Object RDT

Object is a basic last-write-wins data type. An object has named fields which can be set to arbitrary JSON values. One op sets one field. Out of two edits of the same field, the one with greater timestamp wins. Hence, Object does field-by-field last-write-wins merge. To simplify op format and processing, the op name matches the field name. Hence, object fields have Base64 names up to 10 characters. Reserved op names can not be used as Object field names (on, off, ~, 0, error). Tilde can not be used in field names. There are no other restrictions on field name vocabulary.

A serialized Object op looks like this:

object.put(100, 200, "value")

turns into

/Object#TimeCreatd+author!TimeModifd+origin.fieldName    "value"

In addition to field name based access, Object also supports array-like access by integer indexes, either one- or two-dimensional. Indexed values are serialized using regular Base64 encoding (leading zeroes skipped) and ~ as an op name prefix.

matrix_object[1][65] = "cell"

turns into

/Object#TimeCreatd+author!TimeModifd+origin.~1+~11    "cell"

Note that single-indexed accesses read/write a vector which does not belong to the two-indexed matrix. Single-indexed accesses get serialized as .~0, .~1, .~3... while two-indexed turn into .~0+~0, .~1+~0... Indexed access supports vectors up to 2^30 cells (more than a billion). Matrices are up to 2^30 cells in each dimension.

Object state snapshot is serialized as normalized JSON map where keys are op names and values are values:

{"fieldName":"value","~0":12345,"~1+~11":"cell"}