Re: publishing


Subject: Re: publishing
From: Craig Chambers (chambers@cs.washington.edu)
Date: Tue Oct 31 2000 - 17:13:47 PST


Thanks, Jonathan.

Jonathan Aldrich wrote:
> * added an ipca_debug option (default: false) to trigger breakpoints
> during IP analysis if there are no possible callees at a call site. This
> is "wrong" with respect to the program's dynamic behavior , but sometimes
> occurs in practice because the analysis is conservative.

I don't understand this. No possible callees corresponds to unreachable code or
a program with a latent msg-not-understood error. A conservative analysis would
only overestimate callees, not underestimate them. What you're describing might
correspond to a real bug in the analysis, not an overly conservative analysis.

> * added a check for analyzing functions that have a pure_empty argument
> when in SCS IPCA analysis. It is safe (and improves results) to not
> analyze the function call, because if the function can never be called
> without an argument at run time.

I agree with this; it matches my discussion above.

-- Craig



This archive was generated by hypermail 2b25 : Tue Oct 31 2000 - 17:13:54 PST