1874776 - Consider ignoring web extension background pages from use counter telemetry.

Status: RESOLVED FIXED

(Core :: DOM: Core & HTML, defect)

Description

[Emilio Cobos Álvarez (:emilio)]

As noted by Chris in bug 1874439, we see a lot of top level pages in telemetry just in a clean profile. Debugging a little bit I saw:

moz-extension://e05ac0da-4b3f-454e-b42c-da21e4d4d5ed/_generated_background_page.html
moz-extension://48940584-81a3-482c-b28a-b2fd68a508ce/_generated_background_page.html
moz-extension://6816b494-87ca-416f-a1bb-979cf0032d78/_generated_background_page.html
moz-extension://9ad7fc62-b15f-46d2-9ebf-bd8e9171f3cf/_generated_background_page.html
moz-extension://9ec6fc95-76af-4dea-8d47-31101807fa8d/_generated_background_page.html
https://www.mozilla.org/en-US/privacy/firefox/

So web extension background pages can use web APIs, but given they're going to be mostly blank, and that they are more "in control" than regular websites, we might want to ignore those?

Comment 1

[Emilio Cobos Álvarez (:emilio)]

Attachment: Bug 1874776 - Ignore built-in extension pages from telemetry. r=smaug,nika

Just like we ignore chrome pages. This should be uncontroversial, even
though it might be the case that we don't want any of these to show up.

Not sure how to best test this...

Comment 2

[Rob Wu [:robwu]]

(In reply to Emilio Cobos Álvarez (:emilio) from comment #0)

So web extension background pages can use web APIs, but given they're going to be mostly blank,

What makes you think that they are mostly blank? While functional extensions can be as simple as a few dozen lines of code, larger extensions are not uncommon.

and that they are more "in control" than regular websites, we might want to ignore those?

By "in control", do you mean that we have complete coverage of the set of public add-ons? This is true.

Nevertheless, I am not opposed to dropping extensions from use counters. The URLs (in the example above) consist of UUIDs that cannot easily be mapped back to the original extension URL.

Comment 3

[Emilio Cobos Álvarez (:emilio)]

So it seems to me comment 1 should be uncontroversial...

Though I'd lean to just ignore extension pages altogether. It doesn't seem they were measured intentionally (the code was changed in https://hg.mozilla.org/mozilla-central/rev/dc8d6a216ffa28165acaf67ddc2e88974d95a8e9#l6.19, but to preserve use counter behavior).

I don't think these are likely to bring much useful data, and I suspect they skew the data towards not using features (because pages that do nothing in their background script still get a background page, and that gets loaded every browser session).

So I propose ignoring extensions for use counters too... Wdyt? Alternatively we could stash them in a different bucket, but that seems overkill to me.

(In reply to Rob Wu [:robwu] from comment #2)

Nevertheless, I am not opposed to dropping extensions from use counters. The URLs (in the example above) consist of UUIDs that cannot easily be mapped back to the original extension URL.

(We also by design do not store the URL anyways)

Comment 4

[Emilio Cobos Álvarez (:emilio)]

Thoughts on the above?

Comment 5

[Chris H-C :chutten]

Seeing as part of our intended order of work is to create a public dataset of use counters to (e.g.) give folks a look at what the shape of the web platform looks like (at least as Firefox users see it), it'd be nice to make this as close to "intuitive" to our intended audience (people who know about the web platform, but probably not the inner workings of browsers) as possible.

If we wanted a new bucket, that'd be cool (though that'd be another (at time of writing) 356 new CSS counters for page and another 356 for document, plus however many deprecated ops and so forth which isn't a small number). We could definitely handle it and keep it separate for our own uses... but until and unless someone comes up with a extension-background-page-specific use case for these counters, I think we'd best hold off.

As for continuity of data, we're only talking back to Firefox 121 for the Glean use counter metrics and Firefox 118 for Legacy Telemetry, so there's not much to preserve.

I think it's best to make the document and page counts (both the denominators and the use counters) reflect documents of web content specifically. We should remember to annotate the "use-counters" ping to note that "Prior to Firefox version <whichever it is when it lands> Document and Page use counters also included Web Extension background pages in their counts."

Or, succinctly: I think your proposed change is good.

Comment 6

[Rob Wu [:robwu]]

(In reply to Chris H-C :chutten from comment #5)

I think it's best to make the document and page counts (both the denominators and the use counters) reflect documents of web content specifically. We should remember to annotate the "use-counters" ping to note that "Prior to Firefox version <whichever it is when it lands> Document and Page use counters also included Web Extension background pages in their counts."

FYI: the scope is not limited to background pages only - the counters covered any moz-extension document (including extension popup panels, sidebars and actual tabs). The background pages just stood out because they are around by default.

Comment 7

[Nika Layzell [:nika] (ni? for response)]

On first blush, it seems like it might be OK to ignore extension pages for use counters, though it might still be nice to know what APIs extensions end up actually using in some cases.

For an extension though, I'm not sure that "page loads" are the correct granularity to be counting this sort of thing by anyway.

Comment 8

[Emilio Cobos Álvarez (:emilio)]

Attachment: Bug 1874776 - Ignore all extension pages from use counters. r=smaug

It seems the discussion in the bug goes in this direction, and I think
it's the right call as described there.

Depends on D198630

Comment 9

[Emilio Cobos Álvarez (:emilio)]

Ok given the comments above, I'll proceed with landing both patches, and writing a test for this.

Comment 10

[Emilio Cobos Álvarez (:emilio)]

Attachment: Bug 1874776 - Refactor browser_use_counters.js to avoid some duplication. r=chutten

This will make testing the patches in this bug easier too.

Comment 11

[Emilio Cobos Álvarez (:emilio)]

Attachment: Bug 1874776 - Add a test for extension use counters being ignored. r=chutten

Depends on D198716

Comment 12

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/80324d436ea5
Ignore built-in extension pages from telemetry. r=smaug

Comment 13

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/262aeaa318d3
Ignore all extension pages from use counters. r=smaug

Comment 14

[Cristian Tuns]

https://hg.mozilla.org/mozilla-central/rev/80324d436ea5
https://hg.mozilla.org/mozilla-central/rev/262aeaa318d3

Comment 15

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/64f1b6b86676
Refactor browser_use_counters.js to avoid some duplication. r=chutten
https://hg.mozilla.org/integration/autoland/rev/362d9aca9e10
Add a test for extension use counters being ignored. r=chutten

Comment 16

[Norisz Fay [:noriszfay]]

https://hg.mozilla.org/mozilla-central/rev/64f1b6b86676
https://hg.mozilla.org/mozilla-central/rev/362d9aca9e10