Real purpose for failure in "Too restrictive sanity check"

[Jeffrey A Law]

  In message < 39017AD6.4C52906@rcp.co.uk >you write:
  > Jan
  > 
  > With your modified patch which handles zero-length
  > sequences applied the stage1 compiler miscompiles
  > genattrtab.c on x86-linux when bootstraping with
  > -funroll-loops.
  > 
  > Here's a testcase derived from the expand_units()
  > routine in genattrtab.c
[ ... ]
I added this to the testsuite.
jeff