You'll need to hide the async generator test from the 3.5 builds. Otherwise looks good.

Codecov Report

Merging #1192 into master will decrease coverage by <.01%.
The diff coverage is 94.73%.

@@            Coverage Diff             @@
##           master    #1192      +/-   ##
==========================================
- Coverage   99.57%   99.56%   -0.01%     
==========================================
  Files         106      106              
  Lines       12800    12818      +18     
  Branches      983      984       +1     
==========================================
+ Hits        12745    12762      +17     
- Misses         35       36       +1     
  Partials       20       20

So apparently my first version didn't work at all, because we were replacing coro after creating the Task object, so the Task kept going with the original coroutine... fortunately codecov caught it. I fixed that and made the test more robust, so I think this is good now?

We did some code review for this PR on gitter, starting here: https://gitter.im/python-trio/general?at=5d58b8434e17537f525114d7