1085044 - Add telemetry probe for Flash content type classification

Status: RESOLVED WONTFIX

(Firefox Graveyard :: Shumway, defect, P1)

Description

[Till Schneidereit [:till]]

As a follow-up to bug 1075088, we want to add some more information to help us understand different Flash content types' importance.

These probes record whether, based on a very rough heuristic, a page contained no Flash content, only ads (i.e., banners), only non-ads (such as video players or games), or both. Additionally, counts of ads and non-ads are recorded.

I also implemented the change requested via IRC and in bug 1075088 comment 26, so we gate recording this information on the telementry.enabled pref now. (There is a very slight issue here in that a user might toggle that pref after a page has been loaded, so we'll under-count plugin instances. I don't think it matters, though.)

Comment 1

[Till Schneidereit [:till]]

Attachment: Add telemetry probe for Flash content type classification

Comment 2

[Georg Fritzsche [:gfritzsche]]

Comment on attachment 8507445 [details] [diff] [review]
Add telemetry probe for Flash content type classification

Review of attachment 8507445 [details] [diff] [review]:
-----------------------------------------------------------------

A few quick comments here.

::: browser/modules/PluginContent.jsm
@@ +98,1 @@
>      if (Services.telemetry.canSend) {

|if (pref) {|
Can we make the pref name a module-wide const?

@@ +386,1 @@
>      if (Services.telemetry.canSend && this._getPluginInfo(plugin).mimetype ===

|if (pref && ...|

@@ +419,5 @@
>  
> +  _matchesCommonAdDimensions(width, height) {
> +    // Formats mostly taken from
> +    // http://www.iab.net/guidelines/508676/508767/displayguidelines
> +    var commonAdDimensions = [

|let| or |const| please. This doesn't need to be a member function?

@@ +446,5 @@
> +      "970x66",
> +      "970x90",
> +      "970x250"
> +    ];
> +    return commonAdDimensions.indexOf(width + 'x' + height) > -1;

Nit: Easy to overlook the |> -1| there, can we switch this around?

@@ +475,5 @@
> +    let isAd = this._matchesCommonAdDimensions(width, height);
> +    if (isAd)
> +      this.flashPluginStats.adsCount++;
> +    else
> +      this.flashPluginStats.nonAdsCount++;

This is redundant due to |stats.instancesCount - stats.adsCount|.
Nit: bracing.

::: toolkit/components/telemetry/Histograms.json
@@ +2648,5 @@
> +  "FLASH_PLUGIN_ADS_ON_PAGE": {
> +    "expires_in_version": "50",
> +    "kind": "enumerated",
> +    "n_values": 30,
> +    "description": "Flash object instances classified as ads count on page"

Worth mentioning the classification criteria being dimensions?

Comment 4

[Michael Bebenita [:mbx]]

We also need a more granular view of SWF height / width. The current telemetry data suggests that 33% of SWF heights are between 1-43 pixels. If we get more granular information, we could identify exactly what these are.

http://telemetry.mozilla.org/#filter=nightly%2F36%2FFLASH_PLUGIN_HEIGHT&aggregates=multiselect-all!Submissions!Mean!5th%20percentile!25th%20percentile!median!75th%20percentile!95th%20percentile&evoOver=Builds&locked=false&sanitize=true&renderhistogram=Graph

Comment 5

[Chris Peterson [:cpeterson]]

The SWF area numbers are interesting. ~30% of SWFs have an area of 1–256 pixels (16x16 or narrower or wider), smaller than any of the IAB's standard ad sizes. This suggests that these tiny SWFs are not ads, but utility SWFs (e.g. for clipboard access, file uploading, or sound playback) or trackers.

http://telemetry.mozilla.org/#filter=nightly%2F36%2FFLASH_PLUGIN_AREA&aggregates=multiselect-all!Submissions!Mean!5th%20percentile!25th%20percentile!median!75th%20percentile!95th%20percentile&evoOver=Builds&locked=false&sanitize=true&renderhistogram=Graph

https://upload.wikimedia.org/wikipedia/commons/4/43/Standard_web_banner_ad_sizes.svg

Comment 6

[Till Schneidereit [:till]]

(In reply to Chris Peterson (:cpeterson) from comment #5)
> The SWF area numbers are interesting. ~30% of SWFs have an area of 1–256
> pixels (16x16 or narrower or wider), smaller than any of the IAB's standard
> ad sizes. This suggests that these tiny SWFs are not ads, but utility SWFs
> (e.g. for clipboard access, file uploading, or sound playback) or trackers.

Note that the bin includes 0x0, too, so the most likely explanation for most of this is swfobject creating an unsized instance as part of its Flash detection routine. If you take a look at the (way less representative, but still) Shumway probe for SWF area, you can see it showing considerably fewer entries in the lowest bin:
http://telemetry.mozilla.org/#filter=nightly%2F36%2FSHUMWAY_SWF_AREA&renderhistogram=Graph

Shumway without jsplugins doesn't support swfobject, so these numbers (as far as they are representative) will be a cleaner version of the lowest bin in some sense. That still leaves 8% at or below 16x16, though.

Comment 7

[Till Schneidereit [:till]]

We'll have something like it, but not this, exactly.