[cfe-dev] Dataflow analysis with Libtooling and LiveVariables

[Anna Zaks]

> On Dec 15, 2014, at 5:01 AM, Manuel Klimek <klimek at google.com> wrote:
> 
> On Fri Dec 05 2014 at 1:44:04 AM Saheel Godhane <srgodhane at ucdavis.edu <mailto:srgodhane at ucdavis.edu>> wrote:
> Hello everyone!
> 
> I am a PhD student and have been working with Clang with the aim to do an intra-procedural dataflow analysis. I am aware of a recent discussion <http://lists.cs.uiuc.edu/pipermail/cfe-dev/2014-November/039938.html> on this mailing list regarding the same but my questions are slightly more specific. Pardon me if this is the wrong forum to ask them, but anyway, here they are:
> 
> 1. Implementation question
> 
> I have been trying to use the Libtooling clang interface (based on this <http://clang.llvm.org/docs/RAVFrontendAction.html> tutorial) and the existing LiveVariables analysis. My goal is to answer the question: given a VarDecl, is it live at a particular SourceLocation? (is this even possible with LiveVariables, which somehow looks like block-level liveness analysis?)
> 
> +Anna for the LiveVariables question.

LiveVariables operates on the front-end CFG, which contains the references to the AST nodes in its basic blocks.  You should be getting precise liveness information for every statement. 

LiveVariables is currently used by the DeadStoresChecker, so that would be a good point of reference. DeadStoresChecker talks to the LiveVariables through the Observer.

>  
> So far, although I have been able to get some code on a LiveVariables analysis working, it is not displaying the expected result. Specifically, I have overridden VisitFunctionDecl() so as to calculate a LiveVariables analysis on every function. When I dump the live values, it shows correctly all the blocks in my function but it doesn't show any of the live variables. I tried using "clang -cc1 -analyze -analyzer-checker=debug.DumpLiveVars test.c" which works as expected, displaying all the variables live in different blocks.
> 
> Here is my code for VisitFunctionDecl. Pastebin version here <http://pastebin.com/bvgP3iy4>. Full cpp file for my clang tool here <http://pastebin.com/BPfY1Q5e>.
> 
> virtual bool VisitFunctionDecl(FunctionDecl *func) {
>     errs() << "Inside " << ++numFunctions << " " << func->getNameInfo().getName().getAsString() << "\n";
> 
>     clang::AnalysisDeclContextManager *ADCM = new clang::AnalysisDeclContextManager(false, true, true, true, true, true);
>     clang::AnalysisDeclContext *func_ADC = ADCM->getContext(func);
>     clang::LiveVariables *func_LV = clang::LiveVariables::computeLiveness(*func_ADC, false);
>     clang::LiveVariables::Observer *obs = new clang::LiveVariables::Observer();

In order to use the observer, you need to implement your own observer here which will get notified on observeStmt. 

> 
>     func_LV->runOnAllBlocks(*obs);
>     func_LV->dumpBlockLiveness((func_ADC->getASTContext()).getSourceManager());

I am not sure why liveness information is not dumped... I would try to debug this as well as the analyzer's version to see what's missing.

> 
>     return true;
> }
> 
> Surely I am missing something here but I am not able to figure out. Perhaps my instantiation of AnalysisDeclContextManager is incorrect?
> I started out with overriding VisitVarDecl() in the RecursiveASTVisitor but then realized that to call LiveVariables::computeLiveness() I need an AnalysisDeclContext object which works only with block-level declarations (am I correct?), so switched to VisitFunctionDecl().
> 
> Any help on how to get this working, or whether I should just switch over to writing a Checker, will be greatly appreciated! :)
> 
> That depends:
> If what you want in the end is a static analysis check, then yes, write a checker. If you want  to drive refactorings from the results, then the right approach is libTooling; but as you discovered libTooling is not integrated well with the CFG yet (that needs some work)

+1

> 
>  
> 
> Sincerely,
> Saheel.
> 
> _______________________________________________
> cfe-dev mailing list
> cfe-dev at cs.uiuc.edu <mailto:cfe-dev at cs.uiuc.edu>
> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev <http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20141215/e5a64299/attachment.html>