Do not respect application overrides when saving the base class. (#7)

* Do not respect application overrides when saving the base class.

This is more predictable and avoids multiple calls to overriding methods when
traversing the class hierarchy via saveObject/readObject.

setup.cfg

@@ -49,8 +49,8 @@ python_requires = >=3.8
 # For more information, check out https://semver.org/.
 install_requires =
     importlib-metadata; python_version<"3.8"
-    dolomite-base>=0.2.4
-    dolomite-matrix>=0.2.1
+    dolomite-base>=0.3.0
+    dolomite-matrix>=0.3.0
     dolomite-ranges>=0.1.3
     summarizedexperiment>=0.4.4
 

src/dolomite_se/read_ranged_summarized_experiment.py

@@ -3,6 +3,7 @@
 import dolomite_base as dl
 from dolomite_base.read_object import read_object_registry
 from summarizedexperiment import RangedSummarizedExperiment
+from .read_summarized_experiment import read_summarized_experiment
 
 read_object_registry["ranged_summarized_experiment"] = (
     "dolomite_se.read_ranged_summarized_experiment"
@@ -34,8 +35,13 @@ def read_ranged_summarized_experiment(
         :py:class:`~summarizedexperiment.RangedSummarizedExperiment.RangedSummarizedExperiment`
         with file-backed arrays in the assays.
     """
-    metadata["type"] = "summarized_experiment"
-    se = dl.alt_read_object(path, metadata=metadata, **kwargs)
+
+    # We don't try to respect application overrides when loading the base
+    # instance. Application developers should just pretend that we copied the
+    # code from read_summarized_experiment, rather than trying to inject in
+    # custom code at this point, which gets too complicated - see the
+    # associated commentary for save_ranged_summarized_experiment.
+    se = read_summarized_experiment(path, metadata=metadata, **kwargs)
 
     rse = RangedSummarizedExperiment(
         assays=se.get_assays(),

src/dolomite_se/save_ranged_summarized_experiment.py

@@ -2,6 +2,7 @@
 
 import dolomite_base as dl
 from summarizedexperiment import RangedSummarizedExperiment, SummarizedExperiment
+from .save_summarized_experiment import save_summarized_experiment
 
 
 @dl.save_object.register
@@ -46,25 +47,33 @@ def save_ranged_summarized_experiment(
     if assay_args is None:
         assay_args = {}
 
-    # convert to SE
-    _se = SummarizedExperiment(
-        assays=x.get_assays(),
-        row_data=x.get_row_data(),
-        column_data=x.get_column_data(),
-        row_names=x.get_row_names(),
-        column_names=x.get_column_names(),
-        metadata=x.get_metadata(),
-    )
-    dl.alt_save_object(
-        _se, path, data_frame_args=data_frame_args, assay_args=assay_args, **kwargs
+    # We do not respond to application overrides for the SummarizedExperiment
+    # base class. Developers should just pretend that method copied all of the
+    # code from the save_SE function; this call is just an implementation
+    # detail, nothing special is done about the fact that it's the base class.
+    #
+    # This simplifies the dispatch and ensures that an override is only called
+    # once. Consider the alternative - namely, casting to the next subclass and
+    # then calling alt_save_object to respect the override. This would call
+    # the override's SE method repeatedly for every step from the subclass to
+    # SE. If the override's behavior is not idempotent, we have a problem.
+    # 
+    # So, if an application wants to set an override for all SEs, then it
+    # should register an SE method for alt_save_object and then call it. If the
+    # override is slightly different for particular SE subclasses, developers
+    # should just duplicate the common override logic in the alt_save_object
+    # methods for affected subclasses, rather than expecting some injection of
+    # the overriding method into the save_object dispatch hierarchy.
+    save_summarized_experiment(
+        x, path, data_frame_args=data_frame_args, assay_args=assay_args, **kwargs
     )
 
     # save row_ranges
     _ranges = x.get_row_ranges()
     if _ranges is not None:
         dl.alt_save_object(_ranges, path=os.path.join(path, "row_ranges"), **kwargs)
 
-    # Modify OBJECT
+    # modify OBJECT
     _info = dl.read_object_file(path)
     _info["ranged_summarized_experiment"] = {"version": "1.0"}
     dl.save_object_file(path, "ranged_summarized_experiment", _info)