[llvm-commits] [llvm] r49243 - /llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp

[Duncan Sands]

> Is this fixing a regression from Dan's patch? Can you separate out the  
> flushing functionality from getroot? It seems a bit strange to use  
> getRoot for its side effects.

It looks like Dale's patch fixed two Ada testsuite failures.  They started
failing about a week ago, and were both due to exception handling problems.
That fits with them being caused by Dan's patch r48893.  An additional test
that started failing longer ago (the failure not being obviously related to
exception handling), and was still failing yesterday, starting passing again
today.

Having read Dan's patch, I now understand what Dale is doing, and it seems
sensible.  I agree that it would be better to have an explicit flush call.

Ciao,

Duncan.