Re: Repost: RFA: re-instate struct_equiv code

[Bernd Schmidt <bernds_cb1 at t-online dot de>]

Kenneth Zadeck wrote:
> You either end up having to analyze the control flow graph to predict
> the places where the 1 bit should be deleted (Danny's #1), or you end
> up with too many places that have 1 bits (Danny's point #2).  It is
> also possible to have a third outcome, you end up with no enough bits
> turned on, in which case the solution is wrong. 

#1 and #2 I can see - do you have an example for #3?

> The problem is that there is no easy way to tell which of the 1 bits
> came from the assignment being deleted and which came from some other
> assignment.  This is not a problem that you are going to see addressed
> in the near future or even at all.  Over the last 25 years there have
> been a lot of attempts to do incremental data flow, but I would
> suggest implementing any of them (my own phd thesis included).  The

Missing "not"?

> algorithms are complex to implement and do not do much better than
> starting from scratch.

Thanks for the input,


Bernd