855019 - Eideticker CNN test failing on LG G2X

Status: RESOLVED WORKSFORME

(Firefox for Android Graveyard :: General, defect)

Description

[William Lachance (:wlach)]

For some reason, starting on the 21st, the LG G2X we have in Mountain View (running Android 2.3) is not able to completely load the CNN page, and thus is not able to complete the test.

It seems to die after getting this set of resources:

10:54:48.652340 GET: /ep1/cnn/i.cdn.turner.com/cnn/.e/img/3.0/global/shade/TL_shadow_7x6.gif
10:54:48.653458 GET: /ep1/cnn/i.cdn.turner.com/cnn/.e/img/3.0/global/shade/BR_shadow_1000x6.gif
10:54:48.653846 GET: /ep1/cnn/i.cdn.turner.com/cnn/.e/img/3.0/global/shade/BL_shadow_7x6.gif
10:54:48.658523 GET: /ep1/cnn/i.cdn.turner.com/cnn/.e/img/3.0/global/shade/cnn_shdsectbin.990px.bg.gif
10:54:48.659474 GET: /ep1/cnn/i.cdn.turner.com/cnn/.e/img/3.0/global/misc/filter.line.100px.gif
10:54:48.665030 GET: /ep1/cnn/i.cdn.turner.com/cnn/.e/img/3.0/global/icons/red.carrot.jpg

In the "good" builds, we continue on to get:

10:52:32.550437 GET: /ep1/cnn/i.cdn.turner.com/cnn/.e/img/3.0/search/search_btn_footer.gif
10:52:32.666762 GET: /ep1/cnn/i.cdn.turner.com/cnn/.e/img/3.0/global/misc/Numbers_Sprite.gif
10:52:33.059895 GET: /ep1/cnn/www.cnn.com/favicon.ie9.ico
10:52:33.274969 GET: /.element/ssi/www/breaking_news/3.0/banner.html
10:52:33.281548 GET: /ep1/cnn/i2.cdn.turner.com/cnn/dam/assets/120415013659-nat-tornadoes-day-00000211-video-tease.jpg
 10:52:33.281771 GET: /ep1/cnn/i2.cdn.turner.com/cnn/dam/assets/120412085412-zimmerman-charged-video-tease.jpg
10:52:33.282785 GET: /ep1/cnn/i2.cdn.turner.com/cnn/dam/assets/120414055723-sbt-brad-angelina-engaged-00003907-video-tease.jpg
 10:52:33.282938 GET: /ep1/cnn/i2.cdn.turner.com/cnn/dam/assets/111001095936-teen-texting-video-tease.jpg
10:52:33.283911 GET: /ep1/cnn/i2.cdn.turner.com/cnn/dam/assets/120414021613-piers-yearwood-country-food-00005515-video-tease.jpg
10:52:33.291292 GET: /ep1/cnn/i2.cdn.turner.com/cnn/dam/assets/120412090134-nr-ok-grandma-marijuana-arrest-00002410-video-tease.jpg

The logcat has some scary messages in it, but nothing that wasn't there before:

03-26 16:22:50.149 I/GeckoToolbar(27958): zerdatime 89521398 - Throbber start
03-26 16:22:50.149 D/GeckoFavicons(27958): Requesting cancelation of favicon load (1)
03-26 16:22:54.027 W/GeckoGlobalHistory(27958): Rebuilding visited link set...
03-26 16:22:54.167 E/GeckoConsole(27958): [JavaScript Error: "NS_ERROR_DOM_BAD_DOCUMENT_DOMAIN: Illegal document.domain value" {file: "http://10.250.120.74:45130/ep1/cnn/z.cdn.turner.com/cnn/tmpl_asset/static/www_homepage/1205/js/hplib-min.js" line: 42}]
03-26 16:22:55.969 E/GeckoConsole(27958): [JavaScript Error: "TypeError: cnnad_adTileIDGroup is undefined" {file: "http://10.250.120.74:45130/ep1/cnn/z.cdn.turner.com/cnn/tmpl_asset/static/www_homepage/1205/js/hplib-min.js" line: 70}]
03-26 16:22:55.989 E/GeckoConsole(27958): [JavaScript Error: "Error: PrototypeException for url [/.element/ssi/www/breaking_news/3.0/banner.html]: undefined" {file: "http://10.250.120.74:45130/ep1/cnn/z.cdn.turner.com/cnn/tmpl_asset/static/www_homepage/1205/js/hplib-min.js" line: 40}]
03-26 16:22:56.089 W/GeckoGlobalHistory(27958): Rebuilding visited link set...
03-26 16:22:57.687 W/GeckoGlobalHistory(27958): Rebuilding visited link set...

I've spent a few hours trying to figure out what's going on with no luck. Suppose the next step is to do some bisection of builds on the 21st, but I don't really have time to set that up now.

Comment 1

[William Lachance (:wlach)]

Ok, the problem was introduced somewhere in this set of commits:

https://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=d6a51ac10751&tochange=0e9badd3cf39

The central build for "d6a51ac10751" works fine. The weird thing is that nothing looks particularly suspicious in the set of commits that come after. Going to do some further bisection on inbound.

Comment 2

[William Lachance (:wlach)]

Ok, running builds on inbound I've narrowed it down further.

Last known good revision: f8e71f41c30f
First known bad revision: 74c06a7308d1

Thus two possibilities:

* 74c06a7308d1 - Chris Lord — Bug 852565 - Don't expand the viewport for small pages. r=kats Only expand the CSS viewport when a page reaches the screen size. If it's smaller than the screen size, lock the dynamic toolbar and keep the same, smaller CSS viewport. This 'fixes' sites that try to size themselves to the size of the screen and get it wrong. 
* b09f11bca869 - Olli Pettay — Bug 849601, no need to have explicit AutoMutationBatch when setting outerHTML, r=bz 

Neither really seem like likely culprits, but I suppose bug 852565 is the more likely of the two.

Comment 3

[William Lachance (:wlach)]

Ok, just tried reverting the patch in bug 852565 and rebuilt m-c locally, and CNN finished loading again (it doesn't with the patch applied). So something in there is causing this regression.

I suspect this bug is reproducible on any phone with the same screen specs as the LG G2X (480x800). 

To reproduce this without the eideticker harness just:

  git clone git://wrla.ch/ep1.git

Then serve the content from "cnn" from a static webserver and load "cnn.com".

Comment 4

[Kartikaya Gupta (email:[e-mail])]

Cwiiis, any thoughts on this? I'm a little concerned that the patch, combined with content scripts from the cnn page, might be going into an infinite loop somehow, as that's the only thing I can think of that would prevent the page from loading. If so it should be relatively easy to track down, and we should fix it as soon as possible.

Comment 5

[Chris Lord [:cwiiis]]

This is very likely the same bug as bug 853986 and bug 855240 - I'll add this as a blocker and I'm looking at it today. Might need to add some heuristics for pages to always get the larger viewport if this situation happens (but I hope not...)

Comment 6

[Chris Lord [:cwiiis]]

Does this still happen?

Comment 7

[William Lachance (:wlach)]

No, it appears not to happen anymore with the latest nightly. Should we just close this one as non reproducible?

Comment 8

[Chris Lord [:cwiiis]]

It was likely fixed with bug 855240, but I'm happy to just mark this as worksforme unless we have a pressing need to know exactly what fixed this.

Comment 9

[William Lachance (:wlach)]

Eugh, I spoke too soon. It looks like this is still reproducible most of the time. :(

Comment 10

[William Lachance (:wlach)]

How do you want to proceed on this? I could fedex a LG G2X to someone if it would help.

Comment 11

[Chris Lord [:cwiiis]]

Do you see the issue if you just load cnn.com, or the static resources that eideticker uses outside of eideticker? STR and a device would definitely make this easier for me to debug :)

Comment 12

[William Lachance (:wlach)]

I've been doing some testing with my device, and it appears to currently only be reproducible when I try to load the CNN test from the eideticker harness. Also, it appears that the issue is triggered from some javascript that is loaded/run on startup. If I remove a reference to the main CNN.com script (../z.cdn.turner.com/cnn/tmpl_asset/static/www_homepage/1205/js/hplib-min.js) from the webpage that's loaded, the problem disappears.

Summary: this seems like a really weird edge case, albeit a reproducible one. The only disturbing thing is that it seems to have just started occurring when the status bar stuff got committed, but I'm beginning to wonder if that's just a timing issue.

The reproduction steps are something like:

1. Checkout eideticker and bootstrap it (git clone git://github.com/mozilla/eideticker.git && cd eideticker && ./bootstrap.sh)
2. Run sutagent on the device and set appropriate environment variables (export DM_TRANS=sut && export TEST_DEVICE=<ip of device>)
3. Run CNN test: ./bin/runtest.py --no-capture cnn --app-name org.mozilla.fennec

Your device needs to be rooted, and have sutagent and orangutan installed for this to work properly. The LG device I'm reproducing this with has this setup already, as mentioned I could fedex it somewhere so long as I know I'll get it back at some point in the not-so-distant future (I need it for eideticker development), though I wonder if the hassle involved is worth it for what seems like such an obscure problem. I'd personally be tempted to just "fix" the CNN test here and call it a day.

Comment 13

[Chris Lord [:cwiiis]]

Does this still happen? I've checked in a fix that might affect this and I assume it's easy for you to test this - if it's not, just cancel the needinfo and I'll check :)

Comment 14

[William Lachance (:wlach)]

Yeah, things seems to be working ok. Resolving as worksforme.