eclipse-aascw / aas-core-codegen / 35297640074

90%

push

github

Name the Java XML writers without a kind enum (#712) (#712)

The XML writers were indexed by a ``_Kind`` enum: five members, three
mappings keyed by them, and a ``_kind_of`` which answered what a value is
written through.

    class _Kind(enum.Enum):
        STRINGIFIED = "stringified"
        BYTES = "bytes"
        ENUM = "IEnum"
        CLASS = "IClass"
        UNION = "IUnion"

    _VALUE_TYPE_BY_KIND = {_Kind.STRINGIFIED: "Object", ...}
    _ARGUMENT_TYPE_BY_KIND = {_Kind.STRINGIFIED: "?", ...}
    _CONTENT_WRITER_BY_KIND = {_Kind.STRINGIFIED: "writeStringifiedContent", ...}

We can ask the type annotation directly, through functions which
chain: a moniker names the writer, the value type follows the moniker,
and the argument type follows the value type.
``_written_leaf_moniker``, ``_written_value_type`` and
``_written_argument_type``, the twins of ``_serialized_leaf_moniker``,
``_serialized_value_type`` and ``_serialized_argument_type``.

The gating grows stricter on the way. ``_Needed`` used to keep a set of
the kinds written, so the two mappings and the gate all agreed only
because each was keyed on ``_kind_of``. It now keeps the names of the
shared content writers something calls, registered by the very
``_content_writer_name`` which emits those calls, so the gate and the
call can not be spelled apart. The set also stops carrying members no
one reads: ``CLASS`` and ``UNION`` could never enter it, as an instance
writes its own element and never reaches ``register_content``.

Two spellings are kept deliberately rather than derived. A scalar's
argument type stays the unbounded ``?`` instead of widening into
``? extends Object``, which is the same type to ``javac`` but not the
same text, and the four scalars still collapse onto one ``stringified``
moniker -- a scalar is written through ``toString``.

44 of 44 new or added lines in 1 file covered. (100.0%)

34666 of 40413 relevant lines covered (85.78%)

2.57 hits per line