allow option for deserializing with missing, extra, or renamed fields

[timotheecour]

happy to discuss design of this

this is one of the best features of protocol buffers, allowing one to be forward and backward compatible with schema changes.

Eg:

struct Dataset{int age; int height;}
Dataset(13).pack.writeToFile("dataset.mpck");

// later on, Dataset is updated but we still want to use the saved data
struct Dataset{int age; string address = "bar";}
auto dataset="dataset.mpck".read.unpack!Dataset;
// currently, error: incompatible type

• NOTE: error is actually uninformative, it should also show which fields are wrong

// desired behavior:
struct MsgpackOption{
 bool allow_missing_fields=false;
 bool allow_added_fields=false;
}

MsgpackOption opt={allow_missing_fields:true, allow_added_fields:true};
auto dataset="dataset.mpck".read.unpack!Dataset(opt);
assert(dataset==Dataset(13, "bar"));

• NOTE: should also work with reordered fields (eg: struct Dataset{int height;int age; }, possibly modulo an option bool allow_reordered_fields

[repeatedly]

Allowing missing / extra fields seems good for me. msgpack doesn't have a schema unlike protocol buffers, so it seems to work with only tail fields. I will try to implement it.

For reorder support, user should provide serialized data index to actual struct field index. For such case, msgpack seems to not fit...