I can't just replace the entire method. I have to just inject a reassignment of a local var after the last time it normally gets set but before it gets used near the end of the method.
Here's some pseudocode
void test() {
/* stuff */
String thing = null;
if (case1) {
thing = "case1"
}
if (case2) {
thing = "case2"
}
if (case3) {
thing = "case3"
}
if (thing == null) {
thing = "default";
}
/* I want to insert this code below with ASM */
thing = "Injected by ASM";
/* stuff */
}
I could also just replace the default assignment when the code reaches that point where it checks if thing == null. But the byte code for the default assignment is a pretty long StringBuilder with lots of appends. There's a LDC that I can use to uniquely identify that but line I dunno how to replace the whole thing assignment for that line. I only know how to replace the LDC (which is not enough).
The idea is I want to ignore all the case1-3 so that thing is always what I tell ASM to set it as
But the /* stuff */ at the top and bottom of the method cannot be removed
While you could visit the instructions in the method to find the last place it is set, and then pass over the code again to inject the change, the simplest approach is to translate the code so that every time the variable is set, you set it to the value you want. This might not even need more code, just replace what is there.
Related
I am using Kotlin in a webserver app and I have a line of code as follows:
.onComplete { jsonResult: AsyncResult<JsonObject>? ->
Now what I want to do is change the underlying JsonObject wrapped in the AsyncResult, so that it is going to be reflected further downstream.
var res: JsonObject? = jsonResult?.result()
if (res != null) {
if (res.getInteger("files_uploaded") > 0) {
res.put("URL", "Some URL")
}
}
I was then imagining to update the underlying JSON object in the result but not sure how to do that.
please take note that single quotes are missing and ` appear as \` because the code formatting. I tried to leave what seemed least confusing...
You should be able to make changes in the conditional statement
if (res !=null) {
res being the JsonObject:
console.log(res);
would show you what's in there. You may need to use
let resXmodifiedX = JSON.parse(res);
One approach is to write a function and pass res to that function which you can do if it is in the console.log(res).
Some notes on what's below:
place the function somewhere consistent maybe at the bottom of the file...
objects often have multiple levels res.person.name, res.contact.email, or whatever...
use multiple for loops:
let level = res[key]; for(child in level) {
you don't need to do this if you know exactly what object attributes you need to update.
you can set the value directly but you always want to test for it before trying to set it to avoid errors that stop execution.
let toBe = toBe =>`${toBe}`;
let update = (res)?toBe(update(res)):toBe('not Found');
This option is really only if you know for sure that data will be there and you can't proceed without it. Which is not uncommon but also not how JSON is designed to be used.
The code below is a concise way to make some simple changes but may not be an ideal solution. To use it xModify(res) replaces console.log(res) above.
function xModify(x) {
let resXmodifiedX = JSON.parse(x);
let res = resXmodifiedX;
for (key in res) {
res[key] = key=='name'? \`My change ${res[key]}\`: key=='other'? \`My Change ${res[key]}\`:res[key];
resXmodifiedX = JSON.stringify(res);
return resXmodifiedX;
}
That will update res.name and res.other otherwise res[key] is unchanged. If you do not need to parse res change let res = xModifiedx; to let res = x; remove the first line and change the last two lines to return res;
function xModify(x) {
let res = x;
for (key in res) {
res[key] = key=='name'? \`My change ${res[key]}\`: key=='other'? \`My Change ${res[key]}\`:res[key];
return res;
}
If your data is numeric which is not generally the case in a web server response scenario this is a terrible approach. Because it is probably a string I used the template variable as a way to easily add a complex pattern in place of a string. My change ${res[key]} not a real world example. Any valid JS code can go in the ${ } (template variable). I've been defaulting to the first pattern more and more.
let me = (bestCase)?`${'the best version'} of myself`:`${'someone'} I'm ok with`;
So I have a Method
public modifiers Foo foo(Bar bar){
blah;
blah;
veryInterestingStmt;
moreBlah();
return XYZ;
}
I now want to split this method s.t. everything in its body is extracted into a separate method (programmatically).
I.e.
public modifiers Foo foo(Bar bar){
return trulyFoo(bar);
}
public modifiers Foo trulyFoo(Bar bar){
blah;
blah;
veryInterestingStmt;
moreBlah();
return XYZ;
}
How do I do that, though?
The naive
private void fracture(SootMethod sm) {
SootClass sc = sm.getDeclaringClass();
String auxMethodName = sm.getName() + FRACTURE_SUFFIX;
Type auxReturnType = sm.getReturnType();
List<Type>auxParamTypes = new LinkedList<>(sm.getParameterTypes());
int auxModifiers = sm.getModifiers();
SootMethod auxMethod = sc.addMethod(new SootMethod(auxMethodName,auxParamTypes,auxReturnType,auxModifiers));
Body body = sm.getActiveBody();
Body auxBody = Jimple.v().newBody(auxMethod);
auxMethod.setActiveBody(auxBody);
for(Local l : body.getLocals()){
auxBody.getLocals().add(l);
}
PatchingChain<Unit> units = body.getUnits();
PatchingChain<Unit> auxUnits = auxBody.getUnits();
Iterator<Unit> it = body.getUnits().snapshotIterator();
boolean passedFirstNonidentity = false;
while(it.hasNext()){
Stmt stmt = (Stmt) it.next();
if(!passedFirstNonidentity && !(stmt instanceof IdentityStmt)) {
passedFirstNonidentity = true;
//TODO: if added more parameters than original method had, add their identity stmts here
}
auxUnits.add(stmt);
// if(passedFirstNonidentity) units.remove(stmt); //TODO: uncomment this and later add call to {#code auxMethod}
}
}
}
Doesn't work. If I run, say
DirectedGraph dg = new ExceptionalUnitGraph(auxMethod.getActiveBody());
I get a
java.lang.RuntimeException: Unit graph contains jump to non-existing target
at soot.toolkits.graph.UnitGraph.buildUnexceptionalEdges(UnitGraph.java:128)
at soot.toolkits.graph.ExceptionalUnitGraph.initialize(ExceptionalUnitGraph.java:258)
at soot.toolkits.graph.ExceptionalUnitGraph.<init>(ExceptionalUnitGraph.java:159)
at soot.toolkits.graph.ExceptionalUnitGraph.<init>(ExceptionalUnitGraph.java:192)
The technique of moving code without altering the behavior of the code is called Refactoring and is nicely covered in a book by Martin Fowler.
In your case, I would take the following multi-step approach:
Stand up a "do nothing" function in the function you wish to split, just above the lines of code you wish to move.
Move one or two of those lines of code from the surrounding function int the "do nothing" function, splitting the function, but having the split be a nested call.
Move the split function up (or down) to the edge of the block in the surronding function.
Move teh slpit function out of the block, placing new calls to it either prior to every call of the original function, or after every call of the original function. Note that you may have to rework the handling of return parameters, depending on the details.
It is strongly suggested that you write a set of tests to validate some, if not most, of the overall functionality of this block first. Then, after each change run your tests to verify that you didn't change behavior.
What you are seeing now is a change in behavior which came about by modifying the text of the code in such a manner that it did change behavior. The set of safe transformations of source code is likely smaller than you previously believed, or maybe you just made a simple error. However, the work you are attempting requires more knowledge than can be expressed in a StackOverflow style, question / answer, format. That's why I made the book reference.
If you can narrow the scope, you might get a better response in a future resubmission.
It seems that moving stmts just doesn't work. In contrast, completely replacing the body
Body originalBody = sm.getActiveBody();
originalBody.setMethod(auxMethod);
auxMethod.setActiveBody(originalBody);
Body newBody = Jimple.v().newBody(sm);
sm.setActiveBody(newBody);
and then regenerating the locals, identity stmts (and other stmts you may need) in the newBody looks like a sensible way to go.
I am implementing a ray tracer and am having trouble with basic Java references :/ I have been staring at this for a while and I can't see the problem...
IntersectResult ir = new IntersectResult();
root.intersect(ray, ir);
if(r.material!=null)
System.out.println(result.material.diffuse);
// Doesn't print at all!!
// in my Node Class...
#Override
public void intersect(Ray ray, IntersectResult result) {
IntersectResult i = new IntersectResult();
for (Intersectable child:children){
child.intersect(ray, i);
if (result.t>i.t)
result = new IntersectResult(i);
}
if(result.material!=null)
System.out.println(result.material.diffuse); // prints correctly!
}
Basically my question is why is result.material null after the intersect method call when the print statements within the method call show that it is not?
This is a classic case of how the "pass-by-value" approach works and it is not specific to the code in question.
With respect to the latter, the ir reference passed to the intersect() method is redirected by the statement
result = new IntersectResult(i);
So, whenever the above statement is executed, the material object created is not saved in the original material variable, but to a local one, which is lost after the intersect method returns.
If you want to propagate that change, make the intersect() method always return the result object at its end and change the statement at the start of the code to
ir = root.intersect(ray, ir);
and correct the typo in the following if statement (it is ir.material, not r.material).
This is how I understand method getUser below :
Return a User object or null
Get a Set of users and assign them to userSer.
If the set is not empty begin iterating over the set but
return the first user within the set.
Here is the method :
private User getUser(UserDet arg)
{
Set<User> userSet = arg.getUsers(User.class);
if (CollectionUtils.isNotEmpty(userSet))
{
for (User user : userSet)
{
return user;
}
}
return null;
}
I think I could replace the method with this :
private User getUser(UserDet arg)
{
Set<User> userSet = arg.getUsers(User.class);
if (CollectionUtils.isNotEmpty(userSet))
{
return userSet.iterator().next();
}
else {
return null;
}
}
This new method removes the loop and just returns the first element in the set, same as original implemention. Is it correct?
Yes. Actually, it's pretty much almost the same thing, as a foreach loop is syntactic sugar for using an iterator from an Iterable.
Note, however, that you don't need the nonempty check in the first variant, since the loop won't iterate in the case of an empty set anyway.
yes both are same. in first implementation, control will return on first iteration of the loop from the function and consequently loop will end.
Yes it is correct, I'd even go for removing the CollectionUtils.isNotEmptySet and use the Iterator's hasNext method... If the set is guaranteed to be non-null.
It seems to be correct, but it will only make the method a bit easier to read, it will not optimize it in terms of performance. Still I think the change is good and you should do it.
Yes, it does pretty much the same, but if your spec says to start iterating then maybe you should - maybe this method will be extended in the future.
BTW: it is a good convention that your method has only one return statement (i.e. you can create a variable, which will be returned, assigned a null at the beginning and assign a user inside your loop)
Yes. Both the methods return the first element in the set. The first method seems to have been written for something else previously and changed then keeping the for loop intact.
In anycase, the second method that you're proposing won't give any significant performance benefit but should be a better way than the first one.
So in case, UserDet#getUsers(Class) never returns null (but an empty Set in case no user could be found), the shortest (and in my opinion most readable) form is:
private User getUser(UserDet arg) {
Set<User> userSet = arg.getUsers(User.class);
return userSet.isEmpty() ? null : userSet.iterator().next();
}
I would do this.
I won't run a loop and more over I'l add a null check.
private User getUser(UserDet arg) {
Set<User> userSet = arg.getUsers(User.class);
if (userSet != null && userSet.size() > 0) {
return userSet.iterator().next();
}
return null;
}
I'm sure there must be a standard way to do this, but my attempts to search Stackoverflow have failed.
I have a method like:
public void processSomeWidgetsForUser(int userItemId) {
Iterator<Widgets> iter = allWidgets.values().iterator();
while(iter.hasNext()) {
Widget thisWidget = iter.next();
if (userItemId == -1 || thisWidget.getUsersItemId() == userItemId) {
widget.process();
}
}
}
As you can see -1 is a "special value" meaning process all. Doing this saves repeating the loop code in another method called processSomeWidgetsForAllUsers.
But I dislike special values like this because they are easy to misuse or misunderstand, which is exactly the situation what I'm having to fix now (where someone thought -1 meant something else).
I can only think of two ways to improve this.
have a constant, containing -1 called something like
Widget.ALLWIDGETS which at least is self-documenting, but doesn't
stop code from using a -1 (if someone integrates old code in, for
example)
change the method to take a list of all user ids to
process, which can be empty, but that doesn't seem great
performance-wise (would need to retrieve all user ids first and then loop through
removing. Also what happens if the number of widgets in the list changes between
retreiving the ids and removing
Is there a better way? I'm sure I'm missing something obvious.
The above code has been changed slightly, so may not compile, but you should get the gist.
Although somewhat redundant, a fairly neat self-documenting approach could be to have 3 methods rather than one;
Make your original method private, and make one small change which would be to add your static final int EXECUTE_ALL = -1 and use that in your original method, then add the two new methods;
public void processWidget(int wID) throws IllegalArgumentException {
if(wID == EXECUTE_ALL) throw new IllegalArgumentException();
originalMethod(wID);
}
public void processAllWidgets() {
originalMethod(EXECUTE_ALL);
}
It makes your class a little more cluttered, but as far as the exposed methods go, it is clearer and hopefully foolproof. You could alter it not to throw an exception and just ignore any invalid ids, that just depends on your situation.
This approach of course has the major downside that it changes how the class appears to other classes, breaking everything that currently uses the, now private, originalMethod().
Number 1 would work very nicely. Be sure to document what the variable is though, so future coders (possibly yourself) know what it means.
/**This is the explanation for the below variable*/
public final static int ALL_WIDGETS = -1;
Have an external method like so:
static boolean idRepresentsAll(int id) {
return id == -1;
}
In this case, if you decide to replace it with a different mechanism, you only replace your magic number one place in your code.
At the very least, you would want to do something like this:
public static final int ID_REPRESENTING_ALL = -1;
You can change the method signature to accept a boolean for when you want to process them all.
public void processSomeWidgets(boolean doAll, int userItemId) {
Iterator<Widgets> iter = allWidgets.values().iterator();
while(iter.hasNext()) {
Widget thisWidget = iter.next();
if (doAll || thisWidget.getUsersItemId() == userItemId) {
widget.process();
}
}
}
This makes it more explicit, and easier to read in my opinion as there are no special values.