Android: Decrease cost of reflection

[rigdern]

This change suppresses access checking during reflection which makes reflection faster by decreasing its overhead.

Test plan (required)

My team uses this change in our app.

Adam Comella
Microsoft Corp.

[facebook-github-bot]

By analyzing the blame information on this pull request, we identified @astreet and @lexs to be potential reviewers.

[astreet]

@facebook-github-bot import

Going to perf test this internally -- did you all measure any perf wins with this?

[astreet]

(I mean, this is probably strictly better than what we currently have so I'll land it anyway :) )

[facebook-github-bot]

@astreet has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

[rigdern]

@astreet One of my teammates commented on the performance improvement we saw here:

dryganets wrote:

I was optimizing particular flow in our application. We have VirtualListView component and I was testing scrolling inside. In addition to that we have a lot of ongoing activity through the bridge. Mostly timers.

I was using Android tracing to profile performance.

After the change reflection takes 15% of CPU. Before around 60% if I remember correctly.

I also had a synthetic test for reflection only. I was able to made the same count of calls in half of the time with property set.

For sure it would depend on your exact scenario.
For simple calls it has bigger impact.

AaaChiuuu ran a benchmark but said he didn't see any improvement:

AaaChiuuu wrote:

I ran 71c5bb6 on an internal RN cold start benchmark which has about 1000 method calls via reflection on the startup path of a fairly complex UI and it showed no statistically significant gains. I'm curious to understand how this was profiled to show an 80% win. Seems like there's a chance this win is far less than 80% on reflection calls or the cost of reflection is almost nominal compared to other costs.