v3.0.1 compile error in icc 16.0.4

[mark-99]

While this is maybe a compiler defect in icc 16 (icc 17.0.5 worked ok), perhaps there's an easy fix. Note than json v2.x did not have this error.

Source line is GTEST macro:
EXPECT_EQ(j["name"], "Name");

json/3.0.1/include/nlohmann/json.hpp(4177): error: the generated default constructor for "nlohmann::detail::internal_iterator<nlohmann::basic_json<std::map, std::vector, std::string, bool, int64_t={long}, uint64_t={unsigned long}, double, std::allocator, nlohmann::adl_serializer>>" cannot be used in an initializer for its own data member
      internal_iterator<typename std::remove_const<BasicJsonType>::type> m_it = {};
                                                                                 ^
          detected during:
            instantiation of class "nlohmann::detail::iter_impl<BasicJsonType> [with BasicJsonType=const nlohmann::basic_json<std::map, std::vector, std::string, bool, int64_t={long}, uint64_t={unsigned long}, double, std::allocator, nlohmann::adl_serializer>]" at line 376 of "googletest/1.7.0.1/include/gtest/gtest-printers.h"
            instantiation of "void testing::internal::DefaultPrintTo(testing::internal::IsContainer={int}, testing::internal::false_type, const C &, std::ostream *) [with C=nlohmann::basic_json<std::map, std::vector, std::string, bool, int64_t={long}, uint64_t={unsigned long}, double, std::allocator, nlohmann::adl_serializer>]" at line 477 of "googletest/1.7.0.1/include/gtest/gtest-printers.h"
            instantiation of "void testing::internal::PrintTo(const T &, std::ostream *) [with T=nlohmann::basic_json<std::map, std::vector, std::string, bool, int64_t={long}, uint64_t={unsigned long}, double, std::allocator, nlohmann::adl_serializer>]" at line 707 of "googletest/1.7.0.1/include/gtest/gtest-printers.h"
            instantiation of "void testing::internal::UniversalPrinter<T>::Print(const T &, std::ostream *) [with T=nlohmann::basic_json<std::map, std::vector, std::string, bool, int64_t={long}, uint64_t={unsigned long}, double, std::allocator, nlohmann::adl_serializer>]" at line 856 of "googletest/1.7.0.1/include/gtest/gtest-printers.h"
            instantiation of "void testing::internal::UniversalPrint(const T &, std::ostream *) [with T=nlohmann::basic_json<std::map, std::vector, std::string, bool, int64_t={long}, uint64_t={unsigned long}, double, std::allocator, nlohmann::adl_serializer>]" at line 784 of "googletest/1.7.0.1/include/gtest/gtest-printers.h"
            [ 2 instantiation contexts not shown ]
            instantiation of "std::string testing::internal::FormatForComparison<ToPrint, OtherOperand>::Format(const ToPrint &) [with ToPrint=nlohmann::basic_json<std::map, std::vector, std::string, bool, int64_t={long}, uint64_t={unsigned long}, double, std::allocator, nlohmann::adl_serializer>, OtherOperand=char [5]]" at line 351 of "googletest/1.7.0.1/include/gtest/gtest-printers.h"
            instantiation of "std::string testing::internal::FormatForComparisonFailureMessage(const T1 &, const T2 &) [with T1=nlohmann::basic_json<std::map, std::vector, std::string, bool, int64_t={long}, uint64_t={unsigned long}, double, std::allocator, nlohmann::adl_serializer>, T2=char [5]]" at line 1385 of "googletest/1.7.0.1/include/gtest/gtest.h"
            instantiation of "testing::AssertionResult testing::internal::CmpHelperEQFailure(const char *, const char *, const T1 &, const T2 &) [with T1=nlohmann::basic_json<std::map, std::vector, std::string, bool, int64_t={long}, uint64_t={unsigned long}, double, std::allocator, nlohmann::adl_serializer>, T2=char [5]]" at line 1403 of "googletest/1.7.0.1/include/gtest/gtest.h"
            instantiation of "testing::AssertionResult testing::internal::CmpHelperEQ(const char *, const char *, const T1 &, const T2 &) [with T1=nlohmann::basic_json<std::map, std::vector, std::string, bool, int64_t={long}, uint64_t={unsigned long}, double, std::allocator, nlohmann::adl_serializer>, T2=char [5]]" at line 1428 of "googletest/1.7.0.1/include/gtest/gtest.h"
            instantiation of "testing::AssertionResult testing::internal::EqHelper<lhs_is_null_literal>::Compare(const char *, const char *, const T1 &, const T2 &) [with lhs_is_null_literal=false, T1=nlohmann::basic_json<std::map, std::vector, std::string, bool, int64_t={long}, uint64_t={unsigned long}, double, std::allocator, nlohmann::adl_serializer>, T2=char [5]]" at line 74 of "hmd_lib/hmd_lib/utils/FusionToJson.t.cpp"

[theodelrieu]

Could you try editting the source file and replace the m_it = {} with m_it{}?

[mark-99]

Same :/

include/nlohmann/json.hpp(4177): error: the generated default constructor for "nlohmann::detail::internal_iterator<nlohmann::basic_json<std::map, std::vector, std::string, bool, int64_t={long}, uint64_t={unsigned long}, double, std::allocator, nlohmann::adl_serializer>>" cannot be used in an initializer for its own data member
      internal_iterator<typename std::remove_const<BasicJsonType>::type> m_it{};

[mark-99]

Removing the {} entirely works however - does this type actually need explicit value-init? It looks like the fields would be initialized anyway.

[theodelrieu]

Yep you're right, it does seem like an ICC bug, but we can simply remove the {}.

@nlohmann since this is a trivial fix, could it be part of 3.1.0?

[nlohmann]

Would we also need

internal_iterator() = default;

in template<typename BasicJsonType> struct internal_iterator?

Edit: No, I don't think so.

[theodelrieu]

I don't think this is needed.

[mark-99]

I don't think so... generally you only need that if the default ctor is not implicitly generated, e.g. due to the presence of another ctor with params, to force it to make one. It doesn't hurt, but the compiler will do it anyway.

[nlohmann]

@mark-99 Could you please check whether the latest commit fixed the issue?

[mark-99]

I have 3.0.1 rather than HEAD so the line numbers are different, but yes this seems good:
internal_iterator<typename std::remove_const<BasicJsonType>::type> m_it;

Verified on icc 16.0.4, icc 17.0.5, gcc 5.4, gcc 6.3, clang 3.9, clang 4.0.

BTW do you have a timescale on v3.1 release? So I know whether to go to the trouble of patching the copy in our local package manager?

[nlohmann]

If @theodelrieu finishes #944 tomorrow morning, then the release could be made tomorrow evening. :)

[mark-99]

Awesome - thx.

[nlohmann]

You're welcome!