linkml 1.8.1__py3-none-any.whl → 1.8.2__py3-none-any.whl

linkml/generators/pydanticgen/pydanticgen.py

@@ -1,20 +1,20 @@
 import inspect
 import logging
 import os
+import re
 import textwrap
 from collections import defaultdict
-from copy import copy, deepcopy
 from dataclasses import dataclass, field
 from enum import Enum
 from pathlib import Path
 from types import ModuleType
-from typing import Dict, List, Literal, Optional, Set, Type, TypeVar, Union, overload
+from typing import ClassVar, Dict, List, Literal, Optional, Set, Tuple, Type, TypeVar, Union, overload
 
 import click
-from jinja2 import ChoiceLoader, Environment, FileSystemLoader
+from jinja2 import ChoiceLoader, Environment, FileSystemLoader, Template
 from linkml_runtime.linkml_model.meta import (
-    Annotation,
     ClassDefinition,
+    ElementName,
     SchemaDefinition,
     SlotDefinition,
     TypeDefinition,
@@ -25,13 +25,13 @@ from linkml_runtime.utils.schemaview import SchemaView
 from pydantic.version import VERSION as PYDANTIC_VERSION
 
 from linkml._version import __version__
+from linkml.generators.common.lifecycle import LifecycleMixin
 from linkml.generators.common.type_designators import get_accepted_type_designator_values, get_type_designator_value
 from linkml.generators.oocodegen import OOCodeGenerator
 from linkml.generators.pydanticgen import includes
 from linkml.generators.pydanticgen.array import ArrayRangeGenerator, ArrayRepresentation
-from linkml.generators.pydanticgen.build import SlotResult
+from linkml.generators.pydanticgen.build import ClassResult, SlotResult, SplitResult
 from linkml.generators.pydanticgen.template import (
-    ConditionalImport,
     Import,
     Imports,
     ObjectImport,
@@ -39,7 +39,7 @@ from linkml.generators.pydanticgen.template import (
     PydanticBaseModel,
     PydanticClass,
     PydanticModule,
-    TemplateModel,
+    PydanticTemplateModel,
 )
 from linkml.utils import deprecation_warning
 from linkml.utils.generator import shared_arguments
@@ -84,9 +84,7 @@ DEFAULT_IMPORTS = (
             ObjectImport(name="Union"),
         ],
     )
-    + Import(module="pydantic.version", objects=[ObjectImport(name="VERSION", alias="PYDANTIC_VERSION")])
-    + ConditionalImport(
-        condition="int(PYDANTIC_VERSION[0])>=2",
+    + Import(
         module="pydantic",
         objects=[
             ObjectImport(name="BaseModel"),
@@ -95,20 +93,16 @@ DEFAULT_IMPORTS = (
             ObjectImport(name="RootModel"),
             ObjectImport(name="field_validator"),
         ],
-        alternative=Import(
-            module="pydantic",
-            objects=[ObjectImport(name="BaseModel"), ObjectImport(name="Field"), ObjectImport(name="validator")],
-        ),
     )
 )
 
-DEFAULT_INJECTS = {1: [includes.LinkMLMeta_v1], 2: [includes.LinkMLMeta_v2]}
+DEFAULT_INJECTS = [includes.LinkMLMeta]
 
 
 class MetadataMode(str, Enum):
     FULL = "full"
     """
-    all metadata from the source schema will be included, even if it is represented by the template classes, 
+    all metadata from the source schema will be included, even if it is represented by the template classes,
     and even if it is represented by some child class (eg. "classes" will be included with schema metadata
     """
     EXCEPT_CHILDREN = "except_children"
@@ -118,7 +112,7 @@ class MetadataMode(str, Enum):
     """
     AUTO = "auto"
     """
-    Only the metadata that isn't represented by the template classes or excluded with ``meta_exclude`` will be included 
+    Only the metadata that isn't represented by the template classes or excluded with ``meta_exclude`` will be included
     """
     NONE = None
     """
@@ -126,16 +120,55 @@ class MetadataMode(str, Enum):
     """
 
 
+class SplitMode(str, Enum):
+    FULL = "full"
+    """
+    Import all classes defined in imported schemas
+    """
+
+    AUTO = "auto"
+    """
+    Only import those classes that are actually used in the generated schema as
+
+    * parents (``is_a``)
+    * mixins
+    * slot ranges
+    """
+
+
 DefinitionType = TypeVar("DefinitionType", bound=Union[SchemaDefinition, ClassDefinition, SlotDefinition])
 TemplateType = TypeVar("TemplateType", bound=Union[PydanticModule, PydanticClass, PydanticAttribute])
 
 
 @dataclass
-class PydanticGenerator(OOCodeGenerator):
+class PydanticGenerator(OOCodeGenerator, LifecycleMixin):
     """
     Generates Pydantic-compliant classes from a schema
 
     This is an alternative to the dataclasses-based Pythongen
+
+    Lifecycle methods (see :class:`.LifecycleMixin` ) supported:
+
+    * :meth:`~.LifecycleMixin.before_generate_enums`
+
+    Slot generation is nested within class generation, since the pydantic generator currently doesn't
+    create an independent representation of slots aside from their materialization as class fields.
+    Accordingly, the ``before_`` and ``after_generate_slots`` are called before and after each class's
+    slot generation, rather than all slot generation.
+
+    * :meth:`~.LifecycleMixin.before_generate_classes`
+    * :meth:`~.LifecycleMixin.before_generate_class`
+    * :meth:`~.LifecycleMixin.after_generate_class`
+    * :meth:`~.LifecycleMixin.after_generate_classes`
+
+    * :meth:`~.LifecycleMixin.before_generate_slots`
+    * :meth:`~.LifecycleMixin.before_generate_slot`
+    * :meth:`~.LifecycleMixin.after_generate_slot`
+    * :meth:`~.LifecycleMixin.after_generate_slots`
+
+    * :meth:`~.LifecycleMixin.before_render_template`
+    * :meth:`~.LifecycleMixin.after_render_template`
+
     """
 
     # ClassVar overrides
@@ -150,12 +183,12 @@ class PydanticGenerator(OOCodeGenerator):
     """
     If black is present in the environment, format the serialized code with it
     """
-    pydantic_version: int = int(PYDANTIC_VERSION[0])
+
     template_dir: Optional[Union[str, Path]] = None
     """
-    Override templates for each TemplateModel.
+    Override templates for each PydanticTemplateModel.
     
-    Directory with templates that override the default :attr:`.TemplateModel.template`
+    Directory with templates that override the default :attr:`.PydanticTemplateModel.template`
     for each class. If a matching template is not found in the override directory,
     the default templates will be used.
     """
@@ -164,62 +197,62 @@ class PydanticGenerator(OOCodeGenerator):
     injected_classes: Optional[List[Union[Type, str]]] = None
     """
     A list/tuple of classes to inject into the generated module.
-    
+
     Accepts either live classes or strings. Live classes will have their source code
     extracted with inspect.get - so they need to be standard python classes declared in a
-    source file (ie. the module they are contained in needs a ``__file__`` attr, 
+    source file (ie. the module they are contained in needs a ``__file__`` attr,
     see: :func:`inspect.getsource` )
     """
     injected_fields: Optional[List[str]] = None
     """
     A list/tuple of field strings to inject into the base class.
-    
+
     Examples:
-    
+
     .. code-block:: python
 
         injected_fields = (
             'object_id: Optional[str] = Field(None, description="Unique UUID for each object")',
         )
-    
+
     """
     imports: Optional[List[Import]] = None
     """
-    Additional imports to inject into generated module. 
-    
+    Additional imports to inject into generated module.
+
     Examples:
-        
+
     .. code-block:: python
-    
+
         from linkml.generators.pydanticgen.template import (
             ConditionalImport,
             ObjectImport,
             Import,
             Imports
         )
-        
-        imports = (Imports() + 
-            Import(module='sys') + 
-            Import(module='numpy', alias='np') + 
+
+        imports = (Imports() +
+            Import(module='sys') +
+            Import(module='numpy', alias='np') +
             Import(module='pathlib', objects=[
                 ObjectImport(name="Path"),
                 ObjectImport(name="PurePath", alias="RenamedPurePath")
-            ]) + 
+            ]) +
             ConditionalImport(
                 module="typing",
                 objects=[ObjectImport(name="Literal")],
                 condition="sys.version_info >= (3, 8)",
                 alternative=Import(
-                    module="typing_extensions", 
+                    module="typing_extensions",
                     objects=[ObjectImport(name="Literal")]
                 ),
             ).imports
         )
-        
+
     becomes:
-    
+
     .. code-block:: python
-    
+
         import sys
         import numpy as np
         from pathlib import (
@@ -230,14 +263,72 @@ class PydanticGenerator(OOCodeGenerator):
             from typing import Literal
         else:
             from typing_extensions import Literal
-        
+
     """
     metadata_mode: Union[MetadataMode, str, None] = MetadataMode.AUTO
     """
     How to include schema metadata in generated pydantic models.
-    
+
     See :class:`.MetadataMode` for mode documentation
     """
+    split: bool = False
+    """
+    Generate schema that import other schema as separate python modules
+    that import from one another, rather than rolling all into a single
+    module (default, ``False``).
+    """
+    split_pattern: str = ".{{ schema.name }}"
+    """
+    When splitting generation, imported modules need to be generated separately
+    and placed in a python package and import from each other. Since the
+    location of those imported modules is variable -- e.g. one might want to
+    generate schema in multiple packages depending on their version -- this
+    pattern is used to generate the module portion of the import statement.
+
+    These patterns should generally yield a relative module import,
+    since functions like :func:`.generate_split` will generate and write files
+    relative to some base file, though this is not a requirement since custom
+    split generation logic is also allowed.
+
+    The pattern is a jinja template string that is given the ``SchemaDefinition``
+    of the imported schema in the environment. Additional variables can be passed
+    into the jinja environment with the :attr:`.split_context` argument.
+
+    Further modification is possible by using jinja filters.
+
+    After templating, the string is passed through a :attr:`SNAKE_CASE` pattern
+    to replace whitespace and other characters that can't be used in module names.
+
+    See also :meth:`.generate_module_import`, which is used to generate the
+    module portion of the import statement (and can be overridden in subclasses).
+
+    Examples:
+
+        for a schema named ``ExampleSchema`` and version ``1.2.3`` ...
+
+        ``".{{ schema.name }}"`` (the default) becomes
+
+        ``from .example_schema import ClassA, ...``
+
+        ``"...{{ schema.name }}.v{{ schema.version | replace('.', '_') }}"`` becomes
+
+        ``from ...example_schema.v1_2_3 import ClassA, ...``
+
+    """
+    split_context: Optional[dict] = None
+    """
+    Additional variables to pass into ``split_pattern`` when
+    generating imported module names.
+
+    Passed in as ``**kwargs`` , so e.g. if ``split_context = {'myval': 1}``
+    then one would use it in a template string like ``{{ myval }}``
+    """
+    split_mode: SplitMode = SplitMode.AUTO
+    """
+    How to filter imports from imported schema.
+
+    See :class:`.SplitMode` for description of options
+    """
 
     # ObjectVars (identical to pythongen)
     gen_classvars: bool = True
@@ -245,10 +336,16 @@ class PydanticGenerator(OOCodeGenerator):
     genmeta: bool = False
     emit_metadata: bool = True
 
+    # ClassVars
+    SNAKE_CASE: ClassVar[str] = r"(((?<!^)(?<!\.))(?=[A-Z][a-z]))|([^\w\.]+)"
+    """Substitute CamelCase and non-word characters with _"""
+
+    # Private attributes
+    _predefined_slot_values: Optional[Dict[str, Dict[str, str]]] = None
+    _class_bases: Optional[Dict[str, List[str]]] = None
+
     def __post_init__(self):
         super().__post_init__()
-        if int(self.pydantic_version) == 1:
-            deprecation_warning("pydanticgen-v1")
 
     def compile_module(self, **kwargs) -> ModuleType:
         """
@@ -263,8 +360,20 @@ class PydanticGenerator(OOCodeGenerator):
             logging.error(f"Error compiling generated python code: {e}")
             raise e
 
+    def _get_classes(self, sv: SchemaView) -> Tuple[List[ClassDefinition], Optional[List[ClassDefinition]]]:
+        all_classes = sv.all_classes(imports=True).values()
+
+        if self.split:
+            local_classes = sv.all_classes(imports=False).values()
+            imported_classes = [c for c in all_classes if c not in local_classes]
+            return list(local_classes), imported_classes
+        else:
+            return list(all_classes), None
+
     @staticmethod
-    def sort_classes(clist: List[ClassDefinition]) -> List[ClassDefinition]:
+    def sort_classes(
+        clist: List[ClassDefinition], imported: Optional[List[ClassDefinition]] = None
+    ) -> List[ClassDefinition]:
         """
         sort classes such that if C is a child of P then C appears after P in the list
 
@@ -272,6 +381,9 @@ class PydanticGenerator(OOCodeGenerator):
 
         TODO: This should move to SchemaView
         """
+        if imported is not None:
+            imported = [i.name for i in imported]
+
         clist = list(clist)
         slist = []  # sorted
         while len(clist) > 0:
@@ -283,6 +395,11 @@ class PydanticGenerator(OOCodeGenerator):
                     candidates = [candidate.is_a] + candidate.mixins
                 else:
                     candidates = candidate.mixins
+
+                # remove blocking classes imported from other schemas if in split mode
+                if imported:
+                    candidates = [c for c in candidates if c not in imported]
+
                 if not candidates:
                     can_add = True
                 else:
@@ -296,82 +413,165 @@ class PydanticGenerator(OOCodeGenerator):
                 raise ValueError(f"could not find suitable element in {clist} that does not ref {slist}")
         return slist
 
-    def get_predefined_slot_values(self) -> Dict[str, Dict[str, str]]:
+    def generate_class(self, cls: ClassDefinition) -> ClassResult:
+        pyclass = PydanticClass(
+            name=camelcase(cls.name),
+            bases=self.class_bases.get(camelcase(cls.name), PydanticBaseModel.default_name),
+            description=cls.description.replace('"', '\\"') if cls.description is not None else None,
+        )
+
+        imports = self._get_imports(cls) if self.split else None
+
+        result = ClassResult(cls=pyclass, source=cls, imports=imports)
+
+        # Gather slots
+        slots = [self.schemaview.induced_slot(sn, cls.name) for sn in self.schemaview.class_slots(cls.name)]
+        slots = self.before_generate_slots(slots, self.schemaview)
+
+        slot_results = []
+        for slot in slots:
+            slot = self.before_generate_slot(slot, self.schemaview)
+            slot = self.generate_slot(slot, cls)
+            slot = self.after_generate_slot(slot, self.schemaview)
+            slot_results.append(slot)
+            result = result.merge(slot)
+
+        slot_results = self.after_generate_slots(slot_results, self.schemaview)
+        attributes = {slot.attribute.name: slot.attribute for slot in slot_results}
+
+        result.cls.attributes = attributes
+        result.cls = self.include_metadata(result.cls, cls)
+
+        return result
+
+    def generate_slot(self, slot: SlotDefinition, cls: ClassDefinition) -> SlotResult:
+        slot_args = {
+            k: slot._as_dict.get(k, None)
+            for k in PydanticAttribute.model_fields.keys()
+            if slot._as_dict.get(k, None) is not None
+        }
+        slot_args["name"] = underscore(slot.name)
+        slot_args["description"] = slot.description.replace('"', '\\"') if slot.description is not None else None
+        predef = self.predefined_slot_values.get(camelcase(cls.name), {}).get(slot.name, None)
+        if predef is not None:
+            slot_args["predefined"] = str(predef)
+
+        pyslot = PydanticAttribute(**slot_args)
+        pyslot = self.include_metadata(pyslot, slot)
+
+        slot_ranges = []
+        # Confirm that the original slot range (ignoring the default that comes in from
+        # induced_slot) isn't in addition to setting any_of
+        any_of_ranges = [a.range if a.range else slot.range for a in slot.any_of]
+        if any_of_ranges:
+            # list comprehension here is pulling ranges from within AnonymousSlotExpression
+            slot_ranges.extend(any_of_ranges)
+        else:
+            slot_ranges.append(slot.range)
+
+        pyranges = [self.generate_python_range(slot_range, slot, cls) for slot_range in slot_ranges]
+
+        pyranges = list(set(pyranges))  # remove duplicates
+        pyranges.sort()
+
+        if len(pyranges) == 1:
+            pyrange = pyranges[0]
+        elif len(pyranges) > 1:
+            pyrange = f"Union[{', '.join(pyranges)}]"
+        else:
+            raise Exception(f"Could not generate python range for {cls.name}.{slot.name}")
+
+        pyslot.range = pyrange
+
+        imports = self._get_imports(slot) if self.split else None
+
+        result = SlotResult(attribute=pyslot, source=slot, imports=imports)
+
+        if slot.array is not None:
+            results = self.get_array_representations_range(slot, result.attribute.range)
+            if len(results) == 1:
+                result.attribute.range = results[0].range
+            else:
+                result.attribute.range = f"Union[{', '.join([res.range for res in results])}]"
+            for res in results:
+                result = result.merge(res)
+
+        elif slot.multivalued:
+            if slot.inlined or slot.inlined_as_list:
+                collection_key = self.generate_collection_key(slot_ranges, slot, cls)
+            else:
+                collection_key = None
+            if slot.inlined is False or collection_key is None or slot.inlined_as_list is True:
+                result.attribute.range = f"List[{result.attribute.range}]"
+            else:
+                simple_dict_value = None
+                if len(slot_ranges) == 1:
+                    simple_dict_value = self._inline_as_simple_dict_with_value(slot)
+                if simple_dict_value:
+                    # simple_dict_value might be the range of the identifier of a class when range is a class,
+                    # so we specify either that identifier or the range itself
+                    if simple_dict_value != result.attribute.range:
+                        simple_dict_value = f"Union[{simple_dict_value}, {result.attribute.range}]"
+                    result.attribute.range = f"Dict[str, {simple_dict_value}]"
+                else:
+                    result.attribute.range = f"Dict[{collection_key}, {result.attribute.range}]"
+        if not (slot.required or slot.identifier or slot.key) and not slot.designates_type:
+            result.attribute.range = f"Optional[{result.attribute.range}]"
+        return result
+
+    @property
+    def predefined_slot_values(self) -> Dict[str, Dict[str, str]]:
         """
         :return: Dictionary of dictionaries with predefined slot values for each class
         """
-        sv = self.schemaview
-        slot_values = defaultdict(dict)
-        for class_def in sv.all_classes().values():
-            for slot_name in sv.class_slots(class_def.name):
-                slot = sv.induced_slot(slot_name, class_def.name)
-                if slot.designates_type:
-                    target_value = get_type_designator_value(sv, slot, class_def)
-                    slot_values[camelcase(class_def.name)][slot.name] = f'"{target_value}"'
-                    if slot.multivalued:
-                        slot_values[camelcase(class_def.name)][slot.name] = (
-                            "[" + slot_values[camelcase(class_def.name)][slot.name] + "]"
-                        )
-                    slot_values[camelcase(class_def.name)][slot.name] = slot_values[camelcase(class_def.name)][
-                        slot.name
-                    ]
-                elif slot.ifabsent is not None:
-                    value = ifabsent_value_declaration(slot.ifabsent, sv, class_def, slot)
-                    slot_values[camelcase(class_def.name)][slot.name] = value
-                # Multivalued slots that are either not inlined (just an identifier) or are
-                # inlined as lists should get default_factory list, if they're inlined but
-                # not as a list, that means a dictionary
-                elif "linkml:elements" in slot.implements:
-                    slot_values[camelcase(class_def.name)][slot.name] = None
-                elif slot.multivalued:
-                    has_identifier_slot = self.range_class_has_identifier_slot(slot)
-
-                    if slot.inlined and not slot.inlined_as_list and has_identifier_slot:
-                        slot_values[camelcase(class_def.name)][slot.name] = "default_factory=dict"
-                    else:
-                        slot_values[camelcase(class_def.name)][slot.name] = "default_factory=list"
-
-        return slot_values
-
-    def range_class_has_identifier_slot(self, slot):
-        """
-        Check if the range class of a slot has an identifier slot, via both slot.any_of and slot.range
-        Should return False if the range is not a class, and also if the range is a class but has no
-        identifier slot
-
-        :param slot: SlotDefinition
-        :return: bool
-        """
-        sv = self.schemaview
-        has_identifier_slot = False
-        if slot.any_of:
-            for slot_range in slot.any_of:
-                any_of_range = slot_range.range
-                if any_of_range in sv.all_classes() and sv.get_identifier_slot(any_of_range, use_key=True) is not None:
-                    has_identifier_slot = True
-        if slot.range in sv.all_classes() and sv.get_identifier_slot(slot.range, use_key=True) is not None:
-            has_identifier_slot = True
-        return has_identifier_slot
-
-    def get_class_isa_plus_mixins(self) -> Dict[str, List[str]]:
+        if self._predefined_slot_values is None:
+            sv = self.schemaview
+            slot_values = defaultdict(dict)
+            for class_def in sv.all_classes().values():
+                for slot_name in sv.class_slots(class_def.name):
+                    slot = sv.induced_slot(slot_name, class_def.name)
+                    if slot.designates_type:
+                        target_value = get_type_designator_value(sv, slot, class_def)
+                        slot_values[camelcase(class_def.name)][slot.name] = f'"{target_value}"'
+                        if slot.multivalued:
+                            slot_values[camelcase(class_def.name)][slot.name] = (
+                                "[" + slot_values[camelcase(class_def.name)][slot.name] + "]"
+                            )
+                        slot_values[camelcase(class_def.name)][slot.name] = slot_values[camelcase(class_def.name)][
+                            slot.name
+                        ]
+                    elif slot.ifabsent is not None:
+                        value = ifabsent_value_declaration(slot.ifabsent, sv, class_def, slot)
+                        slot_values[camelcase(class_def.name)][slot.name] = value
+
+            self._predefined_slot_values = slot_values
+
+        return self._predefined_slot_values
+
+    @property
+    def class_bases(self) -> Dict[str, List[str]]:
         """
         Generate the inheritance list for each class from is_a plus mixins
         :return:
         """
-        sv = self.schemaview
-        parents = {}
-        for class_def in sv.all_classes().values():
-            class_parents = []
-            if class_def.is_a:
-                class_parents.append(camelcase(class_def.is_a))
-            if self.gen_mixin_inheritance and class_def.mixins:
-                class_parents.extend([camelcase(mixin) for mixin in class_def.mixins])
-            if len(class_parents) > 0:
-                # Use the sorted list of classes to order the parent classes, but reversed to match MRO needs
-                class_parents.sort(key=lambda x: self.sorted_class_names.index(x))
-                class_parents.reverse()
-                parents[camelcase(class_def.name)] = class_parents
-        return parents
+        if self._class_bases is None:
+            sv = self.schemaview
+            parents = {}
+            for class_def in sv.all_classes().values():
+                class_parents = []
+                if class_def.is_a:
+                    class_parents.append(camelcase(class_def.is_a))
+                if self.gen_mixin_inheritance and class_def.mixins:
+                    class_parents.extend([camelcase(mixin) for mixin in class_def.mixins])
+                if len(class_parents) > 0:
+                    # Use the sorted list of classes to order the parent classes, but reversed to match MRO needs
+                    class_parents.sort(
+                        key=lambda x: self.sorted_class_names.index(x) if x in self.sorted_class_names else -1
+                    )
+                    class_parents.reverse()
+                    parents[camelcase(class_def.name)] = class_parents
+            self._class_bases = parents
+        return self._class_bases
 
     def get_mixin_identifier_range(self, mixin) -> str:
         sv = self.schemaview
@@ -442,6 +642,10 @@ class PydanticGenerator(OOCodeGenerator):
                 + ",".join(['"' + x + '"' for x in get_accepted_type_designator_values(sv, slot_def, class_def)])
                 + "]"
             )
+        elif slot_def.equals_string:
+            pyrange = f'Literal["{slot_def.equals_string}"]'
+        elif slot_def.equals_string_in:
+            pyrange = "Literal[" + ", ".join([f'"{a_string}"' for a_string in slot_def.equals_string_in]) + "]"
         elif slot_range in sv.all_classes():
             pyrange = self.get_class_slot_range(
                 slot_range,
@@ -497,8 +701,18 @@ class PydanticGenerator(OOCodeGenerator):
             return list(collection_keys)[0]
         return None
 
-    @staticmethod
-    def _inline_as_simple_dict_with_value(slot_def: SlotDefinition, sv: SchemaView) -> Optional[str]:
+    def _clean_injected_classes(self, injected_classes: List[Union[str, Type]]) -> Optional[List[str]]:
+        """Get source, deduplicate, and dedent injected classes"""
+        if len(injected_classes) == 0:
+            return None
+
+        injected_classes = list(
+            dict.fromkeys([c if isinstance(c, str) else inspect.getsource(c) for c in injected_classes])
+        )
+        injected_classes = [textwrap.dedent(c) for c in injected_classes]
+        return injected_classes
+
+    def _inline_as_simple_dict_with_value(self, slot_def: SlotDefinition) -> Optional[str]:
         """
         Determine if a slot should be inlined as a simple dict with a value.
 
@@ -521,21 +735,21 @@ class PydanticGenerator(OOCodeGenerator):
         :return: str
         """
         if slot_def.inlined and not slot_def.inlined_as_list:
-            if slot_def.range in sv.all_classes():
-                id_slot = sv.get_identifier_slot(slot_def.range, use_key=True)
+            if slot_def.range in self.schemaview.all_classes():
+                id_slot = self.schemaview.get_identifier_slot(slot_def.range, use_key=True)
                 if id_slot is not None:
-                    range_cls_slots = sv.class_induced_slots(slot_def.range)
+                    range_cls_slots = self.schemaview.class_induced_slots(slot_def.range)
                     if len(range_cls_slots) == 2:
                         non_id_slots = [slot for slot in range_cls_slots if slot.name != id_slot.name]
                         if len(non_id_slots) == 1:
                             value_slot = non_id_slots[0]
-                            value_slot_range_type = sv.get_type(value_slot.range)
+                            value_slot_range_type = self.schemaview.get_type(value_slot.range)
                             if value_slot_range_type is not None:
-                                return _get_pyrange(value_slot_range_type, sv)
+                                return _get_pyrange(value_slot_range_type, self.schemaview)
         return None
 
     def _template_environment(self) -> Environment:
-        env = TemplateModel.environment()
+        env = PydanticTemplateModel.environment()
         if self.template_dir is not None:
             loader = ChoiceLoader([FileSystemLoader(self.template_dir), env.loader])
             env.loader = loader
@@ -548,7 +762,7 @@ class PydanticGenerator(OOCodeGenerator):
         array_reps = []
         for repr in self.array_representations:
             generator = ArrayRangeGenerator.get_generator(repr)
-            result = generator(slot.array, range, self.pydantic_version).make()
+            result = generator(slot.array, range).make()
             array_reps.append(result)
 
         if len(array_reps) == 0:
@@ -572,7 +786,7 @@ class PydanticGenerator(OOCodeGenerator):
         Metadata inclusion mode is dependent on :attr:`.metadata_mode` - see:
 
         - :class:`.MetadataMode`
-        - :meth:`.TemplateModel.exclude_from_meta`
+        - :meth:`.PydanticTemplateModel.exclude_from_meta`
 
         """
         if self.metadata_mode is None or self.metadata_mode == MetadataMode.NONE:
@@ -591,13 +805,15 @@ class PydanticGenerator(OOCodeGenerator):
                     continue
 
                 model_attr = getattr(model, k)
-                if isinstance(model_attr, list) and not any([isinstance(item, TemplateModel) for item in model_attr]):
+                if isinstance(model_attr, list) and not any(
+                    [isinstance(item, PydanticTemplateModel) for item in model_attr]
+                ):
                     meta[k] = v
                 elif isinstance(model_attr, dict) and not any(
-                    [isinstance(item, TemplateModel) for item in model_attr.values()]
+                    [isinstance(item, PydanticTemplateModel) for item in model_attr.values()]
                 ):
                     meta[k] = v
-                elif not isinstance(model_attr, (list, dict, TemplateModel)):
+                elif not isinstance(model_attr, (list, dict, PydanticTemplateModel)):
                     meta[k] = v
 
         elif self.metadata_mode in (MetadataMode.FULL, MetadataMode.FULL.value):
@@ -611,155 +827,145 @@ class PydanticGenerator(OOCodeGenerator):
         model.meta = meta
         return model
 
+    def _get_imports(self, element: Union[ClassDefinition, SlotDefinition, None] = None) -> Imports:
+        """
+        Get imports that are implied by their usage in slots or classes
+        (and thus need to be imported when generating schemas in :attr:`.split` == ``True`` mode).
+
+        **Note:**
+        Since in pydantic (currently) the only things that are materialized are classes, we don't
+        import class slots from imported schemas and abandon slots, directly expressing them
+        in the model.
+
+        This is a parent placeholder method in case that changes, "give me something and return
+        a set of imports" that calls subordinate methods. If slots become materialized, keep
+        this as the directly called method rather than spaghetti-ing out another
+        independent method. This method is also isolated in anticipation of structured imports,
+        where we will need to revise our expectations of what is imported when.
+
+        Args:
+            element (:class:`.ClassDefinition` , :class:`.SlotDefinition` , None): The element
+                to get import for. If ``None`` , get all needed imports (see :attr:`.split_mode`
+        """
+        # import from local references, rather than serializing every class in every file
+        if not self.split or (self.split_mode == SplitMode.FULL and element is not None):
+            # we are either compiling this whole thing in one big file (default)
+            # or going to import all classes from the imported schemas,
+            # so we don't import anything
+            return Imports()
+
+        # gather a list of class names,
+        # remove local classes and transform to Imports later.
+        needed_classes = []
+
+        # fine to call rather than pass bc it's cached
+        all_classes = self.schemaview.all_classes(imports=True)
+        local_classes = self.schemaview.all_classes(imports=False)
+
+        if isinstance(element, ClassDefinition):
+            if element.is_a:
+                needed_classes.append(element.is_a)
+            if element.mixins:
+                needed_classes.extend(element.mixins)
+
+        elif isinstance(element, SlotDefinition):
+            # collapses `slot.range`, `slot.any_of`, and `slot.one_of` to a list
+            slot_ranges = self.schemaview.slot_range_as_union(element)
+            needed_classes.extend([a_range for a_range in slot_ranges if a_range in all_classes])
+
+        elif element is None:
+            # get all imports
+            needed_classes.extend([cls for cls in all_classes if cls not in local_classes])
+
+        else:
+            raise ValueError(f"Unsupported type of element to get imports from: f{type(element)}")
+
+        # SPECIAL CASE: classes that are not generated for structural reasons.
+        # TODO: Do we want to have a general means of skipping class generation?
+        skips = ("AnyType",)
+
+        class_imports = [
+            self._get_element_import(cls) for cls in needed_classes if (cls not in local_classes and cls not in skips)
+        ]
+        imports = Imports(imports=class_imports)
+
+        return imports
+
+    def generate_module_import(self, schema: SchemaDefinition, context: Optional[dict] = None) -> str:
+        """
+        Generate the module string for importing from python modules generated from imported schemas
+        when in :attr:`.split` mode.
+
+        Use the :attr:`.split_pattern` as a jinja template rendered with the :class:`.SchemaDefinition`
+        and any passed ``context``. Apply the :attr:`.SNAKE_CASE` regex to substitute matches with
+        ``_`` and ensure lowercase.
+        """
+        if context is None:
+            context = {}
+        module = Template(self.split_pattern).render(schema=schema, **context)
+        module = re.sub(self.SNAKE_CASE, "_", module) if self.SNAKE_CASE else module
+        module = module.lower()
+        return module
+
+    def _get_element_import(self, class_name: ElementName) -> Import:
+        """
+        Make an import object for an element from another schema, using the
+        :attr:`.split_import_pattern` to generate the module import part.
+        """
+        schema_name = self.schemaview.element_by_schema_map()[class_name]
+        schema = [s for s in self.schemaview.schema_map.values() if s.name == schema_name][0]
+        module = self.generate_module_import(schema, self.split_context)
+        return Import(module=module, objects=[ObjectImport(name=camelcase(class_name))], is_schema=True)
+
     def render(self) -> PydanticModule:
         sv: SchemaView
         sv = self.schemaview
-        schema = sv.schema
-        pyschema = SchemaDefinition(
-            id=schema.id,
-            name=schema.name,
-            description=schema.description.replace('"', '\\"') if schema.description else None,
-        )
-        enums = self.generate_enums(sv.all_enums())
-        injected_classes = copy(DEFAULT_INJECTS[self.pydantic_version])
-        if self.injected_classes is not None:
-            injected_classes += self.injected_classes
 
+        # imports
         imports = DEFAULT_IMPORTS
         if self.imports is not None:
             for i in self.imports:
                 imports += i
+        if self.split_mode == SplitMode.FULL:
+            imports += self._get_imports()
 
-        sorted_classes = self.sort_classes(list(sv.all_classes().values()))
-        self.sorted_class_names = [camelcase(c.name) for c in sorted_classes]
+        # injected classes
+        injected_classes = DEFAULT_INJECTS.copy()
+        if self.injected_classes is not None:
+            injected_classes += self.injected_classes.copy()
+
+        # enums
+        enums = self.before_generate_enums(list(sv.all_enums().values()), sv)
+        enums = self.generate_enums({e.name: e for e in enums})
+
+        base_model = PydanticBaseModel(extra_fields=self.extra_fields, fields=self.injected_fields)
 
+        # schema classes
+        class_results = []
+        source_classes, imported_classes = self._get_classes(sv)
+        source_classes = self.sort_classes(source_classes, imported_classes)
         # Don't want to generate classes when class_uri is linkml:Any, will
         # just swap in typing.Any instead down below
-        sorted_classes = [c for c in sorted_classes if c.class_uri != "linkml:Any"]
-
-        for class_original in sorted_classes:
-            class_def: ClassDefinition
-            class_def = deepcopy(class_original)
-            class_name = class_original.name
-            class_def.name = camelcase(class_original.name)
-            if class_def.is_a:
-                class_def.is_a = camelcase(class_def.is_a)
-            class_def.mixins = [camelcase(p) for p in class_def.mixins]
-            if class_def.description:
-                class_def.description = class_def.description.replace('"', '\\"')
-            pyschema.classes[class_def.name] = class_def
-            for attribute in list(class_def.attributes.keys()):
-                del class_def.attributes[attribute]
-            for sn in sv.class_slots(class_name):
-                # TODO: fix runtime, copy should not be necessary
-                s = deepcopy(sv.induced_slot(sn, class_name))
-                # logging.error(f'Induced slot {class_name}.{sn} == {s.name} {s.range}')
-                s.name = underscore(s.name)
-                if s.description:
-                    s.description = s.description.replace('"', '\\"')
-                class_def.attributes[s.name] = s
-
-                slot_ranges: List[str] = []
-
-                # Confirm that the original slot range (ignoring the default that comes in from
-                # induced_slot) isn't in addition to setting any_of
-                any_of_ranges = [a.range if a.range else s.range for a in s.any_of]
-                if any_of_ranges:
-                    # list comprehension here is pulling ranges from within AnonymousSlotExpression
-                    slot_ranges.extend(any_of_ranges)
-                else:
-                    slot_ranges.append(s.range)
+        source_classes = [c for c in source_classes if c.class_uri != "linkml:Any"]
+        source_classes = self.before_generate_classes(source_classes, sv)
+        self.sorted_class_names = [camelcase(c.name) for c in source_classes]
+        for cls in source_classes:
+            cls = self.before_generate_class(cls, sv)
+            result = self.generate_class(cls)
+            result = self.after_generate_class(result, sv)
+            class_results.append(result)
+            if result.imports is not None:
+                imports += result.imports
+            if result.injected_classes is not None:
+                injected_classes.extend(result.injected_classes)
 
-                pyranges = [self.generate_python_range(slot_range, s, class_def) for slot_range in slot_ranges]
+        class_results = self.after_generate_classes(class_results, sv)
 
-                pyranges = list(set(pyranges))  # remove duplicates
-                pyranges.sort()
+        classes = {r.cls.name: r.cls for r in class_results}
 
-                if len(pyranges) == 1:
-                    pyrange = pyranges[0]
-                elif len(pyranges) > 1:
-                    pyrange = f"Union[{', '.join(pyranges)}]"
-                else:
-                    raise Exception(f"Could not generate python range for {class_name}.{s.name}")
-
-                if s.array is not None:
-                    # TODO add support for xarray
-                    results = self.get_array_representations_range(s, pyrange)
-                    # TODO: Move results unpacking to own function that is used after each slot build stage :)
-                    for res in results:
-                        if res.injected_classes:
-                            injected_classes += res.injected_classes
-                        if res.imports:
-                            imports += res.imports
-                    if len(results) == 1:
-                        pyrange = results[0].annotation
-                    else:
-                        pyrange = f"Union[{', '.join([res.annotation for res in results])}]"
-
-                    if "linkml:ColumnOrderedArray" in class_def.implements:
-                        raise NotImplementedError("Cannot generate Pydantic code for ColumnOrderedArrays.")
-                elif s.multivalued:
-                    if s.inlined or s.inlined_as_list:
-                        collection_key = self.generate_collection_key(slot_ranges, s, class_def)
-                    else:
-                        collection_key = None
-                    if s.inlined is False or collection_key is None or s.inlined_as_list is True:
-                        pyrange = f"List[{pyrange}]"
-                    else:
-                        simple_dict_value = None
-                        if len(slot_ranges) == 1:
-                            simple_dict_value = self._inline_as_simple_dict_with_value(s, sv)
-                        if simple_dict_value:
-                            # inlining as simple dict
-                            pyrange = f"Dict[str, {simple_dict_value}]"
-                        else:
-                            pyrange = f"Dict[{collection_key}, {pyrange}]"
-                if not (s.required or s.identifier or s.key) and not s.designates_type:
-                    pyrange = f"Optional[{pyrange}]"
-                ann = Annotation("python_range", pyrange)
-                s.annotations[ann.tag] = ann
-
-        # TODO: Make cleaning injected classes its own method
-        injected_classes = list(
-            dict.fromkeys([c if isinstance(c, str) else inspect.getsource(c) for c in injected_classes])
-        )
-        injected_classes = [textwrap.dedent(c) for c in injected_classes]
-
-        base_model = PydanticBaseModel(
-            pydantic_ver=self.pydantic_version, extra_fields=self.extra_fields, fields=self.injected_fields
-        )
-
-        classes = {}
-        predefined = self.get_predefined_slot_values()
-        bases = self.get_class_isa_plus_mixins()
-        for k, c in pyschema.classes.items():
-            attrs = {}
-            for attr_name, src_attr in c.attributes.items():
-                src_attr = src_attr._as_dict
-                new_fields = {
-                    k: src_attr.get(k, None)
-                    for k in PydanticAttribute.model_fields.keys()
-                    if src_attr.get(k, None) is not None
-                }
-                predef_slot = predefined.get(k, {}).get(attr_name, None)
-                if predef_slot is not None:
-                    predef_slot = str(predef_slot)
-                new_fields["predefined"] = predef_slot
-                new_fields["name"] = attr_name
-
-                attrs[attr_name] = PydanticAttribute(**new_fields, pydantic_ver=self.pydantic_version)
-                attrs[attr_name] = self.include_metadata(attrs[attr_name], src_attr)
-
-            new_class = PydanticClass(
-                name=k, attributes=attrs, description=c.description, pydantic_ver=self.pydantic_version
-            )
-            new_class = self.include_metadata(new_class, c)
-            if k in bases:
-                new_class.bases = bases[k]
-            classes[k] = new_class
+        injected_classes = self._clean_injected_classes(injected_classes)
 
         module = PydanticModule(
-            pydantic_ver=self.pydantic_version,
             metamodel_version=self.schema.metamodel_version,
             version=self.schema.version,
             python_imports=imports.imports,
@@ -768,22 +974,166 @@ class PydanticGenerator(OOCodeGenerator):
             enums=enums,
             classes=classes,
         )
-        module = self.include_metadata(module, schema)
+        module = self.include_metadata(module, self.schemaview.schema)
+        module = self.before_render_template(module, self.schemaview)
         return module
 
-    def serialize(self) -> str:
-        module = self.render()
-        return module.render(self._template_environment(), self.black)
+    def serialize(self, rendered_module: Optional[PydanticModule] = None) -> str:
+        """
+        Serialize the schema to a pydantic module as a string
+
+        Args:
+            rendered_module ( :class:`.PydanticModule` ): Optional, if schema was previously
+                rendered with :meth:`.render` , use that, otherwise :meth:`.render` fresh.
+        """
+        if rendered_module is not None:
+            module = rendered_module
+        else:
+            module = self.render()
+        serialized = module.render(self._template_environment(), self.black)
+        serialized = self.after_render_template(serialized, self.schemaview)
+        return serialized
 
     def default_value_for_type(self, typ: str) -> str:
         return "None"
 
+    @classmethod
+    def generate_split(
+        cls,
+        schema: Union[str, Path, SchemaDefinition],
+        output_path: Union[str, Path] = Path("."),
+        split_pattern: Optional[str] = None,
+        split_context: Optional[dict] = None,
+        split_mode: SplitMode = SplitMode.AUTO,
+        **kwargs,
+    ) -> List[SplitResult]:
+        """
+        Generate a schema that imports from other schema as a set of python modules that
+        import from one another, rather than generating all imported classes in a single schema.
+
+        Uses ``output_path`` for the main schema from ``schema`` , and then
+        generates any imported schema (from which classes are actually used)
+        to modules whose locations are determined by the module names generated
+        by the ``split_pattern`` (see :attr:`.PydanticGenerator.split_pattern` ).
+
+        For example, for
+
+        * a ``output_path`` of ``my_dir/v1_2_3/main.py``
+        * a schema ``main`` with a version ``v1.2.3``
+        * that imports from ``s2`` with version ``v4.5.6``,
+        * and a ``split_pattern`` of ``..{{ schema.version | replace('.', '_') }}.{{ schema.name }}``
+
+        One would get:
+        * ``my_dir/v1_2_3/main.py`` , as expected
+        * that imports ``from ..v4_5_6.s2``
+        * a module at ``my_dir/v4_5_6/s2.py``
+
+        ``__init__.py`` files are generated for any directories that are between
+        the generated modules and their highest common directory.
+
+        Args:
+            schema (str, :class:`.Path` , :class:`.SchemaDefinition` ): Main schema to generate
+            output_path (str, :class:`.Path` ): Python ``.py`` module to generate main schema to
+            split_pattern (str): Pattern to use to generate module names, see :attr:`.PydanticGenerator.split_pattern`
+            split_context (dict): Additional variables to pass into jinja context when generating module import names.
+
+        Returns:
+            list[:class:`.SplitResult`]
+        """
+        output_path = Path(output_path)
+        if not output_path.suffix == ".py":
+            raise ValueError(f"output path must be a python file to write the main schema to, got {output_path}")
+
+        results = []
+
+        # --------------------------------------------------
+        # Main schema
+        # --------------------------------------------------
+        gen_kwargs = kwargs
+        gen_kwargs.update(
+            {"split": True, "split_pattern": split_pattern, "split_context": split_context, "split_mode": split_mode}
+        )
+        generator = cls(schema, **gen_kwargs)
+        # Generate the initial schema to figure out which of the imported schema actually need
+        # to be generated
+        rendered = generator.render()
+        # write schema - we use the ``output_path`` for the main schema, and then
+        # interpret all imported schema paths as relative to that
+        output_path.parent.mkdir(parents=True, exist_ok=True)
+        serialized = generator.serialize(rendered_module=rendered)
+        with open(output_path, "w") as ofile:
+            ofile.write(serialized)
+
+        results.append(
+            SplitResult(main=True, source=generator.schemaview.schema, path=output_path, serialized_module=serialized)
+        )
+
+        # --------------------------------------------------
+        # Imported schemas
+        # --------------------------------------------------
+        imported_schema = {
+            generator.generate_module_import(sch): sch for sch in generator.schemaview.schema_map.values()
+        }
+        for generated_import in [i for i in rendered.python_imports if i.is_schema]:
+            import_generator = cls(imported_schema[generated_import.module], **gen_kwargs)
+            serialized = import_generator.serialize()
+            rel_path = _import_to_path(generated_import.module)
+            abs_path = (output_path.parent / rel_path).resolve()
+            abs_path.parent.mkdir(parents=True, exist_ok=True)
+            with open(abs_path, "w") as ofile:
+                ofile.write(serialized)
+
+            results.append(
+                SplitResult(
+                    main=False,
+                    source=imported_schema[generated_import.module],
+                    path=abs_path,
+                    serialized_module=serialized,
+                    module_import=generated_import.module,
+                )
+            )
+
+        _ensure_inits([r.path for r in results])
+        return results
+
 
 def _subclasses(cls: Type):
     return set(cls.__subclasses__()).union([s for c in cls.__subclasses__() for s in _subclasses(c)])
 
 
-_TEMPLATE_NAMES = sorted(list(set([c.template for c in _subclasses(TemplateModel)])))
+_TEMPLATE_NAMES = sorted(list(set([c.template for c in _subclasses(PydanticTemplateModel)])))
+
+
+def _import_to_path(module: str) -> Path:
+    """Make a (relative) ``Path`` object from a python module import string"""
+    # handle leading .'s separately..
+    _, dots, module = re.split(r"(^\.*)(?=\w)", module, maxsplit=1)
+    # treat zero or one dots as a relative import to the current directory
+    dir_pieces = ["../" for _ in range(max(len(dots) - 1, 0))]
+    dir_pieces.extend(module.split("."))
+    dir_pieces[-1] = dir_pieces[-1] + ".py"
+    return Path(*dir_pieces)
+
+
+def _ensure_inits(paths: List[Path]):
+    """For a set of paths, find the common root and it and all the subdirectories have an __init__.py"""
+    # if there is only one file, there is no relative importing to be done
+    if len(paths) <= 1:
+        return
+    common_path = Path(os.path.commonpath(paths))
+
+    if not (ipath := (common_path / "__init__.py")).exists():
+        with open(ipath, "w") as ifile:
+            ifile.write(" \n")
+
+    for path in paths:
+        # ensure __init__ for each directory from this path up to the common path
+        path = path.parent
+        while path != common_path:
+            if not (ipath := (path / "__init__.py")).exists():
+                with open(ipath, "w") as ifile:
+                    ifile.write(" \n")
+            path = path.parent
 
 
 @shared_arguments(PydanticGenerator)
@@ -795,22 +1145,16 @@ _TEMPLATE_NAMES = sorted(list(set([c.template for c in _subclasses(TemplateModel
 Optional jinja2 template directory to use for class generation.
 
 Pass a directory containing templates with the same name as any of the default 
-:class:`.TemplateModel` templates to override them. The given directory will be 
+:class:`.PydanticTemplateModel` templates to override them. The given directory will be 
 searched for matching templates, and use the default templates as a fallback 
 if an override is not found
-  
+
 Available templates to override:
 
 \b
 """
     + "\n".join(["- " + name for name in _TEMPLATE_NAMES]),
 )
-@click.option(
-    "--pydantic-version",
-    type=click.IntRange(1, 2),
-    default=int(PYDANTIC_VERSION[0]),
-    help="Pydantic version to use (1 or 2)",
-)
 @click.option(
     "--array-representations",
     type=click.Choice([k.value for k in ArrayRepresentation]),
@@ -839,7 +1183,7 @@ Available templates to override:
     "Default (auto) is to include all metadata that can't be otherwise represented",
 )
 @click.version_option(__version__, "-V", "--version")
-@click.command()
+@click.command(name="pydantic")
 def cli(
     yamlfile,
     template_file=None,
@@ -849,7 +1193,6 @@ def cli(
     classvars=True,
     slots=True,
     array_representations=list("list"),
-    pydantic_version=int(PYDANTIC_VERSION[0]),
     extra_fields: Literal["allow", "forbid", "ignore"] = "forbid",
     black: bool = False,
     meta: MetadataMode = "auto",
@@ -870,7 +1213,6 @@ def cli(
 
     gen = PydanticGenerator(
         yamlfile,
-        pydantic_version=pydantic_version,
         array_representations=[ArrayRepresentation(x) for x in array_representations],
         extra_fields=extra_fields,
         emit_metadata=head,