Pre-process parsed docstrings to add hints on how to resolve names

pydoctor/astbuilder.py

@@ -176,6 +176,7 @@ def visit_Module(self, node: ast.Module) -> None:
 
     def depart_Module(self, node: ast.Module) -> None:
         self.builder.pop(self.module)
+        epydoc2stan.transform_parsed_names(self.module)
 
     def visit_ClassDef(self, node: ast.ClassDef) -> None:
         # Ignore classes within functions.
@@ -1072,6 +1073,8 @@ def _annotation_for_elements(sequence: Iterable[object]) -> Optional[ast.expr]:
 class ASTBuilder:
     """
     Keeps tracks of the state of the AST build, creates documentable and adds objects to the system.
+
+    One ASTBuilder instance is only suitable to build one Module.
     """
     ModuleVistor = ModuleVistor
 

pydoctor/epydoc2stan.py

@@ -4,6 +4,8 @@
 
 from collections import defaultdict
 import enum
+import inspect
+from itertools import chain
 from typing import (
     TYPE_CHECKING, Any, Callable, ClassVar, DefaultDict, Dict, Generator,
     Iterator, List, Mapping, Optional, Sequence, Tuple, Union,
@@ -12,8 +14,11 @@
 import re
 
 import attr
+from docutils.transforms import Transform
+from docutils import nodes
 
 from pydoctor import model, linker, node2stan
+from pydoctor.node2stan import parse_reference
 from pydoctor.astutils import is_none_literal
 from pydoctor.epydoc.markup import Field as EpydocField, ParseError, get_parser_by_name, processtypes
 from twisted.web.template import Tag, tags
@@ -884,7 +889,10 @@
     # Only Attribute instances have the 'annotation' attribute.
     annotation: Optional[ast.expr] = getattr(obj, 'annotation', None)
     if annotation is not None:
-        return colorize_inline_pyval(annotation)
+        parsed_type = colorize_inline_pyval(annotation)
+        # cache parsed_type attribute
+        obj.parsed_type = parsed_type
+        return parsed_type
 
     return None
 
@@ -1143,3 +1151,80 @@
             extra_epytext += '`%s <%s>`' % (short_text, c.fullName())
 
         cls.extra_info.append(parse_docstring(cls, extra_epytext, cls, 'restructuredtext', section='constructor extra'))
+
+class _ReferenceTransform(Transform):
+
+    def __init__(self, document:nodes.document, ctx:'model.Documentable'):
+        super().__init__(document)
+        self.ctx = ctx
+
+    def apply(self):
+        ctx = self.ctx
+        module = self.ctx.module
+        for node in self.document.findall(nodes.title_reference):
+            _, target = parse_reference(node)
+            if target == node.attributes.get('refuri', target):
+                name, *rest = target.split('.')
+                # Only apply transformation to non-ambigous names, 
+                # because we don't know if we're dealing with an annotation
+                # or an interpreted, so we must go with the conservative approach.
+                if ((module.isNameDefined(name) and
+                    not ctx.isNameDefined(name, only_locals=True))
+                    or (ctx.isNameDefined(name, only_locals=True) and 
+                    not module.isNameDefined(name))):
+
+                    node.attributes['refuri'] = '.'.join(chain(
+                        ctx._localNameToFullName(name).split('.'), rest))
+
+def _apply_reference_transform(doc:ParsedDocstring, ctx:'model.Documentable') -> None:
+    """
+    Runs L{_ReferenceTransform} on the underlying docutils document. 
+    No-op if L{to_node} raises L{NotImplementedError}.
+    """
+    try:
+        document = doc.to_node()
+    except NotImplementedError:
+        return
+    else:
+        _ReferenceTransform(document, ctx).apply()
+
+def transform_parsed_names(node:'model.Module') -> None:
+    """
+    Walk this module's content and apply in-place transformations to the 
+    L{ParsedDocstring} instances that olds L{obj_reference} or L{title_reference} nodes. 
+
+    Fixing "Lookup of name in annotation fails on reparented object #295".
+    The fix is not 100% complete at the moment: attribute values and decorators
+    are not handled.
+    """
+    from pydoctor import model, astbuilder
+    # resolve names early when possible
+    for ob in model.walk(node):
+        # resolve names in parsed_docstring, do not forget field bodies
+        if ob.parsed_docstring:
+            _apply_reference_transform(ob.parsed_docstring, ob)
+            for f in ob.parsed_docstring.fields:
+                _apply_reference_transform(f.body(), ob)
+        if isinstance(ob, model.Function):
+            if ob.signature:
+                for p in ob.signature.parameters.values():
+                    ann = p.annotation if p.annotation is not inspect.Parameter.empty else None                    
+                    if isinstance(ann, astbuilder._ValueFormatter):
+                        _apply_reference_transform(ann._colorized, ob)
+                    default = p.default if p.default is not inspect.Parameter.empty else None
+                    if isinstance(default, astbuilder._ValueFormatter):
+                        _apply_reference_transform(default._colorized, ob)
+            # TODO: resolve function's annotations, they are currently presented twice
+            # we can only change signature, annotations in param table must be handled by
+            # introducing attribute parsed_annotations
+        elif isinstance(ob, model.Attribute):
+            # resolve attribute annotation with parsed_type attribute
+            parsed_type = get_parsed_type(ob)
+            if parsed_type:
+                _apply_reference_transform(parsed_type, ob)
+            # TODO: resolve parsed_value
+            # TODO: resolve parsed_decorators
+        elif isinstance(ob, model.Class):
+            # TODO: resolve parsed_class_signature
+            # TODO: resolve parsed_decorators
+            pass

pydoctor/linker.py

@@ -133,9 +133,13 @@
     def link_to(self, identifier: str, label: "Flattenable") -> Tag:
         fullID = self.obj.expandName(identifier)
 
-        target = self.obj.system.objForFullName(fullID)
-        if target is not None:
-            return taglink(target, self.page_url, label)
+        try:
+            target = self.obj.system.find_object(fullID)
+        except LookupError:
+            pass
+        else:
+            if target is not None:
+                return taglink(target, self.page_url, label)
 
         url = self.look_for_intersphinx(fullID)
         if url is not None:
@@ -184,8 +188,18 @@
         if target is not None:
             return target
 
-        # Check if the fullID exists in an intersphinx inventory.
         fullID = self.obj.expandName(identifier)
+
+        # Try fetching the name with it's outdated fullname
+        try:
+            target = self.obj.system.find_object(fullID)
+        except LookupError:
+            pass
+        else:
+            if target is not None:
+                return target
+
+        # Check if the fullID exists in an intersphinx inventory.
         target_url = self.look_for_intersphinx(fullID)
         if not target_url:
             # FIXME: https://github.com/twisted/pydoctor/issues/125

pydoctor/model.py

@@ -111,6 +111,19 @@ class DocumentableKind(Enum):
     PROPERTY            = 150
     VARIABLE            = 100
 
+def walk(node:'Documentable') -> Iterator['Documentable']:
+    """
+    Recursively yield all descendant nodes in the tree starting at *node*
+    (including *node* itself), in no specified order.  This is useful if you
+    only want to modify nodes in place and don't care about the context.
+    """
+    from collections import deque
+    todo = deque([node])
+    while todo:
+        node = todo.popleft()
+        todo.extend(node.contents.values())
+        yield node
+
 class Documentable:
     """An object that can be documented.
 
@@ -271,7 +284,7 @@ def _handle_reparenting_post(self) -> None:
     def _localNameToFullName(self, name: str) -> str:
         raise NotImplementedError(self._localNameToFullName)
 
-    def isNameDefined(self, name:str) -> bool:
+    def isNameDefined(self, name:str, only_locals:bool=False) -> bool:
         """
         Is the given name defined in the globals/locals of self-context?
         Only the first name of a dotted name is checked.
@@ -411,13 +424,13 @@ def setup(self) -> None:
         super().setup()
         self._localNameToFullName_map: Dict[str, str] = {}
 
-    def isNameDefined(self, name: str) -> bool:
+    def isNameDefined(self, name: str, only_locals:bool=False) -> bool:
         name = name.split('.')[0]
         if name in self.contents:
             return True
         if name in self._localNameToFullName_map:
             return True
-        if not isinstance(self, Module):
+        if not isinstance(self, Module) and not only_locals:
             return self.module.isNameDefined(name)
         else:
             return False
@@ -811,8 +824,8 @@ def docsources(self) -> Iterator[Documentable]:
     def _localNameToFullName(self, name: str) -> str:
         return self.parent._localNameToFullName(name)
 
-    def isNameDefined(self, name: str) -> bool:
-        return self.parent.isNameDefined(name)
+    def isNameDefined(self, name: str, only_locals:bool=False) -> bool:
+        return self.parent.isNameDefined(name, only_locals=only_locals)
 
 class Function(Inheritable):
     kind = DocumentableKind.FUNCTION

pydoctor/node2stan.py

@@ -3,7 +3,7 @@
 """
 import re
 import optparse
-from typing import Any, Callable, ClassVar, Iterable, List, Optional, Union, TYPE_CHECKING
+from typing import Any, Callable, ClassVar, Iterable, List, Optional, Sequence, Tuple, Union, TYPE_CHECKING
 from docutils.writers import html4css1
 from docutils import nodes, frontend, __version_info__ as docutils_version_info
 
@@ -52,6 +52,25 @@ def gettext(node: Union[nodes.Node, List[nodes.Node]]) -> List[str]:
             filtered.extend(gettext(child))
     return filtered
 
+def parse_reference(node:nodes.title_reference) -> Tuple[Union[str, Sequence[nodes.Node]], str]:
+    """
+    Split a reference into (label, target).
+    """
+    label: "Flattenable"
+    if 'refuri' in node.attributes:
+        # Epytext parsed or manually constructed nodes.
+        label, target = node.children, node.attributes['refuri']
+    else:
+        # RST parsed.
+        m = _TARGET_RE.match(node.astext())
+        if m:
+            label, target = m.groups()
+        else:
+            label = target = node.astext()
+    # Support linking to functions and methods with () at the end
+    if target.endswith('()'):
+        target = target[:len(target)-2]
+    return label, target
 
 _TARGET_RE = re.compile(r'^(.*?)\s*<(?:URI:|URL:)?([^<>]+)>$')
 _VALID_IDENTIFIER_RE = re.compile('[^0-9a-zA-Z_]')
@@ -105,22 +124,12 @@ def visit_obj_reference(self, node: nodes.Node) -> None:
         self._handle_reference(node, link_func=self._linker.link_to)
 
     def _handle_reference(self, node: nodes.Node, link_func: Callable[[str, "Flattenable"], "Flattenable"]) -> None:
+        node_label, target = parse_reference(node)
         label: "Flattenable"
-        if 'refuri' in node.attributes:
-            # Epytext parsed or manually constructed nodes.
-            label, target = node2stan(node.children, self._linker), node.attributes['refuri']
+        if not isinstance(node_label, str):
+            label = node2stan(node_label, self._linker)
         else:
-            # RST parsed.
-            m = _TARGET_RE.match(node.astext())
-            if m:
-                label, target = m.groups()
-            else:
-                label = target = node.astext()
-
-        # Support linking to functions and methods with () at the end
-        if target.endswith('()'):
-            target = target[:len(target)-2]
-
+            label = node_label
         self.body.append(flatten(link_func(target, label)))
         raise nodes.SkipNode()
 

pydoctor/test/test_epydoc2stan.py

@@ -2106,3 +2106,81 @@ def func():
     assert docstring2html(mod.contents['func'], docformat='plaintext') == expected
     captured = capsys.readouterr().out
     assert captured == ''
+
+def test_parsed_names_partially_resolved_early() -> None:
+    """
+    Test for issue #295
+
+    Annotations are first locally resolved when we reach the end of the module, 
+    then again when we actually resolve the name when generating the stan for the annotation.
+    """
+    typing = '''\
+    List = ClassVar = TypeVar = object()
+    '''
+
+    base = '''\
+    import ast
+    class Vis(ast.NodeVisitor):
+        ...
+    '''
+    src = '''\
+    from typing import List
+    import typing as t
+
+    from .base import Vis
+
+    class Cls(Vis, t.Generic['_T']):
+        """
+        L{Cls}
+        """
+        clsvar:List[str]
+        clsvar2:t.ClassVar[List[str]]
+
+        def __init__(self, a:'_T'):
+            self._a:'_T' = a
+
+    C = Cls
+    _T = t.TypeVar('_T')
+    unknow: i|None|list
+    ann:Cls
+    '''
+
+    top = '''\
+    # the order matters here
+    from .src import C, Cls, Vis
+    __all__ = ['Cls', 'C', 'Vis']
+    '''
+
+    system = model.System()
+    builder = system.systemBuilder(system)
+    builder.addModuleString(top, 'top', is_package=True)
+    builder.addModuleString(base, 'base', 'top')
+    builder.addModuleString(src, 'src', 'top')
+    builder.addModuleString(typing, 'typing')
+    builder.buildModules()
+
+    Cls = system.allobjects['top.Cls']
+    clsvar = Cls.contents['clsvar']
+    clsvar2 = Cls.contents['clsvar2']
+    a = Cls.contents['_a']
+    assert clsvar.expandName('typing.List')=='typing.List'
+    assert '<obj_reference refuri="typing.List">' in clsvar.parsed_type.to_node().pformat()
+    assert 'href="typing.html#List"' in flatten(clsvar.parsed_type.to_stan(clsvar.docstring_linker))
+    assert 'href="typing.html#ClassVar"' in flatten(clsvar2.parsed_type.to_stan(clsvar2.docstring_linker))
+    assert 'href="top.src.html#_T"' in flatten(a.parsed_type.to_stan(clsvar.docstring_linker))
+
+    # the reparenting/alias issue
+    ann = system.allobjects['top.src.ann']
+    assert 'href="top.Cls.html"' in  flatten(ann.parsed_type.to_stan(ann.docstring_linker))
+    assert 'href="top.Cls.html"' in flatten(Cls.parsed_docstring.to_stan(Cls.docstring_linker))
+
+    unknow = system.allobjects['top.src.unknow']
+    assert flatten_text(unknow.parsed_type.to_stan(unknow.docstring_linker)) == 'i|None|list'
+
+
+
+    # TODO: test the __init__ signature and the class bases
+
+    # TODO: Fix two new twisted warnings:
+    # twisted/internet/_sslverify.py:330: Cannot find link target for "twisted.internet.ssl.DN", resolved from "twisted.internet._sslverify.DistinguishedName"
+    # twisted/internet/_sslverify.py:347: Cannot find link target for "twisted.internet.ssl.DN", resolved from "twisted.internet._sslverify.DistinguishedName"