Recently a regression was found in a large schema test. (fogbugz 22091).
I have isolated the issue to an ordinary self-contained TDML test that is part of the regular daffodil source base.
The problem was that when we changed InStream and PState to copy less, all our standard regression tests passed, but a large schema test regressed
Test test_backtrack1 recreates the problem. That is to say, if you back out the fix (line 258 of InStream.scala) that put back the old copying behavior, then this test will illustrate the failure. (With the fix, it doesn't fail.) THe point is, the fix is undesirable as it increases copying, so really we don't want this fix.