feat: support metric custom column ordering and sub-setting

Fixes #177

fgpyo/util/inspect.py

@@ -332,7 +332,7 @@ def get_parser() -> partial:  # noqa: C901
         nonlocal type_
         nonlocal parsers
         try:
-            return functools.partial(parsers[type_])
+            return functools.partial(parsers[type_])  # type: ignore[misc]
         except KeyError as ex:
             if (
                 type_ in [str, int, float]

fgpyo/util/metric.py

@@ -111,6 +111,32 @@
    >>> Person(name=Name(first='john', last='doe'), age=42, address=None).formatted_values()
    ["first last", "42"]
 ```
+
+Re-ordering and sub-setting the columns when writing is supported by overriding the `_header()`
+method.  In the example below, the `weight` field is not written, but is optional to support reading
+the metric back in.  Also, the `name` and `age` columns are written in reverse order.
+
+```python
+   >>> from fgpyo.util.metric import Metric
+   >>> import attr
+   >>> @attr.s(auto_attribs=True, frozen=True)
+   ... class Person(Metric["Person"]):
+   ...     name: str
+   ...     age: int
+   ...     weight: Optional[int]
+   ...     @classmethod
+   ...     def _header(cls, field_types: Optional[List[FieldType]] = None) -> List[str]:
+   ...         return ["age", "name"]
+   >>> person = Person(name="john", age=42, weight=180)
+   >>> person.header()
+   ["age", "name"]
+   >>> list(person.values())
+   [42, "john"]
+   >>> Person.write(Path("/path/to/metrics.txt"), person)
+   >>> list(Person.read(Path("/path/to/metrics.txt")))
+   [Person(name="john", age=42, weight=None)]
+```
+
 """
 
 from abc import ABC
@@ -122,10 +148,13 @@
 from typing import Generic
 from typing import Iterator
 from typing import List
+from typing import Optional
 from typing import TypeVar
+from typing import final
 
 from fgpyo import io
 from fgpyo.util import inspect
+from fgpyo.util.inspect import FieldType
 
 MetricType = TypeVar("MetricType", bound="Metric")
 
@@ -141,19 +170,21 @@ class Metric(ABC, Generic[MetricType]):
     [`format_value()`][fgpyo.util.metric.Metric.format_value].
     """
 
-    def values(self) -> Iterator[Any]:
+    def values(self, _header: Optional[List[str]] = None) -> Iterator[Any]:
         """An iterator over attribute values in the same order as the header."""
-        for field in inspect.get_fields(self.__class__):  # type: ignore[arg-type]
-            yield getattr(self, field.name)
+        if _header is None:
+            _header = self.header()
+        for name in _header:
+            yield getattr(self, name)
 
-    def formatted_values(self) -> List[str]:
+    def formatted_values(self, _header: Optional[List[str]] = None) -> List[str]:
         """An iterator over formatted attribute values in the same order as the header."""
-        return [self.format_value(value) for value in self.values()]
+        return [self.format_value(value) for value in self.values(_header=_header)]
 
     @classmethod
     def _parsers(cls) -> Dict[type, Callable[[str], Any]]:
         """Mapping of type to a specific parser for that type.  The parser must accept a string
-        as a single parameter and return a single value of the given type.  Sub-classes may
+        as a single parameter and return a single value of the given type.  Subclasses may
         override this method to support custom types."""
         return {}
 
@@ -245,21 +276,45 @@ def write(cls, path: Path, *values: MetricType) -> None:
         Args:
             path: Path to the output file.
             values: Zero or more metrics.
-
         """
+        header = cls.header()
         with io.to_writer(path) as writer:
             writer.write("\t".join(cls.header()))
             writer.write("\n")
             for value in values:
                 # Important, don't recurse on nested attr classes, instead let implementing class
                 # implement format_value.
-                writer.write("\t".join(cls.format_value(item) for item in value.values()))
+                writer.write(
+                    "\t".join(cls.format_value(item) for item in value.values(_header=header))
+                )
                 writer.write("\n")
 
     @classmethod
+    @final
     def header(cls) -> List[str]:
-        """The list of header values for the metric."""
-        return [a.name for a in inspect.get_fields(cls)]  # type: ignore[arg-type]
+        """The list of header values for the metric.
+
+        This method may be overridden to re-order or subset the columns written to file with
+        `write()` or returned by `values()`.
+        """
+        field_types = list(inspect.get_fields(cls))  # type: ignore[arg-type]
+        field_names = {field.name for field in field_types}
+        header = cls._header(field_types=field_types)
+        extra_fields = [h for h in header if h not in field_names]
+        if len(extra_fields) > 0:
+            raise ValueError("header() returned extra fields: " + ", ".join(extra_fields))
+        return header
+
+    @classmethod
+    def _header(cls, field_types: Optional[List[FieldType]] = None) -> List[str]:
+        """Returns a mapping of a field to the index it should be returned in `values()`
+        based on the field names returned by `header()`.
+
+        This is useful for re-ordering or sub-setting the output columns in `write()`.
+        """
+        if field_types is None:
+            field_types = list(inspect.get_fields(cls))  # type: ignore[arg-type]
+        return [a.name for a in field_types]
 
     @classmethod
     def format_value(cls, value: Any) -> str:  # noqa: C901
@@ -315,7 +370,7 @@ def format_value(cls, value: Any) -> str:  # noqa: C901
 
     @classmethod
     def to_list(cls, value: str) -> List[Any]:
-        """Returns a list value split on comma delimeter."""
+        """Returns a list value split on comma delimiter."""
         return [] if value == "" else value.split(",")
 
     @staticmethod

tests/fgpyo/util/test_metric.py

@@ -26,6 +26,7 @@
 import attr
 import pytest
 
+from fgpyo.util.inspect import FieldType
 from fgpyo.util.inspect import is_attr_class
 from fgpyo.util.inspect import is_dataclasses_class
 from fgpyo.util.metric import Metric
@@ -121,6 +122,24 @@ class NameMetric(Metric["NameMetric"]):
             first: str
             last: str
 
+        @make_dataclass(use_attr=use_attr)
+        class LastFirstMetric(Metric["LastFirstMetric"]):
+            first: str
+            last: str
+
+            @classmethod
+            def _header(cls, field_types: Optional[List[FieldType]] = None) -> List[str]:
+                return ["last", "first"]
+
+        @make_dataclass(use_attr=use_attr)
+        class SubsetMetric(Metric["SubsetMetric"]):
+            written: str
+            hidden: Optional[str]
+
+            @classmethod
+            def _header(cls, field_types: Optional[List[FieldType]] = None) -> List[str]:
+                return ["written"]
+
         @make_dataclass(use_attr=use_attr)
         class NamedPerson(Metric["NamedPerson"]):
             name: Name
@@ -156,6 +175,8 @@ class ListPerson(Metric["ListPerson"]):
         self.Person = Person
         self.Name = Name
         self.NameMetric = NameMetric
+        self.LastFirstMetric = LastFirstMetric
+        self.SubsetMetric = SubsetMetric
         self.NamedPerson = NamedPerson
         self.PersonMaybeAge = PersonMaybeAge
         self.PersonDefault = PersonDefault
@@ -234,6 +255,10 @@ def test_is_correct_dataclass_type(use_attr: bool) -> None:
     assert is_dataclasses_class(data_and_classes.Name) is not use_attr
     assert is_attr_class(data_and_classes.NameMetric) is use_attr
     assert is_dataclasses_class(data_and_classes.NameMetric) is not use_attr
+    assert is_attr_class(data_and_classes.LastFirstMetric) is use_attr
+    assert is_dataclasses_class(data_and_classes.LastFirstMetric) is not use_attr
+    assert is_attr_class(data_and_classes.SubsetMetric) is use_attr
+    assert is_dataclasses_class(data_and_classes.SubsetMetric) is not use_attr
     assert is_attr_class(data_and_classes.NamedPerson) is use_attr
     assert is_dataclasses_class(data_and_classes.NamedPerson) is not use_attr
     assert is_attr_class(data_and_classes.PersonMaybeAge) is use_attr
@@ -368,6 +393,54 @@ def test_metric_read_missing_column_with_default(
         list(PersonDefault.read(path=path))
 
 
+@pytest.mark.parametrize("data_and_classes", (attr_data_and_classes, dataclasses_data_and_classes))
+def test_metric_read_different_column_order(tmp_path: Path, data_and_classes: DataBuilder) -> None:
+    NameMetric: TypeAlias = data_and_classes.NameMetric
+    name = NameMetric(first="Jane", last="Doe")
+    path = tmp_path / "metrics.txt"
+
+    with path.open("w") as writer:
+        writer.write("last\tfirst\n")
+        writer.write("Doe\tJane\n")
+    assert list(NameMetric.read(path=path)) == [name]
+
+
+@pytest.mark.parametrize("data_and_classes", (attr_data_and_classes, dataclasses_data_and_classes))
+def test_metric_write_different_column_order(tmp_path: Path, data_and_classes: DataBuilder) -> None:
+    LastFirstMetric: TypeAlias = data_and_classes.LastFirstMetric
+    name = LastFirstMetric(first="Jane", last="Doe")
+
+    assert LastFirstMetric.header() == ["last", "first"]
+    assert list(name.values()) == ["Doe", "Jane"]
+
+    path = tmp_path / "metrics.txt"
+    LastFirstMetric.write(path, name)
+    with path.open("r") as read:
+        header = read.readline().strip().split("\t")
+        fields = read.readline().strip().split("\t")
+
+    assert header == ["last", "first"]
+    assert fields == [name.last, name.first]
+
+
+@pytest.mark.parametrize("data_and_classes", (attr_data_and_classes, dataclasses_data_and_classes))
+def test_metric_write_subset(tmp_path: Path, data_and_classes: DataBuilder) -> None:
+    SubsetMetric: TypeAlias = data_and_classes.SubsetMetric
+    metric = SubsetMetric(written="present", hidden="absent")
+
+    assert SubsetMetric.header() == ["written"]
+    assert list(metric.values()) == ["present"]
+
+    path = tmp_path / "metrics.txt"
+    SubsetMetric.write(path, metric)
+    with path.open("r") as read:
+        header = read.readline().strip().split("\t")
+        fields = read.readline().strip().split("\t")
+
+    assert header == ["written"]
+    assert fields == [metric.written]
+
+
 @pytest.mark.parametrize("data_and_classes", (attr_data_and_classes, dataclasses_data_and_classes))
 def test_metric_header(data_and_classes: DataBuilder) -> None:
     assert data_and_classes.DummyMetric.header() == [