org.jruby.compiler.ir.IRExecutionScope Maven / Gradle / Ivy
package org.jruby.compiler.ir;
import java.util.ArrayList;
import java.util.HashMap;
import java.util.Iterator;
import java.util.List;
import java.util.Map;
import java.util.Set;
import java.util.Stack;
import java.util.SortedSet;
import java.util.TreeSet;
import org.jruby.compiler.ir.compiler_pass.CompilerPass;
import org.jruby.compiler.ir.instructions.CallInstr;
import org.jruby.compiler.ir.instructions.Instr;
import org.jruby.compiler.ir.instructions.ReceiveClosureInstr;
import org.jruby.compiler.ir.instructions.RubyInternalCallInstr;
import org.jruby.compiler.ir.operands.LocalVariable;
import org.jruby.compiler.ir.operands.Operand;
import org.jruby.compiler.ir.operands.MethAddr;
import org.jruby.compiler.ir.operands.Variable;
import org.jruby.compiler.ir.representations.CFG;
import org.jruby.parser.StaticScope;
/* IR_Method and IR_Closure -- basically scopes that represent execution contexts.
* This is just an abstraction over methods and closures */
public abstract class IRExecutionScope extends IRScopeImpl {
private List instructions; // List of IR instructions for this method
private CFG cfg; // Control flow graph for this scope
private List closures; // List of (nested) closures in this scope
/* *****************************************************************************************************
* Does this execution scope (applicable only to methods) receive a block and use it in such a way that
* all of the caller's local variables need to be materialized into a heap binding?
* Ex:
* def foo(&b)
* eval 'puts a', b
* end
*
* def bar
* a = 1
* foo {} # prints out '1'
* end
*
* Here, 'foo' can access all of bar's variables because it captures the caller's closure.
*
* There are 2 scenarios when this can happen (even this is conservative -- but, good enough for now)
* 1. This method receives an explicit block argument (in this case, the block can be stored, passed around,
* eval'ed against, called, etc.).
* CAVEAT: This is conservative ... it may not actually be stored & passed around, evaled, called, ...
* 2. This method has a 'super' call (ZSuper AST node -- RUBY_INTERNALS_CALL_Instr(MethAddr.ZSUPER, ..) IR instr)
* In this case, the parent (in the inheritance hierarchy) can access the block and store it, etc. So, in reality,
* rather than assume that the parent will always do this, we can query the parent, if we can precisely identify
* the parent method (which in the face of Ruby's dynamic hierarchy, we cannot). So, be pessimistic.
*
* This logic was extracted from an email thread on the JRuby mailing list -- Yehuda Katz & Charles Nutter
* contributed this analysis above.
* ********************************************************************************************************/
private boolean canCaptureCallersBinding;
/* ****************************************************************************
* Does this scope define code, i.e. does it (or anybody in the downward call chain)
* do class_eval, module_eval? In the absence of any other information, we default
* to yes -- which basically leads to pessimistic but safe optimizations. But, for
* library and internal methods, this might be false.
* **************************************************************************** */
private boolean canModifyCode;
/* ****************************************************************************
* Does this scope (if a closure, applies to the nearest method ancestor) require a binding to be materialized?
* Yes if any of the following holds true:
* - calls 'Proc.new'
* - calls 'eval'
* - calls 'call' (could be a call on a stored block which could be local!)
* - calls 'send' and we cannot resolve the message (method name) that is being sent!
* - calls methods that can access the caller's binding
* - calls a method which we cannot resolve now!
* - has a call whose closure requires a binding
* **************************************************************************** */
private boolean requiresBinding;
// NOTE: Since we are processing ASTs, loop bodies are processed in depth-first manner
// with outer loops encountered before inner loops, and inner loops finished before outer ones.
//
// So, we can keep track of loops in a loop stack which keeps track of loops as they are encountered.
// This lets us implement next/redo/break/retry easily for the non-closure cases
private Stack loopStack;
protected int requiredArgs = 0;
protected int optionalArgs = 0;
protected int restArg = -1;
private void init() {
instructions = new ArrayList();
closures = new ArrayList();
loopStack = new Stack();
// All flags are true by default!
canModifyCode = true;
canCaptureCallersBinding = true;
requiresBinding = true;
}
public IRExecutionScope(IRScope lexicalParent, Operand container, String name, StaticScope staticScope) {
super(lexicalParent, container, name, staticScope);
init();
}
public void addClosure(IRClosure c) {
closures.add(c);
}
@Override
public void addInstr(Instr i) {
instructions.add(i);
}
public void startLoop(IRLoop l) {
loopStack.push(l);
}
public void endLoop(IRLoop l) {
loopStack.pop(); /* SSS FIXME: Do we need to check if l is same as whatever popped? */
}
public IRLoop getCurrentLoop() {
return loopStack.isEmpty() ? null : loopStack.peek();
}
public List getClosures() {
return closures;
}
// SSS FIXME: Deprecated! Going forward, all instructions should come from the CFG
@Override
public List getInstrs() {
return instructions;
}
public IRMethod getClosestMethodAncestor() {
IRExecutionScope s = this;
while (!(s instanceof IRMethod)) {
s = (IRExecutionScope)s.getLexicalParent();
}
return (IRMethod) s;
}
public void setCodeModificationFlag(boolean f) {
canModifyCode = f;
}
public boolean modifiesCode() {
return canModifyCode;
}
public boolean requiresBinding() {
return requiresBinding;
}
public boolean canCaptureCallersBinding() {
return canCaptureCallersBinding;
}
public CFG buildCFG() {
cfg = new CFG(this);
cfg.build(instructions);
return cfg;
}
// Get the control flow graph for this scope
public CFG getCFG() {
return cfg;
}
// Nothing to do -- every compiler pass decides whether to
// run it on nested closures or not.
@Override
public void runCompilerPassOnNestedScopes(CompilerPass p) { }
public void computeExecutionScopeFlags() {
// init
canModifyCode = true;
canCaptureCallersBinding = false;
requiresBinding = false;
// recompute flags -- we could be calling this method different times
// definitely once after ir generation and local optimizations propagates constants locally
// but potentially at a later time after doing ssa generation and constant propagation
boolean receivesClosureArg = false;
for (Instr i: getInstrs()) {
if (i instanceof ReceiveClosureInstr)
receivesClosureArg = true;
// SSS FIXME: Should we build a ZSUPER IR Instr rather than have this code here?
if ((i instanceof RubyInternalCallInstr) && (((CallInstr) i).getMethodAddr() == MethAddr.ZSUPER))
canCaptureCallersBinding = true;
if (i instanceof CallInstr) {
CallInstr call = (CallInstr) i;
if (call.requiresBinding())
requiresBinding = true;
// If this method receives a closure arg, and this call is an eval that has more than 1 argument,
// it could be using the closure as a binding -- which means it could be using pretty much any
// variable from the caller's binding!
if (receivesClosureArg && call.canBeEval() && (call.getCallArgs().length > 1))
canCaptureCallersBinding = true;
}
}
}
@Override
public String toStringInstrs() {
StringBuilder b = new StringBuilder();
int i = 0;
for (Instr instr : instructions) {
if (i > 0) b.append("\n");
b.append(" ").append(i).append('\t').append(instr);
i++;
}
if (!closures.isEmpty()) {
b.append("\n\n------ Closures encountered in this scope ------\n");
for (IRClosure c: closures)
b.append(c.toStringBody());
b.append("------------------------------------------------\n");
}
return b.toString();
}
@Override
public String toStringVariables() {
Map ends = new HashMap();
Map starts = new HashMap();
SortedSet variables = new TreeSet();
for (int i = instructions.size() - 1; i >= 0; i--) {
Instr instr = instructions.get(i);
Variable var = instr.result;
if (var != null) {
variables.add(var);
starts.put(var, i);
}
for (Operand operand : instr.getOperands()) {
if (operand != null && operand instanceof Variable && ends.get((Variable)operand) == null) {
ends.put((Variable)operand, i);
variables.add((Variable)operand);
}
}
}
StringBuilder sb = new StringBuilder();
int i = 0;
for (Variable var : variables) {
Integer end = ends.get(var);
if (end != null) { // Variable is actually used somewhere and not dead
if (i > 0) sb.append("\n");
i++;
sb.append(" " + var + ": " + starts.get(var) + "-" + end);
}
}
return sb.toString();
}
@Interp
public Iterator getLiveLocalVariables() {
Map ends = new HashMap();
Map starts = new HashMap();
Set variables = new TreeSet();
for (int i = instructions.size() - 1; i >= 0; i--) {
Instr instr = instructions.get(i);
// TODO: Instruction encode whether arguments are optional/required/block
// TODO: PErhaps this should be part of allocate and not have a generic
// getLiveLocalVariables...perhaps we just need this to setup static scope
Variable variable = instr.result;
if (variable != null && variable instanceof LocalVariable) {
variables.add((LocalVariable) variable);
starts.put((LocalVariable) variable, i);
}
for (Operand operand : instr.getOperands()) {
if (!(operand instanceof LocalVariable)) continue;
variable = (LocalVariable) operand;
if (ends.get((LocalVariable) variable) == null) {
ends.put((LocalVariable) variable, i);
variables.add((LocalVariable) variable);
}
}
}
return variables.iterator();
}
/**
* Create and (re)assign a static scope. In general local variables should
* never change even if we optimize more, but I was not positive so I am
* pretending this can change over time. The obvious secondary benefit
* to storing this on execution scope is we can grab it when we allocate
* static scopes for all closures.
*
* Note: We are missing a distinct life-cycle point to run methods like this
* since this method can be modified at any point. Not being fully set up
* is less of an issue since we are calling this when we construct a live
* runtime version of the method/closure etc, but for profiled optimizations
* this is less clear if/when we can run this. <-- Assumes this ever needs
* changing.
*
* @param parent scope should be non-null for all closures and null for methods
*/
@Interp
public StaticScope allocateStaticScope(StaticScope parent) {
Iterator variables = getLiveLocalVariables();
StaticScope scope = constructStaticScope(parent);
while (variables.hasNext()) {
LocalVariable variable = variables.next();
int destination = scope.addVariable(variable.getName());
System.out.println("Allocating " + variable + " to " + destination);
// Ick: Same Variable objects are not used for all references to the same variable. S
// o setting destination on one will not set them on all
variable.setLocation(destination);
}
return scope;
}
@Interp
public void calculateParameterCounts() {
for (int i = instructions.size() - 1; i >= 0; i--) {
Instr instr = instructions.get(i);
}
}
/**
* Closures and Methods have different static scopes. This returns the
* correct instance.
*
* @param parent scope should be non-null for all closures and null for methods
* @return a newly allocated static scope
*/
@Interp
protected abstract StaticScope constructStaticScope(StaticScope parent);
// ENEBO: Can this always be the same variable? Then SELF comparison could compare against this?
public Variable getSelf() {
return getLocalVariable("%self");
}
public LocalVariable getLocalVariable(String name) {
return getClosestMethodAncestor().getLocalVariable(name);
}
public int getLocalVariablesCount() {
return getClosestMethodAncestor().getLocalVariablesCount();
}
}
© 2015 - 2025 Weber Informatics LLC | Privacy Policy