleCodeExporter]

for the background, you can read #113.

Here are what i have from my tests on trunk js agent :

- i had to update wpt (from 2.7 something) to trunk in order to work, i.e. : 
screenshot not displayed , HAR empty ...
- i had some strange results regarding information like "fully load" and 
"document complete", i.e. :
   - tested on nissan.fr site for example
   - 8.821s DC and 8.822s FL
   - but i have in the HAR view something like 17s 
- there is lot of similar strange output in debug regarding a lot of urls, i.e. 
:
(
D 1218Z12:42:07.520 devtools2har.js:94 Socket.<anonymous>: devtools2har STDERR: 
Dec 18, 2012 1:42:07 PM org.webpagetest.dt2har.converter.HarResource 
createHarContent
WARNING: no content available for resource 
)

Regards,

-- 
Regis A. Despres

Original issue reported on code.google.com by regis.de...@gmail.com on 18 Dec 2012 at 3:29

[GoogleCodeExporter]

Original comment by regis.de...@gmail.com on 18 Dec 2012 at 3:30

Attachments:

• issue.127.tgz

[GoogleCodeExporter]

Original comment by k...@google.com on 18 Dec 2012 at 3:40

• Added labels: Component-agent-desktop, OpSys-All, Type-Defect

[GoogleCodeExporter]

some more infos from latest tests : 

 - video option checked : blank images in filmstrip view

Original comment by regis.de...@gmail.com on 18 Dec 2012 at 8:20

[GoogleCodeExporter]

here is last debug informations regarding time weirdness , blank screenshot on 
freebsd + flash chrome and blank video screenshots.

Please tell me how i can give an hand on this =)

Original comment by regis.de...@gmail.com on 22 Dec 2012 at 4:31

Attachments:

• debug.log
• chromedriver.log
• result.tgz

[GoogleCodeExporter]

and another logs with "-m 8" debug level on agent side

Original comment by regis.de...@gmail.com on 22 Dec 2012 at 4:36

Attachments:

• chromedriver.log
• debug.log
• result.tgz

[GoogleCodeExporter]

regarding the freebsd+flash blank screenshot issue :
 - i've tested with webdriverjs standalone same issue ; i'll fill a issue on github for this one.

Original comment by regis.de...@gmail.com on 23 Dec 2012 at 9:55

[GoogleCodeExporter]

ragarding the blank video screenshot i've found this during uploading results 
in the httpd logs :

==> /var/log/apache2/error_log <==
Not a JPEG file: starts with 0x89 0x50
Not a JPEG file: starts with 0x89 0x50
Not a JPEG file: starts with 0x89 0x50
Not a JPEG file: starts with 0x89 0x50
Not a JPEG file: starts with 0x89 0x50

Original comment by regis.de...@gmail.com on 24 Dec 2012 at 12:54

[GoogleCodeExporter]

As of my latest changes (r1651), video sort of works (OSX 10.8.2) -- the frames 
for nissan.fr are all white until the very last one at ~14s. The reason for 
this is I do not take any screenshots at all between the end of user script 
(log line "Sandbox finished, waiting for browser to coalesce") and the end of 
test run. I should probably be doing that, but right now I don't. The nissan.fr 
site loads basically empty, then a Flash program starts, and I'm guessing it 
begins displaying stuff shortly *after* the page onLoad fires, i.e. after WD 
get() returns.

Right now I take screenshots after each WD call. As a workaround, if you 
artificially for example look for an element (doesn't matter if you find it) 
each 1/2 second for 10 seconds using e.g. setTimeout/setInterval, you will have 
screenshots each 1/2 second for 10 seconds.

The inaccurately reported timings are I'm guessing also because most/all of the 
page is Flash, so I would not be surprised if the HAR-based timing computation 
gets confused, but that's just a wild guess, Pat would have an authoritative 
answer.

Original comment by k...@google.com on 29 Dec 2012 at 7:01

[GoogleCodeExporter]

FYI I just committed r1656 where I switched to DevTools-based screenshots 
instead of WebDriver screenshots. The NodeJS agent now requires Chrome 26 which 
is currently Canary, looks like it should become Dev channel next week. Tested 
on OSX 10.8.

The immediate benefit is now all screenshots are the same size -- showing only 
the visible window viewport, which makes the filmstrip work a lot better. 
WebDriver screenshots show full page even if it's larger than the visible 
viewport, so the frames were ending up different size, which was breaking the 
filmstrip.

The other upcoming benefit is I can now take periodic screenshots independent 
of the WD program, in parallel (almost), because it's a DevTools command and 
does not need to be synchronized with WebDriver commands. I haven't implemented 
that yet, it I'm going to.

Original comment by k...@google.com on 5 Jan 2013 at 6:53

[GoogleCodeExporter]

Pat, just to clarify -- the screenshot change has no effect on the HAR/timings 
logic, which is the main issue in this bug.

Original comment by k...@google.com on 5 Jan 2013 at 6:53

[GoogleCodeExporter]

The latest agent doesn not rely on har at all and uses dev toos traces directly 
for all of the timing.

Original comment by PatMee...@gmail.com on 4 Jul 2013 at 5:56

• Changed state: Fixed