[PATCH] rewrite fwprop's dataflow

[Paolo Bonzini]

> This explanation should be part of the body of the code, not something 
> that will be lost in the mail.   Given that this is effectively a new  
> and unique problem, this is especially important.

Thanks Kenny, I'll copy and paste it with small adjustments to 
df-problems.c/fwprop.c.

> The dataflow code is fine.   I am not authorized to approve the changes 
> to fwprop.c

I think you can, the border between DF and passes using it are murky. 
After all it's just replacement for RD+chains -- I could have made a 
"singleton UD" problem in df-problems.c and that would have qualified as 
dataflow, but the code would be uglier.  Also, you are right now the 
only one besides me to understand the MD problem. :-)

Paolo