[cfe-dev] Adding Support for APINotes

Gábor Márton via cfe-dev cfe-dev at lists.llvm.org
Tue Sep 29 09:11:36 PDT 2020

Hi Saleem,

About the technical aspects. The feature to be able to add extra
information to library functions would be really helpful in the Clang
Static Analyzer. APINotes might be a way forward to achieve that.

We'd like to add the following to functions

   - Argument constraints
      - Not null
      - Range, e.g. isalpha(int x) x must be in [0,255].
      - Buffer size, e.g. fread(x1, x2, x3)  x1 is a buffer with a size
      denoted by x2 * x3
   - Taint rules
      - propagation
          int x; // x is tainted
        int y;
        myPropagator(x, &y); // y is tainted
      - filter
          int x; // x is tainted
        isOutOfRange(&x); // x is not tainted anymore
      - sink
          int x; // x is tainted
        myNamespace::mySink(x); // It will warn
      - Error return rules, e.g. the return value of mktime() should always
   be checked against -1.
   (D72705 <https://reviews.llvm.org/D72705>)
   - Indicate which functions we should consider with statistical checkers
   <https://www.youtube.com/watch?v=IM1DYZ-xn-k> (WIP, not upstreamed yet).

Note to taint analysis: we might want to tag global variables as sources.
E.g. std::cin.

I think that we could extend the APINote implementation that you have in
the referenced commit, but with a CSA specific attribute:
def CSANotes : InheritableAttr {
  let Spellings = [GNU<"csa">];
  let Args = [StringArgument<"FreeFormatOrSomeYamlMaybe">];
  let Subjects = SubjectList<[Tag, Function, Var],
The string argument could be parsed specifically to the CSA developers'
needs in a CSA specific implementation file. IMHO it would be a flexible
solution because we could avoid adding many CSA specific attributes to
Attrs.td and it would allow us to make experiments.

My concerns about the referenced APINotes implementation:

   - Specifically with the `'Name' key`. How could we match in C++
      - overloaded functions
      - member functions
      - functions in namespaces
      - function templates
      - member functions of class templates (e.g. std::vector::begin)
   - Do we have a mechanism to test/indicate if a note is not applied but
   it should have been?

Adding CSA guys who might want to comment.
+Artem Dergachev <noqnoqneo at gmail.com>
+Gábor Horváth <xazax.hun at gmail.com>


On Mon, Sep 28, 2020 at 11:10 PM Saleem Abdulrasool via cfe-dev <
cfe-dev at lists.llvm.org> wrote:

> Hi!
> I'd like to revive the effort around merging APINotes support into the
> llvm.org repository.  This was previously discussed at
> http://lists.llvm.org/pipermail/cfe-dev/2017-May/053860.html nearly 3
> years ago.  The overall consensus seemed neutral to slightly positive.  Now
> that the Swift specific attributes have been merged, the APINotes seem like
> a good next step towards converging the fork.
> I've put up https://reviews.llvm.org/D88446 to add initial documentation
> on the feature before trying to add the actual implementation with the
> goal of gathering commits on the technical aspects of the feature.
> Thanks.
> --
> Saleem Abdulrasool
> compnerd (at) compnerd (dot) org
> _______________________________________________
> cfe-dev mailing list
> cfe-dev at lists.llvm.org
> https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20200929/c78b0645/attachment-0001.html>

More information about the cfe-dev mailing list