clang-7 debugging error about cppad

[bradbell]

Thanks for the bug report. It would help if you created an example.

Perhaps you could
cp bug/template.sh bug/check_simple_vector.sh
and then edit check_simple_vector.sh so that it demonstrates the bug
and I would then be able to reproduce the problem by executing
bug/check_simple_vector.sh

Perhaps check_simple_vector.sh would be very short and you could just past it into the comment window for this bug.

[bradbell]

I added an example test clang compliation of checking simple vectors:
https://github.com/coin-or/CppAD/blob/master/bug/clang_simple.sh

It works fine with clang version 6.0.0 (tags/RELEASE_600/final)
Please try running
bug/clang_simple.sh
to see if it fails with clang 7

[han-so1omon]

I'm going to try to debug this problem today and tomorrow. My suspicion is that the error is somehow expected behavior but with a strange error message. The stack trace seems related to this call, which does some type casting:

    CppAD::vector<CppAD::AD<double>> outputVar(n);
    outputVar = tiny_dnn::convertToAdVec<CppAD::AD<double>>(v->at(i).out);

    CppAD::Independent(outputVar);
   
    CppAD::vector<CppAD::AD<double>> objectiveVar(m);
    objectiveVar[0] = tiny_dnn::convertToAdVec<CppAD::AD<double>>(objective(outputVar,
        v->at(i).advantageValues, v->at(i).in,
        v->at(i).value, v->at(i).valueTarget, policy));
    
    CppAD::ADFun<double> surrogateVars(outputVar,objectiveVar); 

    v->at(i).objectiveValue = (double)objectiveVar[0];

    tiny_dnn::vec_t x(n);
    x = v->at(i).out;
    v->at(i).gradientValues = surrogateVars.Jacobian(x);

tiny_dnn::vec_t is: typedef std::vector<float_t, aligned_allocator<float_t, 64>> vec_t;

[bradbell]

I have fixed the conversion warnings, in the Cppad source, generated by clang 6.0; see 08-13 on
https://coin-or.github.io/CppAD/doc/whats_new_18.htm#08-13

There are still warnings on some of the supporting software, like Ipopt and Adolc
(perhaps CppAD should treat these packages like eigen; i.e., give up on them fixing their warnings
and treat the corresponding include directories like system directories).

You can use clang to build the CppAD tests, that do not use adolc or ipopt, using:
bin/run_cmake.sh --no_adolc --no_ipopt --clang
cd build
make check

You can build and run one test using
bin/test_one.sh dir/file.cpp
where dir/file.cpp is the path the individual test.

[bradbell]

Is this still a problem for you ? If not, would you please close this issue.

[bradbell]

@han-so1omon I am closing this bug. If it is still a problem for you, please re-open it.