<div dir="ltr"><div class="gmail_extra"><div class="gmail_quote">On Mon, Jun 15, 2015 at 4:54 PM, Daniel Berlin <span dir="ltr"><<a href="mailto:dberlin@dberlin.org" target="_blank">dberlin@dberlin.org</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-color:rgb(204,204,204);border-left-style:solid;padding-left:1ex"><span class="">> I'm mostly going from Robert Wilson's 1997 phd thesis, although I'm pretty<br>
> sure I've seen a lot of the same ideas elsewhere as well.<br>
<br>
</span>Yes, using summary/transfer functions has been tried a lot.<br>
Note: the numbers in most of these phd thesis do *not* get born out in practice.<br>
<br>
See, e.g, <a href="https://urldefense.proofpoint.com/v2/url?u=http-3A__www.cse.unsw.edu.au_-7Eysui_papers_spe14.pdf&d=AwMFaQ&c=8hUWFZcy2Z-Za5rBPlktOQ&r=Mfk2qtn1LTDThVkh6-oGglNfMADXfJdty4_bhmuhMHA&m=l4T-n2b8SAz1_q9MzjemJQcwk4dWjc_r9wccyABk9vA&s=MJMihsMxFN9snxB-Z6hQxPI662QBXyPL4lZSXtd04tA&e=" rel="noreferrer" target="_blank">http://www.cse.unsw.edu.au/~ysui/papers/spe14.pdf</a><br>
<br>
vs<br>
<br>
<a href="http://impact.crhc.illinois.edu/shared/report/phd-thesis-erik-nystrom.pdf" rel="noreferrer" target="_blank">http://impact.crhc.illinois.edu/shared/report/phd-thesis-erik-nystrom.pdf</a><br>
<br>
vs<br>
the wilson thesis.<br></blockquote><div><br></div><div>Cool, I'll give those papers a read, thanks.<br></div><div> <br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-color:rgb(204,204,204);border-left-style:solid;padding-left:1ex">In particular, i expect the wilson algorithm is going to fall over<br>
very badly on anything of even medium size.<br></blockquote><div><br></div><div>Interesting.  Do you mean "fall over" in terms of running time, or precision, or something else?</div><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-color:rgb(204,204,204);border-left-style:solid;padding-left:1ex">
Nowadays, the thing of the day seems to be using cfl reachability<br>
based formulations and doing context-sensitivity on demand.<br>
</blockquote></div><br></div><div class="gmail_extra">Out of curiosity, what's the appeal (aside from academic novelty) of the CFL approaches?</div><div class="gmail_extra"><br></div><div class="gmail_extra">From a personal perspective, I'm particularly interested in the maximum analytic precision each AA approach can take, almost without regard to how much time or memory the computation takes to run.  So one thing I found appealing about Wilson's thesis was his "location set" approach for providing field-sensitivity.  </div><div class="gmail_extra"><br></div><div class="gmail_extra">I also liked Alain Deutsch's 1994 paper "Interprocedural may-alias analysis for pointers: Beyond k-limiting" for the same reason.  But I didn't take a crack at implementing *that* because based on the paper's complexity, I thought I'd go clinically insane trying to turn it into code :) </div></div>