Can you provide a more precise example? Show me your source code or something.

The entire code is too long, but this is the main part. The ISignal* was gotten from an IMessage->Signals().

    const dbcppp::ISignal* _sig_def;
    ...
    inline int64_t signal_id() const {
        for (const auto& a : _sig_def->AttributeValues()) {
            if (a.Name() == "SignalID") return std::get<1>(a.Value()); // 1 = FLOAT instead of 0 = INT
        }
        return -1;
    }

Runtime error with std::get<0>, correct result with std::get<1> (double). _sig_def->ExtendedValueType() returns 1 for FLOAT, when the .dbc file says the Attribute is an INT.
For strings, std::get<2> works as expected.

I checked the grammar parser, and it seems like this is due to an error in the grammar (here). The marked line says:

static const auto attribute_value_def = double_ | signed_int | quoted_string;

This line will make the grammar parser first try to parse the value as an double, and only if it fails, continue with trying to parse the value as int. However, since every int is a double, the parser always will parse int as double. Swaping double_ and signed_int should fix your specific problem. However, then a new appears: When the BA_DEF_ is defining the value as floating point but the parser is able to parse the BA_ value as int, the opposite of your error happens, what of course is an error as well. But since the "new error" is an error which must semantically be catched later (e.g. when parsing the AST), swaping is the right thing to do here.

Can you fix this in your local dbcppp version and confirm my theory?

Yes, I can confirm this fixes it. Thanks so much!

The tests are failing, you should use the dirty fix with care.