src.org.python.indexer.Scope Maven / Gradle / Ivy
Show all versions of jython-standalone Show documentation
/**
* Copyright 2009, Google Inc. All rights reserved.
* Licensed to PSF under a Contributor Agreement.
*/
package org.python.indexer;
import org.python.indexer.ast.NName;
import org.python.indexer.ast.NNode;
import org.python.indexer.ast.NUrl;
import org.python.indexer.types.NType;
import org.python.indexer.types.NUnionType;
import org.python.indexer.types.NUnknownType;
import java.util.ArrayList;
import java.util.Collection;
import java.util.Collections;
import java.util.HashSet;
import java.util.LinkedHashMap;
import java.util.List;
import java.util.Map.Entry;
import java.util.Map;
import java.util.Set;
/**
* Symbol table.
*/
public class Scope {
/**
* For preventing circular inheritance from recursing.
*/
private static Set looked = new HashSet();
public enum Type {
CLASS,
INSTANCE,
FUNCTION,
MODULE,
GLOBAL,
SCOPE
}
/**
* XXX: This table is incorrectly overloaded to contain both object
* attributes and lexical-ish scope names, when they are in some cases
* separate namespaces. (In particular, they're effectively the same
* namespace for module scope and class scope, and they're different for
* function scope, which uses the {@code func_dict} namespace for storing
* attributes.)
*/
private Map table; // stays null for most scopes (mem opt)
private Scope parent;
private List supers;
private Set globalNames;
private Type scopeType;
private String path = "";
private int lambdaCounter = 0;
private boolean isBindingPhase = false;
public Scope(Scope parent, Type type) {
if (type == null) {
throw new IllegalArgumentException("'type' param cannot be null");
}
setParent(parent);
setScopeType(type);
}
public void setTable(Map table) {
this.table = table;
}
/**
* Returns an immutable view of the table.
*/
public Map getTable() {
if (table != null) {
return Collections.unmodifiableMap(table);
}
Map map = Collections.emptyMap();
return map;
}
public void setParent(Scope parent) {
this.parent = parent;
}
public Scope getParent() {
return parent;
}
public void addSuper(Scope sup) {
if (supers == null) {
supers = new ArrayList();
}
supers.add(sup);
}
public void setSupers(List supers) {
this.supers = supers;
}
public List getSupers() {
if (supers != null) {
return Collections.unmodifiableList(supers);
}
List list = Collections.emptyList();
return list;
}
public void setScopeType(Type type) {
this.scopeType = type;
}
public Type getScopeType() {
return scopeType;
}
public boolean isFunctionScope() {
return scopeType == Type.FUNCTION;
}
/**
* Mark a name as being global (i.e. module scoped) for name-binding and
* name-lookup operations in this code block and any nested scopes.
*/
public void addGlobalName(String name) {
if (name == null) {
return;
}
if (globalNames == null) {
globalNames = new HashSet();
}
globalNames.add(name);
}
/**
* Returns {@code true} if {@code name} appears in a {@code global}
* statement in this scope or any enclosing scope.
*/
public boolean isGlobalName(String name) {
if (globalNames != null) {
return globalNames.contains(name);
}
return parent == null ? false : parent.isGlobalName(name);
}
/**
* Directly assigns a binding to a name in this table. Does not add a new
* definition or reference to the binding. This form of {@code put} is
* often followed by a call to {@link Indexer#putLocation} to create a reference to
* the binding. When there is no code location associated with {@code id},
* or it is otherwise undesirable to create a reference, the
* {@link Indexer#putLocation} call is omitted.
*/
public void put(String id, NBinding b) {
putBinding(id, b);
}
/**
* Adds a definition and/or reference to the table.
* If there is no binding for {@code id}, creates one and gives it
* {@code type} and {@code kind}.
*
* If a binding already exists, then add either a definition or a reference
* at {@code loc} to the binding. By convention we consider it a definition
* if the type changes. If the passed type is different from the binding's
* current type, set the binding's type to the union of the old and new
* types, and add a definition. If the new type is the same, just add a
* reference.
*
* If the binding already exists, {@code kind} is only updated if a
* definition was added and the binding's type was previously the
* unknown type.
*/
public NBinding put(String id, NNode loc, NType type, NBinding.Kind kind) {
if (type == null) {
throw new IllegalArgumentException("Null type: id=" + id + ", loc=" + loc);
}
NBinding b = lookupScope(id);
return insertOrUpdate(b, id, loc, type, kind);
}
/**
* Same as {@link #put}, but adds the name as an attribute of this scope.
* Looks up the superclass chain to see if the attribute exists, rather
* than looking in the lexical scope chain.
*
* @return the new binding, or {@code null} if the current scope does
* not have a properly initialized path.
*/
public NBinding putAttr(String id, NNode loc, NType type, NBinding.Kind kind) {
if (type == null) {
throw new IllegalArgumentException("Null type: id=" + id + ", loc=" + loc);
}
// Attributes are always part of a qualified name. If there is no qname
// on the target type, it's a bug (we forgot to set the path somewhere.)
if ("".equals(path)) {
Indexer.idx.reportFailedAssertion(
"Attempting to set attr '" + id + "' at location " + loc
+ (loc != null ? loc.getFile() : "")
+ " in scope with no path (qname) set: " + this.toShortString());
return null;
}
NBinding b = lookupAttr(id);
return insertOrUpdate(b, id, loc, type, kind);
}
private NBinding insertOrUpdate(NBinding b, String id, NNode loc, NType t, NBinding.Kind k) {
if (b == null) {
b = insertBinding(new NBinding(id, loc, t, k));
} else {
updateType(b, loc, t, k);
}
return b;
}
/**
* Adds a new binding for {@code id}. If a binding already existed,
* replaces its previous definitions, if any, with {@code loc}. Sets the
* binding's type to {@code type} (not a union with the previous type).
*/
public NBinding update(String id, NNode loc, NType type, NBinding.Kind kind) {
if (type == null) {
throw new IllegalArgumentException("Null type: id=" + id + ", loc=" + loc);
}
return update(id, new Def(loc), type, kind);
}
/**
* Adds a new binding for {@code id}. If a binding already existed,
* replaces its previous definitions, if any, with {@code loc}. Sets the
* binding's type to {@code type} (not a union with the previous type).
*/
public NBinding update(String id, Def loc, NType type, NBinding.Kind kind) {
if (type == null) {
throw new IllegalArgumentException("Null type: id=" + id + ", loc=" + loc);
}
NBinding b = lookupScope(id);
if (b == null) {
return insertBinding(new NBinding(id, loc, type, kind));
}
b.getDefs().clear(); // XXX: what about updating refs & idx.locations?
b.addDef(loc);
b.setType(type);
// XXX: is this a bug? I think he meant to do this check before the
// line above that sets b.type, if it's supposed to be like put().
if (b.getType().isUnknownType()) {
b.setKind(kind);
}
return b;
}
private NBinding insertBinding(NBinding b) {
switch (b.getKind()) {
case MODULE:
b.setQname(b.getType().getTable().path);
break;
case PARAMETER:
b.setQname(extendPathForParam(b.getName()));
break;
default:
b.setQname(extendPath(b.getName()));
break;
}
b = Indexer.idx.putBinding(b);
putBinding(b.getName(), b);
return b;
}
private void putBinding(String id, NBinding b) {
ensureTable();
table.put(id, b);
}
private void updateType(NBinding b, NNode loc, NType type, NBinding.Kind kind) {
NType curType = b.followType();
if (!isNewType(curType, type)) {
if (loc != null
&& !(loc instanceof NUrl)
&& !b.getDefs().contains(loc)) {
Indexer.idx.putLocation(loc, b);
}
return;
}
if (loc != null && !b.getRefs().contains(loc)) {
b.addDef(loc);
b.setProvisional(false);
}
// The union ordering matters here. If they're two different unknown
// types, union() points the first one to the second one. We want to
// keep the binding's existing type iff its table contains provisional
// attribute bindings that we need to look up later.
NType btype = b.getType();
NType t1, t2;
if (btype.isUnknownType() && !btype.getTable().isEmpty()) {
t1 = type;
t2 = btype;
} else {
t1 = btype;
t2 = type;
}
NType newType = NUnionType.union(t1, t2);
b.setType(newType);
if (curType.isUnknownType()) {
b.setKind(kind);
}
retargetReferences(b, curType);
}
/**
* If the current type had a provisional binding, retarget its refs to the
* new type. It probably only works one level deep: need dataflow analysis
* in the general case. However, it does pick up some extra references,
* so it's reasonable for now.
*/
private void retargetReferences(NBinding b, NType curType) {
Scope newScope = b.followType().getTable();
for (Map.Entry e : curType.getTable().entrySet()) {
String attr = e.getKey();
NBinding oldBinding = e.getValue();
if (!oldBinding.isProvisional()) {
continue;
}
Indexer.idx.removeBinding(oldBinding);
NBinding newBinding = newScope.lookupAttr(attr);
if (newBinding != null) {
List refs = new ArrayList(); // avoid ConcurrentModificationException
refs.addAll(oldBinding.getRefs());
for (Ref ref : refs) {
Indexer.idx.updateLocation(ref, newBinding);
}
}
}
}
/**
* Returns {@code true} if the binding is being assigned a new type.
*/
private boolean isNewType(NType curType, NType type) {
// In the bindNames() phase we want all places where a given name
// is bound in the same scope to share the same binding, because
// we haven't resolved the types yet. This takes care of that case.
if (isBindingPhase) {
return false;
}
if (curType.isUnionType()) {
return !curType.asUnionType().contains(type);
}
return curType != type;
}
public void remove(String id) {
if (table != null) {
table.remove(id);
}
}
/**
* Create a copy of the symbol table but without the links to parent, supers
* and children. Useful for creating instances.
*
* @return the symbol table for use by the instance.
*/
public Scope copy(Type tableType) {
Scope ret = new Scope(null, tableType);
if (table != null) {
ret.ensureTable();
ret.table.putAll(table);
}
return ret;
}
public void setPath(String path) {
if (path == null) {
throw new IllegalArgumentException("'path' param cannot be null");
}
this.path = path;
}
public String getPath() {
return path;
}
public void setPath(String a, String b) {
NBinding b1 = lookup(a);
NBinding b2 = lookup(b);
if (b1 != null && b2 != null) {
b1.setQname(b2.getQname());
}
}
/**
* Look up a name (String) in the current symbol table. If not found,
* recurse on the parent table.
*/
public NBinding lookup(String name) {
NBinding b = getModuleBindingIfGlobal(name);
if (b != null) {
return b;
}
if (table != null) {
NBinding ent = table.get(name);
if (ent != null) {
return ent;
}
}
if (getParent() == null) {
return null;
}
return getParent().lookup(name);
}
/**
* Specialized version for the convenience of looking up {@code Name}s.
* For all other types return {@code null}.
*/
public NBinding lookup(NNode n) {
if (n instanceof NName) {
return lookup(((NName)n).id);
}
return null;
}
/**
* Look up a name, but only in the current scope.
* @return the local binding for {@code name}, or {@code null}.
*/
public NBinding lookupLocal(String name) {
NBinding b = getModuleBindingIfGlobal(name);
if (b != null) {
return b;
}
return table == null ? null : table.get(name);
}
/**
* Look up an attribute in the type hierarchy. Don't look at parent link,
* because the enclosing scope may not be a super class. The search is
* "depth first, left to right" as in Python's (old) multiple inheritance
* rule. The new MRO can be implemented, but will probably not introduce
* much difference.
* @param supersOnly search only in the supers' scopes, not in local table.
*/
public NBinding lookupAttr(String name, boolean supersOnly) {
if (looked.contains(this)) {
return null;
}
if (table != null && !supersOnly) {
NBinding b = table.get(name);
if (b != null) {
return b;
}
}
if (supers == null || supers.isEmpty()) {
return null;
}
looked.add(this);
try {
for (Scope p : supers) {
NBinding b = p.lookupAttr(name);
if (b != null) {
return b;
}
}
return null;
} finally {
looked.remove(this);
}
}
/**
* Look up an attribute in the local scope and superclass scopes.
* @see #lookupAttr(String,boolean)
*/
public NBinding lookupAttr(String name) {
return lookupAttr(name, false);
}
/**
* Look up the scope chain for a binding named {@code name}
* and if found, return its type.
*/
public NType lookupType(String name) {
return lookupType(name, false);
}
/**
* Look for a binding named {@code name} and if found, return its type.
* @param localOnly {@code true} to look only in the current scope;
* if {@code false}, follows the scope chain.
*/
public NType lookupType(String name, boolean localOnly) {
NBinding b = localOnly ? lookupLocal(name) : lookup(name);
if (b == null) {
return null;
}
NType ret = b.followType();
// XXX: really need to make ModuleTable polymorphic...
if (this == Indexer.idx.moduleTable) {
if (ret.isModuleType()) {
return ret;
}
if (ret.isUnionType()) {
for (NType t : ret.asUnionType().getTypes()) {
NType realType = t.follow();
if (realType.isModuleType()) {
return realType;
}
}
}
Indexer.idx.warn("Found non-module type in module table: " + b);
return null;
}
return ret;
}
public NType lookupTypeAttr(String name) {
NBinding b = lookupAttr(name);
if (b != null) {
return b.followType();
}
return null;
}
/**
* Look up a name, but the search is bounded by a type and will not proceed
* to an outer scope when reaching a certain type of symbol table.
*
* @param name the name to be looked up
* @param typebound the type we wish the search to be bounded at
* @return a binding, or {@code null} if not found
*/
public NBinding lookupBounded(String name, Type typebound) {
if (scopeType == typebound) {
return table == null ? null : table.get(name);
}
if (getParent() == null) {
return null;
}
return getParent().lookupBounded(name, typebound);
}
/**
* Returns {@code true} if this is a scope in which names may be bound.
*/
public boolean isScope() {
switch (scopeType) {
case CLASS:
case INSTANCE:
case FUNCTION:
case MODULE:
case GLOBAL:
return true;
default:
return false;
}
}
/**
* Find the enclosing scope-defining symbol table.
*
* More precisely, if a form introduces a new name in the "current scope",
* resolving the form needs to search up the symbol-table chain until it
* finds the table representing the scope to which the name should be added.
* Used by {@link org.python.indexer.ast.NameBinder} to create new name
* bindings in the appropriate enclosing table with the appropriate binding
* type.
*/
public Scope getScopeSymtab() {
if (this.isScope()) {
return this;
}
if (getParent() == null) {
Indexer.idx.reportFailedAssertion("No binding scope found for " + this.toShortString());
return this;
}
return getParent().getScopeSymtab();
}
/**
* Look up a name, but bounded by a scope defining construct. Those scopes
* are of type module, class, instance or function. This is used in
* determining the locations of a variable's definition.
*/
public NBinding lookupScope(String name) {
NBinding b = getModuleBindingIfGlobal(name);
if (b != null) {
return b;
}
Scope st = getScopeSymtab();
if (st != null) {
return st.lookupLocal(name);
}
return null;
}
/**
* Find a symbol table of a certain type in the enclosing scopes.
*/
public Scope getSymtabOfType(Type type) {
if (scopeType == type) {
return this;
}
if (parent == null) {
return null;
}
return parent.getSymtabOfType(type);
}
/**
* Returns the global scope (i.e. the module scope for the current module).
*/
public Scope getGlobalTable() {
Scope result = getSymtabOfType(Type.MODULE);
if (result == null) {
Indexer.idx.reportFailedAssertion("No module table found for " + this);
result = this;
}
return result;
}
/**
* Returns the containing lexical scope (which may be this scope)
* for lexical name lookups. In particular, it skips class scopes.
*/
public Scope getEnclosingLexicalScope() {
if (scopeType == Scope.Type.FUNCTION
|| scopeType == Scope.Type.MODULE) {
return this;
}
if (parent == null) {
Indexer.idx.reportFailedAssertion("No lexical scope found for " + this);
return this;
}
return parent.getEnclosingLexicalScope();
}
/**
* If {@code name} is declared as a global, return the module binding.
*/
private NBinding getModuleBindingIfGlobal(String name) {
if (isGlobalName(name)) {
Scope module = getGlobalTable();
if (module != null && module != this) {
return module.lookupLocal(name);
}
}
return null;
}
/**
* Name binding occurs in a separate pass before the name resolution pass,
* building out the scope tree and binding names in the correct scopes.
* In this pass, the name binding and lookup rules are slightly different.
* This condition is transient: no scopes will be in the name-binding phase
* in a completed index (or module).
*/
public boolean isNameBindingPhase() {
return isBindingPhase;
}
public void setNameBindingPhase(boolean isBindingPhase) {
this.isBindingPhase = isBindingPhase;
}
/**
* Merge all records from another symbol table. Used by {@code import from *}.
*/
public void merge(Scope other) {
ensureTable();
this.table.putAll(other.table);
}
public Set keySet() {
if (table != null) {
return table.keySet();
}
Set result = Collections.emptySet();
return result;
}
public Collection values() {
if (table != null) {
return table.values();
}
Collection result = Collections.emptySet();
return result;
}
public Set> entrySet() {
if (table != null) {
return table.entrySet();
}
Set> result = Collections.emptySet();
return result;
}
public boolean isEmpty() {
return table == null ? true : table.isEmpty();
}
/**
* Dismantles all resources allocated by this scope.
*/
public void clear() {
if (table != null) {
table.clear();
table = null;
}
parent = null;
if (supers != null) {
supers.clear();
supers = null;
}
if (globalNames != null) {
globalNames.clear();
globalNames = null;
}
}
public String newLambdaName() {
return "lambda%" + (++lambdaCounter);
}
/**
* Generates a qname for a parameter of a function or method.
* There is not enough context for {@link #extendPath} to differentiate
* params from locals, so callers must use this method when the name is
* known to be a parameter name.
*/
public String extendPathForParam(String name) {
if (path.equals("")) {
throw new IllegalStateException("Not inside a function");
}
return path + "@" + name;
}
/**
* Constructs a qualified name by appending {@code name} to this scope's qname.
*
* The indexer uses globally unique fully qualified names to address
* identifier definition sites. Many Python identifiers are already
* globally addressable using dot-separated package, class and attribute
* names.
*
* Function variables and parameters are not globally addressable in the
* language, so the indexer uses a special path syntax for creating globally
* unique qualified names for them. By convention the syntax is "@" for
* parameters and "&" for local variables.
*
* @param name a name to append to the current qname
* @return the qname for {@code name}. Does not change this scope's path.
*/
public String extendPath(String name) {
if (name.endsWith(".py")) {
name = Util.moduleNameFor(name);
}
if (path.equals("")) {
return name;
}
String sep = null;
switch (scopeType) {
case MODULE:
case CLASS:
case INSTANCE:
case SCOPE:
sep = ".";
break;
case FUNCTION:
sep = "&";
break;
default:
System.err.println("unsupported context for extendPath: " + scopeType);
return path;
}
return path + sep + name;
}
private void ensureTable() {
if (table == null) {
table = new LinkedHashMap();
}
}
@Override
public String toString() {
return "";
}
public String toShortString() {
return "";
}
}