Decide how to handle Marks

[sitespeedio]

Decide how to handle Marks

[tobli]

There are multiple issues with marks and measures. One is that the navigation timing spec uses absolute times, expressed as milliseconds since epoch, whereas user timing uses milliseconds since navigationStart using a milliseconds accurate to a thousandth of a millisecond. Looks like this difference will be addressed in http://w3c-test.org/webperf/specs/NavigationTiming2/. I suggest standardizing on the precision from the user timing spec, and converting all times to be relative to epoch.

A second issue is that it may be useful to get statistics of user timing marks. At the moment however they're not in statistics, since only intervals (measures) are included there. Given the proposed resolution to the first issues, one approach would be to synthesize measures for such user timing marks, using navigationStart as the start time, and the end of the mark as the end. I.e. {duration: 0, startTime: 226.26299997791648, entryType: "mark", name: "mark_above_the_fold"} would also generate a synthesized measure {duration: 216.26299997791648, startTime: 10.0000000, entryType: "measure", name: "mark_above_the_fold"}.

Would that make sense?

[soulgalore]

Well, that sounds good, maybe could be a little tricky? However, remember that I've already added marks in statistics (all the ones except the ones in Navigation Timing API), think we discussed it? :) No problem removing them though, for me it is important to get them as stats, because I think that is a really killer part of the code.

[tobli]

This issue is no longer relevant.