BUG: C++ constant value gets serialized

[arobenko]

Having the following enum definition

<enum name="Enum1" encodingType="uint8">
    <validValue name="Val1">0</validValue>
    <validValue name="Val2">1</validValue>
    <validValue name="Val3">2</validValue>
</enum> 

and message with single constant field

<message name="Msg1" id="1" description="TestMessage">
    <field name="field1" id="20" type="uint8" presence="constant" valueRef="Enum1.Val2"/>
</message>

generates code that serializes the field:

std::uint8_t field1() const
{
    return (*((std::uint8_t *)(m_buffer + m_offset + 0)));
}

Msg1 &field1(const std::uint8_t value)
{
    *((std::uint8_t *)(m_buffer + m_offset + 0)) = (value);
    return *this;
}

while realizing that the field is constant:

    static const char *field1MetaAttribute(const MetaAttribute::Attribute metaAttribute) SBE_NOEXCEPT
    {
        switch (metaAttribute)
        {
            case MetaAttribute::EPOCH: return "unix";
            case MetaAttribute::TIME_UNIT: return "nanosecond";
            case MetaAttribute::SEMANTIC_TYPE: return "";
            case MetaAttribute::PRESENCE: return "constant";   // <----- HERE
        }

        return "";
    }

The only difference from #528 is that the type of the field is specified "uint8" directly instead of "Enum1".

Reproduced with v1.7.4

[mjpt777]

The type provided for the constant is incorrect and thus was generating the wrong code. I've pushed a change to validate the types match.

[arobenko]

Hi there,
At this moment I'm working on my own "schema to C++11 only" compiler, which generates implementation of the protocol in terms of my COMMS Library. I'm using code generated by SBE as a reference implementation and in my unit-testing, to make sure that a message serialized by one generated code can be correctly deserialized by another. I assumed that any example encountered in the spec is a valid one and the compiler is expected to understand it. The spec contains the following one:

<composite name="UTCTimestampNanos" description="UTC timestamp with nanosecond precision" semanticType="UTCTimestamp" >
    <type name="time" primitiveType="uint64" />
    <type name="unit" primitiveType="uint8" presence="constant" valueRef="TimeUnit.nanosecond" />
</composite>

The "unit" is regular "uint8" constant type, but uses "valueRef" to specify its numeric value. I think other developers who read the spec may consider it to be valid and decide to use the same constructs in their schema. Maybe it's a good idea to support it in your compiler as well. I do support it in mine.