Summary:
These got smashed together with some weird rebase snafu. They are pretty intertwined anyway so the value of
separate commits is minimal (e.g. separate commits would not revert cleanly anyway).
== [lists] better fill rate logging (previously D4907958)
After looking through some production data, I think this will address all the issues we're seeing. Now:
- Header/Footer getting no longer counted as blank.
- Avoid floating point for Scuba.
- Compare actual time of blankness, not just samples.
- Include both "any" vs. "mostly" blank (similar to 1 and 4 frame drops).
- Include events where there is no blankness so we have a baseline.
- Remove events with too few samples
**Test Plan: **
A bunch of scrolling in FlatListExample
T17384966
== [Lists] Update SectionSeparatorItem docs (previously D4909526)
Forgot to update the language here when we modified the behavior with the introduction of separator
highlighting support.
** Test Plan: **
nope.
== [Lists] Add renderSectionFooter prop to SectionList (previously D4923353)
Handy for things like "see more" links and such.
The logic here is to render the footer last, *after* the bottom section separator. This is to preserve
the highlighting behavior of the section separator by keeping it adjacent to the items.
**Test Plan: **
Added to snapshot test and example:
{F66635525}
{F66635526}
== [SectionList] Add a bunch more info for rendering items and separators (previously D4923663)
This extra info can be helpful for rending more complex patterns.
**Test Plan: **
Made snapshot test more comprehensive and inspected the output.
== [Lists] reduce render churn (previously D4924639)
I don't think the velocity based leadFactor is helping and might actually be hurting because
it causes a lot of churn in the items we render.
Instead, this diff introduces fillPreference which biases the window expansion in the direction of scroll,
but doesn't actually affect the final bounds of the window at all, so items that are already rendered are
more likely to stay rendered.
**Test Plan: **
Played around in debug mode and watched the overlay - seems better. Also tests all pass.
T16621861
== [Lists] Add initialScrollIndex prop
Makes it easy to load a VirtualizedList at a location in the middle of the content without
wasting time rendering initial rows that aren't relevant, for example when opening an infinite calendar
view to "today".
**Test Plan: **
With debug overlay, set `initialScrollIndex={52}` prop in `FlatListExample` and
and see it immediately render a full screen of items with item 52 aligned at the top of the screen. Note
no initial items are mounted per debug overlay. Scroll around a bunch and everything else seems to work
as normal.
No SectionList impl since `getItemLayout` isn't easy to use there.
T17091314
Reviewed By: bvaughn
Differential Revision: D4907958
fbshipit-source-id: 8b9f1f542f9b240f1e317f3fd7e31c9376e8670e
Summary:
If tracking is enabled and the sampling check passes on a scroll or layout event,
we compare the scroll offset to the layout of the rendered items. If the items don't cover
the visible area of the list, we fire an `onFillRateExceeded` call with relevant stats for
logging the event through an analytics pipeline.
The measurement methodology is a little jank because everything is async, but it seems directionally
useful for getting ballpark numbers, catching regressions, and tracking improvements.
Benchmark testing shows a ~2014 MotoX starts hitting the fill rate limit at about 2500 px / sec,
which is pretty fast scrolling.
This also reworks our frame rate stuff so we can use a shared `SceneTracking` thing and track blankness
globally.
Reviewed By: bvaughn
Differential Revision: D4806867
fbshipit-source-id: 119bf177463c8c3aa51fa13d1a9d03b1a96042aa
Miguel Jimenez Esun
James Burnett
Spencer Ahrens