[cfe-dev] Writing simple checkers for the static analyzer

[Jordan Rose]

So, several weeks later...I poked further into this and found that it was actually a bug in the analyzer, fixed in r211209. So, thanks for letting us know!

Jordan


On May 28, 2014, at 14:19 , Rafael Auler <rafaelauler at gmail.com> wrote:

> Hi Jordan,
> 
> Thanks! Running it with the clang driver made it work. Regarding the clang frontend, I couldn't make it work by using "-analyzer-checker=core,alpha.mychecker", but "-analyzer-checker=core,osx,alpha.mychecker" did work!
> 
> Best regards,
> Rafael
> 
> 
> On Wed, May 28, 2014 at 12:25 AM, Jordan Rose <jordan_rose at apple.com> wrote:
> It's worth noting that running without the core checkers enabled isn't really a sane state for the analyzer. You should either pass "core,alpha.mychecker.MyChecker" to -analyzer-checker, or use "clang --analyze" instead of "clang -cc1 -analyze" and use "-Xanalyzer" to pass down the -analyzer-checker option.
> 
> Jordan
> 
> On May 26, 2014, at 18:36 , Rafael Auler <rafaelauler at gmail.com> wrote:
> 
>> Hi Jordan,
>> 
>> I'm using the tagged version 3.4, do  you think it is an issue that got fixed in trunk? I am quite sure that I have it enabled, since I put some debugging printfs and saw the results:
>> 
>> -1 = doNotCallTwice() has not been called before
>> 0 = doNotCallTwice() has been called before
>> 
>> rafael$ clang -cc1 -analyze -analyzer-checker=alpha.mychecker.MyChecker mytest.c
>> doNotCallTwice! -1
>> doNotCallTwice! -1
>> doNotCallTwice! -1
>> rafael$ vim mytest.c  # change to avoid folding
>> rafael$ clang -cc1 -analyze -analyzer-checker=alpha.mychecker.MyChecker mytest.c
>> doNotCallTwice! -1
>> doNotCallTwice! -1
>> doNotCallTwice! -1
>> doNotCallTwice! 0
>> mytest.c:8:5: warning: Called twice
>>     doNotCallTwice();
>>     ^~~~~~~~~~~~~~~~
>> 1 warning generated.
>> 
>> 
>> In the first case, I use only function calls in the test case. In the second, I put the extra statement that forces the engine to avoid folding, and the detection finally works.
>> 
>> Thanks,
>> Rafael
>> 
>> 
>> On Mon, May 26, 2014 at 5:52 PM, Jordan Rose <jordan_rose at apple.com> wrote:
>> Hm, if I drop this into my clang sources (and update it to match changes in trunk), I don't see any issues with what you've written—building and running it on your sample input works fine. Are you sure you have it enabled? (I forgot to pass -analyzer-checker on my first test, so I have to ask.)
>> 
>> Jordan
>> 
>> On May 25, 2014, at 7:28 , Rafael Auler <rafaelauler at gmail.com> wrote:
>> 
>>> Hi Jordan,
>>> 
>>> Sure, it is attached. Thanks for taking a look at this. 
>>> 
>>> Cheers,
>>> Rafael
>>> 
>>> 
>>> On Sun, May 25, 2014 at 4:13 AM, Jordan Rose <jordan_rose at apple.com> wrote:
>>> Hi, Rafael. From your description, this sounds like a bug in the analyzer—two program states with differing user data should not be folded. Can you attach your checker so I can take a look and see if there are any obvious mistakes? (on your part or ours).
>>> 
>>> Thanks,
>>> Jordan
>>> 
>>> On May 24, 2014, at 22:01 , Rafael Auler <rafaelauler at gmail.com> wrote:
>>> 
>>> > Hello,
>>> >
>>> > I am trying to write a very simple checker for the clang static analyzer for the sake of writing a first exercise on this topic. Its goal is to simply alert whether a specific function has been called twice in a given path. Let's assume the name of this specific function that I am tracking is "doNotCallTwice()".
>>> >
>>> > In order to record state information, I use the REGISTER_TRAIT_WITH_PROGRAMSTATE macro to register an unsigned together with the program state. This integer indicates whether the function "doNotCallTwice()" has been called in a path and, if it is equal to 1 in a node where I detect yet another call, I prepare to report a "double call" bug. I use "checkPostCall" for changing the state.
>>> >
>>> > However, something strange happens. My extra integer registered in the program state is not sufficient to differentiate two ProgramStates with the same ProgramPoint: the engine fold the two nodes anyway, ignoring my new state information. On the other hand, the information *is* propagated. If I use other ways to avoid the nodes being folded, the checker works fine.
>>> >
>>> > An example where it does not work:
>>> >
>>> > void myfunc (int x, int y) {
>>> >   if (x)
>>> >     doNotCallTwice();
>>> >   if (y)
>>> >     doNotCallTwice();
>>> >   doNotCallTwice();
>>> > }
>>> >
>>> > Since programstates get folded in the ExplodedGraph, I never detect any path where two calls to doNotCallTwice() happen. However, change the code in the following way avoids the folding and make my checker work:
>>> >
>>> > void myfunc (int x, int y) {
>>> >   if (x)
>>> >     doNotCallTwice();
>>> >   if (y)
>>> >     doNotCallTwice();
>>> >   y = x;  // Now x and y are not dead anymore and this won't be folded
>>> >   doNotCallTwice();
>>> > }
>>> >
>>> > I based my checker on SimpleStreamChecker.cpp. Am I doing something conceptually wrong?
>>> >
>>> > Best regards,
>>> > Rafael
>>> > _______________________________________________
>>> > cfe-dev mailing list
>>> > cfe-dev at cs.uiuc.edu
>>> > http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev
>>> 
>>> 
>>> <MyChecker.cpp><mytest.c>
>> 
>> 
> 
> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20140618/0082b37b/attachment.html>