explicit_object_extensions provisional -- SIMICS-23313

Author: lwaern-intel

py/dead_dml_methods.py

@@ -76,7 +76,7 @@ def traverse_ast(ast):
         for stmt in stmts:
             yield from traverse_ast(stmt)
     elif ast.kind == 'object':
-        (_, _, _, stmts) = ast.args
+        (_, _, _, _, stmts) = ast.args
         for stmt in stmts:
             yield from traverse_ast(stmt)
     elif ast.kind == 'method':

py/dml/dmlparse.py

@@ -297,8 +297,8 @@ def toplevel_else_if(t):
 
 @prod_dml12
 def object_anonymous_bank(t):
-    'object : BANK object_spec'
-    t[0] = ast.object_(site(t), None, 'bank', [], t[2])
+    'object : maybe_extension BANK object_spec'
+    t[0] = ast.object_(site(t), None, 'bank', [], None, t[3])
 
 @prod
 def array_list_empty(t):
@@ -312,9 +312,9 @@ def array_list(t):
 
 @prod
 def object_regarray(t):
-    'object : REGISTER objident array_list sizespec offsetspec maybe_istemplate object_spec'
-    t[0] = ast.object_(site(t), t[2], 'register', t[3],
-                       t[4] + t[5] + t[6] + t[7])
+    'object : maybe_extension REGISTER objident array_list sizespec offsetspec maybe_istemplate object_spec'
+    t[0] = ast.object_(site(t), t[3], 'register', t[4], t[1],
+                       t[5] + t[6] + t[7] + t[8])
 
 @prod
 def bitrangespec(t):
@@ -328,8 +328,8 @@ def bitrangespec_empty(t):
 
 @prod_dml14
 def object_field(t):
-    'object : FIELD objident array_list bitrangespec maybe_istemplate object_spec'
-    t[0] = ast.object_(site(t), t[2], 'field', t[3], t[4] + t[5] + t[6])
+    'object : maybe_extension FIELD objident array_list bitrangespec maybe_istemplate object_spec'
+    t[0] = ast.object_(site(t), t[3], 'field', t[4], t[1], t[5] + t[6] + t[7])
 
 def endian_translate_bit(expr, width, bitorder):
     if bitorder == 'be':
@@ -368,10 +368,10 @@ def bitrange_2(t):
 
 @prod_dml12
 def object_field_1(t):
-    'object : FIELD objident bitrange maybe_istemplate object_spec'
+    'object : maybe_extension FIELD objident bitrange maybe_istemplate object_spec'
     if logging.show_porting:
-        report(PFIELDRANGE(site(t, 3)))
-    t[0] = ast.object_(site(t), t[2], 'field', [], t[3] + t[4] + t[5])
+        report(PFIELDRANGE(site(t, 4)))
+    t[0] = ast.object_(site(t), t[3], 'field', [], None, t[4] + t[5] + t[6])
 
 @prod_dml12
 def field_array_size_no(t):
@@ -397,8 +397,8 @@ def field_array_size(t):
 
 @prod_dml12
 def object_field_2(t):
-    'object : FIELD objident fieldarraysize bitrangespec maybe_istemplate object_spec'
-    t[0] = ast.object_(site(t), t[2], 'field', t[3], t[4] + t[5] + t[6])
+    'object : maybe_extension FIELD objident fieldarraysize bitrangespec maybe_istemplate object_spec'
+    t[0] = ast.object_(site(t), t[3], 'field', t[4], None, t[5] + t[6] + t[7])
 
 @prod_dml14
 def data(t):
@@ -472,20 +472,20 @@ def saved_decl_many_init(t):
 
 @prod
 def object3(t):
-    '''object : CONNECT   objident array_list maybe_istemplate object_spec
-              | INTERFACE objident array_list maybe_istemplate object_spec
-              | ATTRIBUTE objident array_list maybe_istemplate object_spec
-              | BANK      objident array_list maybe_istemplate object_spec
-              | EVENT     objident array_list maybe_istemplate object_spec
-              | GROUP     objident array_list maybe_istemplate object_spec
-              | PORT      objident array_list maybe_istemplate object_spec
-              | IMPLEMENT objident array_list maybe_istemplate object_spec'''
-    t[0] = ast.object_(site(t), t[2], t[1], t[3], t[4] + t[5])
+    '''object : maybe_extension CONNECT   objident array_list maybe_istemplate object_spec
+              | maybe_extension INTERFACE objident array_list maybe_istemplate object_spec
+              | maybe_extension ATTRIBUTE objident array_list maybe_istemplate object_spec
+              | maybe_extension BANK      objident array_list maybe_istemplate object_spec
+              | maybe_extension EVENT     objident array_list maybe_istemplate object_spec
+              | maybe_extension GROUP     objident array_list maybe_istemplate object_spec
+              | maybe_extension PORT      objident array_list maybe_istemplate object_spec
+              | maybe_extension IMPLEMENT objident array_list maybe_istemplate object_spec'''
+    t[0] = ast.object_(site(t), t[3], t[2], t[4], t[1], t[5] + t[6])
 
 @prod_dml14
 def object_subdevice(t):
-    '''object : SUBDEVICE objident array_list maybe_istemplate object_spec'''
-    t[0] = ast.object_(site(t), t[2], t[1], t[3], t[4] + t[5])
+    '''object : maybe_extension SUBDEVICE objident array_list maybe_istemplate object_spec'''
+    t[0] = ast.object_(site(t), t[3], t[2], t[4], t[1], t[5] + t[6])
 
 @prod_dml12
 def maybe_extern_yes(t):
@@ -884,6 +884,31 @@ def object_desc_none(t):
     'object_desc :'
     t[0] = []
 
+@prod_dml14
+def maybe_extension_yes(t):
+    'maybe_extension : IN'
+    if not site(t).provisional_enabled(
+            provisional.explicit_object_extensions):
+        report(ESYNTAX(site(t), 'in', None))
+        t[0] = None
+    else:
+        t[0] = True
+
+@prod_dml14
+def maybe_extension_no(t):
+    'maybe_extension : '
+    enabled = (provisional.explicit_object_extensions
+               in t.parser.file_info.provisional)
+    t[0] = False if enabled else None
+
+# Note that `maybe_extension` is used even in DML 1.2 exclusive rules, which
+# may seem redundant, but no! Removing those uses would lead to shift/reduce
+# conflicts.
+@prod_dml12
+def maybe_extension(t):
+    'maybe_extension : '
+    t[0] = None
+
 @prod
 def object_spec_none(t):
     'object_spec : object_desc SEMI'

py/dml/messages.py

@@ -1169,6 +1169,41 @@ def log(self):
         DMLError.log(self)
         self.print_site_message(self.other_site, "existing declaration")
 
+
+class EEXTENSION(DMLError):
+    """When the `explict_object_extensions` provisional feature is enabled,
+    any object definition made via `in` syntax is considered an extension such
+    that there must be some other non-extension declaration of the object, or
+    DMLC will reject the extension.
+    To declare and define a new object not already declared, omit the `in`
+    syntax.
+    """
+    fmt = ("object '%s' not declared elsewhere."
+           " To declare and define a new object, omit 'in'.")
+
+class EMULTIOBJDECL(DMLError):
+    """When the `explicit_object_extensions` provisional feature is enabled,
+    any object declaration not made using `in` syntax is considered a
+    declaration of a novel object — because of that, DMLC will reject
+    it if there already is another non-`in` declaration across files enabling
+    `explicit_object_extensions`.
+    """
+    fmt = ("object '%s' already declared."
+           " To extend upon the definition of an object, use 'in %s'")
+    def __init__(self, site, other_site, objtype, name):
+        super().__init__(site, name, f'{objtype} {name} ...')
+        self.other_site = other_site
+
+    def log(self):
+        from . import provisional
+        DMLError.log(self)
+        self.print_site_message(self.other_site, "existing declaration")
+        prov_site = self.site.provisional_enabled(
+            provisional.explicit_object_extensions)
+        self.print_site_message(
+            prov_site,
+            "enabled by the explicit_object_extensions provisional feature")
+
 class EVARPARAM(DMLError):
     """
     The value assigned to the parameter is not a well-defined constant.

py/dml/provisional.py

@@ -135,6 +135,34 @@ class simics_util_vect(ProvisionalFeature):
     stable = True
     dml12 = True
 
+
+@feature
+class explicit_object_extensions(ProvisionalFeature):
+    '''<a id="explicit_object_extensions"/>
+    This feature extends the DML syntax for object declarations to distinguish
+    between an intent to introduce a new object to the model structure, and an
+    intent to extend the definition of an existing object.
+
+    The following form is introduced to mark the intent to extend an object:
+    <pre>
+    in <em>object-type</em> <em>name</em> <em>...</em> { <em>...</em> }
+    </pre>
+    E.g.
+    <pre>
+    in bank some_bank { ... }
+    </pre>
+
+    If this form is used while there is no other non-extension declaration of
+    the named object, then DMLC will signal an error because the definition
+    was not intended to introduce the object to the model structure.
+
+    DMLC will also signal an error if there is more than one non-extension
+    declaration of the object among the files enabling
+    `explicit_object_extensions`.
+    '''
+    short = "Require `in` syntax for additional declarations of an object"
+    stable = False
+
 def parse_provisional(
         provs: list[("Site", str)]) -> dict[ProvisionalFeature, "Site"]:
     ret = {}

py/dml/structure.py

@@ -503,9 +503,9 @@ def wrap_sites(spec, issite, tname):
             else:
                 raise ICE(issite, 'unknown node type %r %r' % (asttype, stmt))
         composite_wrapped = [
-            (objtype, name, arrayinfo,
+            (objtype, name, arrayinfo, is_extension,
              ObjectSpec(*wrap_sites(spec, issite, tname)))
-            for (objtype, name, arrayinfo, spec) in composite]
+            for (objtype, name, arrayinfo, is_extension, spec) in composite]
         blocks.append((preconds, shallow_wrapped, composite_wrapped))
 
     return (TemplateSite(spec.site, issite, tname), spec.rank,
@@ -833,8 +833,8 @@ def sort_method_implementations(implementations, obj_specs):
     return traits.sort_method_implementations(implementations)
 
 def merge_subobj_defs(def1, def2, parent):
-    (objtype, name, arrayinfo, obj_specs1) = def1
-    (objtype2, name2, arrayinfo2, obj_specs2) = def2
+    (objtype, name, arrayinfo, extension_status1, obj_specs1) = def1
+    (objtype2, name2, arrayinfo2, extension_status2, obj_specs2) = def2
     assert name == name2
 
     site1 = obj_specs1[0].site
@@ -844,6 +844,26 @@ def merge_subobj_defs(def1, def2, parent):
         report(ENAMECOLL(site1, site2, name))
         return def1
 
+    # extension status:
+    # None -> dual extension and decl
+    # True -> extension
+    # False -> explicit decl
+    if extension_status1 is extension_status2 is False:
+        # Blame the one with higher rank if possible. Otherwise, blame the
+        # new def.
+        (decl_site_1, decl_site_2) = (
+            (site1, site2) if (
+                obj_specs2[0].rank in obj_specs1[0].rank.inferior)
+            else (site2, site1))
+        report(EMULTIOBJDECL(decl_site_1, decl_site_2, objtype, name))
+    # One is explicit decl: merged is explicit decl
+    elif extension_status1 is False or extension_status2 is False:
+        extension_status1 = False
+    # If either is dual, make dual
+    elif extension_status1 is None or extension_status2 is None:
+        extension_status1 = None
+    # Otherwise, both are explicit extensions. Keep explicit extension.
+
     if len(arrayinfo) != len(arrayinfo2):
         raise EAINCOMP(site1, site2, name,
                        "mixing declarations with different number "
@@ -870,7 +890,8 @@ def merge_subobj_defs(def1, def2, parent):
                 merged_arrayinfo.append((idxvar1, len1))
 
 
-    return (objtype, name, merged_arrayinfo, obj_specs1 + obj_specs2)
+    return (objtype, name, merged_arrayinfo, extension_status1,
+            obj_specs1 + obj_specs2)
 
 def method_is_std(node, methname):
     """
@@ -1708,13 +1729,14 @@ def mkobj2(obj, obj_specs, params, each_stmts):
 
     for (stmts, obj_spec) in composite_subobjs:
         for s in stmts:
-            (objtype, ident, arrayinfo, subobj_spec) = s
+            (objtype, ident, arrayinfo, is_extension, subobj_spec) = s
 
             if ident is None:
                 assert (dml.globals.dml_version == (1, 2)
                         and objtype in {'bank', 'field'})
 
-            subobj_def = (objtype, ident, arrayinfo, [subobj_spec])
+            subobj_def = (objtype, ident, arrayinfo, is_extension,
+                          [subobj_spec])
             if ident in subobj_defs:
                 subobj_defs[ident] = merge_subobj_defs(subobj_defs[ident],
                                                        subobj_def, obj)
@@ -1724,7 +1746,10 @@ def mkobj2(obj, obj_specs, params, each_stmts):
                 symbols[ident] = subobj_spec.site
                 subobj_defs[ident] = subobj_def
 
-    for (_, _, arrayinfo, specs) in subobj_defs.values():
+    for (_, ident, arrayinfo, extension_status, specs) in subobj_defs.values():
+        if extension_status is True:
+            for spec in specs:
+                report(EEXTENSION(spec.site, ident))
         for (i, (idx, dimsize_ast)) in enumerate(arrayinfo):
             if dimsize_ast is None:
                 idxref = (f" (with index variable '{idx.args[0]}')"
@@ -1814,7 +1839,7 @@ def mkobj2(obj, obj_specs, params, each_stmts):
         # the whole register.
         if obj.objtype == 'register' and not any(
                 objtype == 'field'
-                for (objtype, _, _, _) in list(subobj_defs.values())):
+                for (objtype, _, _, _, _) in list(subobj_defs.values())):
             # The implicit field instantiates the built-in field
             # template and does nothing else.
             subobjs.append(mkobj(
@@ -1856,7 +1881,7 @@ def mkobj2(obj, obj_specs, params, each_stmts):
     subobj_name_defs = {}
 
     for name in sorted(subobj_defs, key=lambda name: name or ''):
-        (objtype, ident, arrayinfo, subobj_specs) = subobj_defs[name]
+        (objtype, ident, arrayinfo, _, subobj_specs) = subobj_defs[name]
         if (not obj.accepts_child_type(objtype)
             # HACK: disallow non-toplevel banks in DML 1.2, see SIMICS-19009
             or (dml.globals.dml_version == (1, 2)

py/dml/template.py

@@ -123,7 +123,7 @@ def defined_symbols(self):
                     symbols[sub.args[0]] = (sub.kind, sub.site)
                 else:
                     assert sub.kind == 'error'
-            for (_, name, _, specs) in composite:
+            for (_, name, _, _, specs) in composite:
                 symbols[name] = ('subobj', specs.site)
         return symbols
 
@@ -299,11 +299,12 @@ def obj_from_asts(site, stmts):
             block = []
             for decl_ast in composite:
                 assert decl_ast.kind == 'object'
-                (name, objtype, indices, sub_stmts) = decl_ast.args
+                (name, objtype, indices, is_extension,
+                 sub_stmts) = decl_ast.args
                 spec = obj_from_asts(
                     decl_ast.site, sub_stmts + [ast.is_(
                         decl_ast.site, [(decl_ast.site, objtype)])])
-                block.append((objtype, name, indices, spec))
+                block.append((objtype, name, indices, is_extension, spec))
             blocks.append((preconds, shallow, block))
         return ObjectSpec(site, rank, is_stmts, in_eachs, params, blocks)
     return obj_from_asts(site, stmts)
@@ -331,7 +332,7 @@ def rank_structure(asts):
     while queue:
         (spec, conditional) = queue.pop()
         if spec.kind == 'object':
-            (_, objtype, _, stmts) = spec.args
+            (_, objtype, _, _, stmts) = spec.args
             inferior[objtype] = spec
             queue.extend((stmt, conditional) for stmt in stmts)
         elif spec.kind == 'is':