I've attached an updated version of my patch that better handles cases where pointer arithmetic is done after casting a constant-size array to a pointer for a smaller base type (e.g. casting an int array to char*). Of the pointer arithmetic warnings, about 24% could be considered false positives; however, the actual number of false positives is quite small and 2/3 of them stem from the use of a single macro--if you count those as a single warning & false positive, the rate drops to 17%. Of the false positives most are from semi-questionable pointer arithmetic where a constant greater than the length of the array/pointer is being added to the pointer and some int > 1 being subtracted from it, e.g.:<br>
<br>void foo(int n) {<br> char x[5];<br> if (n > 0) bar(x + 6 - n);<br>}<br><br>Cheers,<br>Kaelyn<br><br><div class="gmail_quote">On Tue, Jul 19, 2011 at 11:44 AM, Kaelyn Uhrain <span dir="ltr"><<a href="mailto:rikka@google.com">rikka@google.com</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">I'm still looking at the pointer arithmetic warnings to determine which are false positives, but I realized I screwed up my previous stats a bit as I forgot to account for duplicated/repeated warnings (e.g. the same header included in multiple compilation units and so generating the same warnings multiple times). For unique warnings, the stats are:<br>
<br>- 16.5% increase in warnings from before my patch (originally reported a 24% increase)<br>- 47% of those new warnings being about pointer arithmetic<br>- 6.7% of all of the warnings emitted with my patch applied are concerning pointer arithmetic. (originally reported 8.3%)<br>
- The new pointer arithmetic warnings represent a 7.7% increase in warnings from before my patch, not 10%.<br><br>I'll send another email once I have a feel for how noisy the pointer arithmetic warnings are.<br><br>Cheers,<br>
<font color="#888888">
Kaelyn</font><div><div></div><div class="h5"><br><br><div class="gmail_quote">On Mon, Jul 18, 2011 at 5:54 PM, Kaelyn Uhrain <span dir="ltr"><<a href="mailto:rikka@google.com" target="_blank">rikka@google.com</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
Ted,<br><br>You're welcome. I'll try to figure out what fraction of the pointer arithmetic warnings are false positives (requires a bit of manual digging on my part to determine if the code is indeed buggy or if it is valid / intended). For the overall 24% increase in warnings, keep in mind that over half of that is the existing bounds checking now being applied to cases where it wasn't before, i.e.:<br>
<br>char *foo[5];<br>foo[77]; // -Warray-bounds already found<br>&foo[77]; // -Warray-bounds currently misses<br>*foo[77]; // -Warray-bounds currently misses<br><br>The function that did the bounds checking would never catch the last two cases because it would see a UnaryOperator (in the above cases for the '&' and the '*') and skip the expression instead of looking inside the UnaryOperator expression for the array subscripting.<br>
<br>The new pointer arithmetic bounds checking only represents a 10% increase in warnings--and IMHO that is the only part where the number of false positives introduced might be an issue.<br><br>Cheers,<br><font color="#888888">Kaelyn</font><div>
<div></div><div><br><br><div class="gmail_quote">
On Mon, Jul 18, 2011 at 5:36 PM, Ted Kremenek <span dir="ltr"><<a href="mailto:kremenek@apple.com" target="_blank">kremenek@apple.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<div style="word-wrap:break-word">Hi Kaelyn,<div><br></div><div>Thanks for the statistics. What would be good to know is what fraction of these are false positives (i.e., are these all real bugs)? A small random sample might be helpful. A 24% increase in warnings is fairly substantial, and we don't want to do that unless there is a real benefit.</div>
<div><br></div><font color="#888888"><div>Ted</div></font><div><div></div><div><div><br><div><div>On Jul 18, 2011, at 5:18 PM, Kaelyn Uhrain wrote:</div><br><blockquote type="cite"><div class="gmail_quote">Ted,<br>
<br>On Thu, Jul 14, 2011 at 5:04 PM, Ted Kremenek <span dir="ltr"><<a href="mailto:kremenek@apple.com" target="_blank">kremenek@apple.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<div style="word-wrap:break-word"><br><div>The only other issue: should this be controlled under a separate warning flag, at least initially so we can experiment with this new warning and see how noisy it is? E.g. "-Warray-bounds-pointer-arithmetic".</div>
</div></blockquote></div><br>I've tested the -Warray-bounds changes against the Google codebase and my patch increases the number of warnings from -Warray-bounds by 24%. Of the new warnings, 57.33% are for array indexes that most likely weren't picked up before because of unary operators like & or * (11.1% of all the warnings now emitted), and the remaining 42.67% are from out-of-bounds pointer arithmetic (8.3% of all the warnings from -Warray-bounds).<br>
<br>Cheers,<br>Kaelyn<br>
</blockquote></div><br></div></div></div></div>
</blockquote></div><br>
</div></div></blockquote></div><br>
</div></div></blockquote></div><br>