Re: New codegen failing test-cases

9 Dec 2010


      On Wed, Dec 08, 2010 at 11:31:19PM -0500, Edward Z. Yang wrote:
...
Ian, I'd love a little guidance with this patch:
Thu Oct 21 13:08:53 BST 2010  Ian Lynagh 
  * Use takeUniqFromSupply in emitProcWithConvention
  We were using the supply's unique, and then passing the same supply to
  initUs_, which sounds like a bug waiting to happen.
    {
    hunk ./compiler/codeGen/StgCmmMonad.hs 607
    -        ; let (offset, entry) = mkEntry (mkBlockId $ uniqFromSupply us) conv args
    -              blks = initUs_ us $ lgraphOfAGraph $ entry <*> blocks
    +        ; let (uniq, us') = takeUniqFromSupply us
    +              (offset, entry) = mkEntry (mkBlockId uniq) conv args
    +              blks = initUs_ us' $ lgraphOfAGraph $ entry <*> blocks
    }
The new codegen has this hunk instead:
; let (offset, entry) = mkCallEntry conv args
              blks = initUs_ us $ lgraphOfAGraph $ entry <*> blocks
and it's not clear to me if this circumvents the previous bug
or mkCallEntry needs to be modified to expose the new supply.
The problem was that "us" was being used twice. If it's now only used
once then no problem.


Thanks
Ian