Re: Bug fix in `tree.c': build_range_type

[Jeffrey A Law <law at cygnus dot com>]

  In message <199807280923.LAA14979@esmeralda.gerwinski.de>you write:
  > Richard Henderson wrote:
  > > TYPE_SIZE_UNIT being NULL should have no ill effects except a loss
  > > of optimization.  I wonder, though, if the bug Peter is reporting
  > > doesn't have to do with TYPE_SIZE_UNIT being completely uninitialized
  > > in the new type node.  Is this true?
  > 
  > Replacing the fix by `TYPE_SIZE_UNIT (itype) = NULL_TREE;' has the same
  > effect (i.e. fixes the bug).  It really seems as if `TYPE_SIZE_UNIT' is
  > completely unitialized otherwise.
But shouldn't TYPE_SIZE_UNIT be initialized to zero by the make_node
call which created the type?  There's a loop in that make_node that is
supposed to zero out the node when it's allocated.

Is the node in question being allocated by some other routine?  If so
we need to know what routine so that we can fix it (or possibly look
into having it call make_node).


  > BTW, although I have subscribed to egcs-bugs on 17 Aug 1997, I did not
  > get my own email back.  However, I am getting about 3 emails per week
  > through the EGCS mailing lists.  Seemingly, something is wrong with the
  > EGCS lists, too.
I don't see gerwinski.de anywhere in the egcs subscription lists.
If you're getting mails through egcs, it's not from us (unless you're
subscribed via some other address).

jeff