"_quirks_mode" metric is always false

[zcorpan]

In HTTPArchive/legacy.httparchive.org#58 we introduced a custom metric to record document.compatMode. I think it has been working in the past, but now it appears to be false for all pages.

SELECT url FROM `httparchive.pages.2019_11_01_desktop` WHERE STRPOS(payload, '"_quirks_mode":false') > 0

Lots of results.

SELECT url FROM `httparchive.pages.2019_11_01_desktop` WHERE STRPOS(payload, '"_quirks_mode":true') > 0

No results.

I wanted this metric now to research whatwg/quirks#48 😐

[rviscomi]

Are there any example pages that return BackCompat that we can test this with? Are there other methods of detecting quirks mode, like the doctype, that can be parsed from the HTML body?

[zcorpan]

HTTPArchive/legacy.httparchive.org#40 (comment) has a query that still works with newer dataset.

I ran it now on httparchive.pages.2019_11_01_desktop resulting in 209,037 pages.

https://docs.google.com/spreadsheets/d/14-44npYPWVgdPQyMzIhdfD1m3UTEsqOSnH93rPNHbi4/edit?usp=sharing

[zcorpan]

I think that result misses out pages that have no doctype at all.

Running a query for WHERE doctype IS NULL gives only 39 results, which seems way too low to represent the number of pages without a doctype.

[zcorpan]

No wait, there's a whole bunch of rows with "" which probably means no doctype.

[rviscomi]

@zcorpan can we close this?

[zcorpan]

No, we should either fix the metric so it works again or remove it.

The original implementation is here

https://github.com/HTTPArchive/legacy.httparchive.org/blob/master/custom_metrics/quirks_mode.js

That repo is now marked legacy. Should it be moved somewhere else?

[rviscomi]

Here's a query that looks at all values of the custom metric:

SELECT
  JSON_EXTRACT(payload, '$._quirks_mode') AS q,
  COUNT(0) AS n
FROM
  `httparchive.pages.2020_06_01_mobile`
GROUP BY
  q

q	n
	100
1	238849
false	5494475

For some reason the JSON encoding of true is being converted to 1, but according to these results it does seem like there are some pages that are in quirks mode.

[zcorpan]

Ooh, so it is working then. Great! Thanks @rviscomi

[zcorpan]

So per this metric, ~4.17% of pages are in quirks mode.