Implement "replace" transformation in Field operation module

docs/croissant-spec-draft.md

@@ -1088,6 +1088,7 @@ Croissant supports a few simple transformations that can be applied on the sourc
 - delimiter: split a string into an array using the supplied character.
 - regex: A regular expression to parse the data.
 - jsonPath: A JSON path to evaluate on the (JSON) data source.
+- replace: A string of the form `pattern/replacement` to replace occurrences of `pattern` with `replacement`. Additional forward slashes (`/`) can be escaped with a backslash, and `pattern` can be a regular expression.
 
 For example, to extract information from a filename using a regular expression, we can write:
 

python/mlcroissant/mlcroissant/_src/operation_graph/operations/field.py

@@ -63,6 +63,20 @@ def _apply_transform_fn(value: Any, transform: Transform, field: Field) -> Any:
             raise ValueError(f"`format` only applies to dates. Got {field.data_type}")
     elif transform.separator is not None:
         return value.split(transform.separator)
+    elif transform.replace is not None:
+        if isinstance(value, pathlib.PurePath):
+            value = os.fspath(value)
+        # Split on unescaped slash.
+        parts = re.split(r"(?<!\\)/", transform.replace)
+        if len(parts) != 2:
+            raise ValueError(
+                "`replace` must have exactly one unescaped slash. "
+                f"Got {transform.replace} which has "
+                f"{len(parts) - 1} unescaped slashes."
+            )
+        parts = [part.replace("\\/", "/") for part in parts]
+        pattern, replacement = parts
+        return re.sub(pattern, replacement, value)
     return value
 
 

python/mlcroissant/mlcroissant/_src/operation_graph/operations/field_test.py

@@ -372,6 +372,27 @@ def test_extract_lines(separator):
             "2024-12-10",
             False,
         ],
+        [
+            "foo",
+            Source(transforms=[Transform(replace="foo/bar")]),
+            DataType.TEXT,
+            "bar",
+            False,
+        ],
+        [
+            "path/to/a/file",
+            Source(transforms=[Transform(replace=r"path\/to\/a/some\/new\/path")]),
+            DataType.TEXT,
+            "some/new/path/file",
+            False,
+        ],
+        [
+            "123",
+            Source(transforms=[Transform(replace=r"[0-9]+([0-9])/abc\1")]),
+            DataType.TEXT,
+            "abc3",
+            False,
+        ],
     ],
 )
 def test_apply_transforms_fn(value, source, data_type, expected_value, repeated):
@@ -381,6 +402,38 @@ def test_apply_transforms_fn(value, source, data_type, expected_value, repeated)
     assert field.apply_transforms_fn(value, f) == expected_value
 
 
+@pytest.mark.parametrize(
+    ["replace", "unescaped_slashes"],
+    [
+        ["two/unescaped/slashes", 2],
+        [r"no\/unescaped\/slashes", 0],
+        [r"f/o/u/r/\/unescaped\/slashes", 4],
+    ],
+)
+def test_apply_replace_exception_fn(replace, unescaped_slashes):
+    f = Field(
+        id="test",
+        name="test",
+        data_types=DataType.TEXT,
+        source=Source(transforms=[Transform(replace=replace)]),
+        repeated=False,
+    )
+
+    has_error = False
+    try:
+        field.apply_transforms_fn("foo", f)
+    except ValueError as e:
+        has_error = True
+        assert (
+            e.args[0]
+            == "`replace` must have exactly one unescaped slash. "
+            f"Got {replace} which has "
+            f"{unescaped_slashes} unescaped slashes."
+        )
+
+    assert has_error
+
+
 def test_apply_multiple_transforms_fn():
     source = Source(
         transforms=[