Recently I saw following piece of code on GitHub:
private static String safeToString(Object obj) {
if (obj == null) return null;
try {
return obj.toString();
} catch (Throwable t) {
return "Error occured";
}
}
I've never placed toString() method invocations inside the try-catch blocks. But now when I think about it, it might make sense. For example someone could overwrite toString() method in it's class that might throw a runtime exception, like NullPointerException. So we can try to catch Exception. But why Throwable? Do you think it makes any sense?
There is almost never a good reason to do this. The contract of toString() does not say it’s permissible to throw an exception from that method. Any code which throws an exception is broken code, and such an exception needs to be exposed and fixed, not suppressed.
In the case where you are converting some “bad” object from a library which is out of your control to a String, it might be appropriate to write catch (RuntimeExcepton e), but such a catch should be accompanied by comments which describe in detail why it is necessary, because under normal circumstances, it is not needed.
Rogue exception-throwing toString methods aside, note that Java already has at least two “safe” ways to convert a possibly null value to a String:
Objects.toString(obj, null)
String.valueOf(obj)
…so I would question whether the safeToString method should exist at all.
There are rare cases where you might want to catch an Error like this. In general it's a bad idea however, in this case it might make sense as this is generally for logging/debugging purposes and not used directly by the application.
I would prefer something more informative such as
private static String safeToString(Object obj) {
if (obj == null) return null;
try {
return obj.toString();
} catch (Throwable t) {
return obj.getClass() + ".toString() threw " + t;
}
}
e.g.
class Element {
Object data;
Element e;
public String toString() {
return data + (e == null ? "" : e.toString());
}
}
Element e = new Element();
e.data = "hi";
e.e = e; // oops
System.out.println("e: " + safeToString(e)); // doesn't kill the thread or JVM.
Throwable is the parent class of Exception and Error.
It is normally a bad idea to try and catch Error, as it is designed to not be caught.
Catching Throwable is just the overachieved and counterproductive version of catching Exception. Nonetheless, if for some reason you created another kind of Throwable you want to catch along with an Exception, that could be a way to do that in a single try/catch block. Not that it would be a clean way to do so, but it would work.
EDIT for the TL;DR : in most cases, catch Exception instead of Throwable.
It is incorrect to catch any Throwable and then continue execution since it includes Error, which is meant to be fatal:
From the Javadocs:
An Error is a subclass of Throwable that indicates serious problems that a reasonable application should not try to catch. Most such errors are abnormal conditions. The ThreadDeath error, though a "normal" condition, is also a subclass of Error because most applications should not try to catch it.
That is, some Errors can be recovered (e.g. LinkageError), but others not so much.
But catching Exception might be a valid use-case for example in logging code where you don't want the execution to break simply because a call to toString() fails:
private static String safeToString(Object obj) {
try {
return obj == null ? "null" : obj.toString();
} catch (Exception e) {
return "<exception: " + e + ">";
}
}
End of try block, or after all blocks?
I read a lot of answers on stack overflow about what will happen in each case. However, I could not find out which one is the commonly accepted "corrected" place to return.
Object o = null;
try {
Connection c = getConnection();
o = c.getThing();
return o;
} catch (Exception e) {
//handle exception
return null;
} finally {
c.close();
}
vs
Object o = null;
try {
Connection c = getConnection();
o = c.getThing();
} catch (Exception e) {
//handle exception
return null;
} finally {
c.close();
}
return o;
It completely depends on what you are trying to do. See the following code for some examples of the possibilities (not all of which should be combined in any single piece of code!).
try {
// do something that may fail
return 0; // return a normal value
} catch(SomeException e) {
// maybe log an error
return -1 // maybe return a default or error value
} finally {
// maybe clean up resources
// finally will be executed even if you return in try or catch
// a return here will trump a return in try/catch. This is generally regarded as a bad idea
// see https://stackoverflow.com/questions/65035/does-finally-always-execute-in-java
}
return 1 // return a normal value here instead of in the try/catch.
// May be clearer than multiple return statements
// Also useful if return value does not depend on the try/catch outcome
Update for your updated question. It's partly a matter of preference, though there are significant reasons for each choice. Many people prefer a single point of return, so would do this:
Object o = null;
try {
Connection c = getConnection();
o = c.getThing();
} catch (Exception e) {
//handle exception; leave o as null
} finally {
c.close();
}
return o;
Although this sounds like it should be clearer, of course one still has to examine the code to determine what value(s) o may end up with, so the advantage isn't great.
However, the other options have the advantage that the returning of null is very explicit. This is good if you are one of the many programmers who consider returning null values to be generally a bad idea (see Tony Hoare's "billion dollar mistake").
Instead of returning null one can use "Optional", "Option" or "Maybe" types (available in e.g. Java 8, Scala, Haskell and the Functional Java library), or use the Null Object pattern.
First, I know the Title is a bit ambiguous. Actually I don't know how to write it better!
Second, I will describe the problem.
Situation:
I am practicing on an online judge (OJ), so in case of wrong output, the OJ shows me the test case which makes my code to fail. Normally, I can copy my code and paste it into Eclipse, then call my function with that test case and debug my code.
But the problem when the test case is a multiple calls for my function (Like testing an enhanced version of a priority queue), let's assume there were n calls till the fail of the code. So to debug the code, I will have to call the function say n times! Which is not logical!
Function the OJ will call:
public void enqueue (int value)
{
// implementation
}
public int dequeue ()
{
// implementation
}
Test case makes the code to fail:
Last executed input: enqueue(5, 3), enqueue(51, 60), enqueue(0, 14), enqueue(364, 16),... dequeue(),...
Action:
I need a way to call the function with an array of test cases in order to be able to debug the code.
OR
By some way call the function with its parameter directly from the string. Something like invoke("enqueue(5, 3)");
After some investigation, I found a way to do the Action I need by using Java Reflection Also a helpful thread What is reflection and why is it useful?
I managed to develop a tool, here you are the steps:
Copy the test case in a String
Split the string on each function call
Remove the function name from each call, and store them in array of String in order.
Split the parameter
Now I have 2 arrays of integers for param1 and param2, and an array of String for invokations
I used Reflection in order to be able to call the methods from a string of calls, inside a for loop.
Then we have this
public class Test
{
public static void main(String[] args)
{
String testCase = "enqueue(5, 3), enqueue(51, 60), enqueue(0, 14), enqueue(364, 16), dequeue()";
// Prepare parameters and invocations
int[] param1; // assuming it is ready
int[] param2; // assuming it is ready
String[] calls; // assuming it is ready
try
{
Class calledClass = Class.forName("OJ.Prob3");
Method calledMethod1 = calledClass.getDeclaredMethod("enqueue", String.class, int.class);
Method calledMethod2 = calledClass.getDeclaredMethod("dequeue", null);
for (int i = 0 ; i < calls.length ; i++)
{
if (calls[i].equalsIgnoreCase("enqueue"))
calledMethod1.invoke(calledClass.newInstance(), param[i], param2[i]);
else if (calls[i].equalsIgnoreCase("dequeue"))
calledMethod2.invoke(calledClass.newInstance())
}
} catch (ClassNotFoundException e)
{
e.printStackTrace();
} catch (NoSuchMethodException e)
{
e.printStackTrace();
} catch (SecurityException e)
{
e.printStackTrace();
} catch (IllegalAccessException e)
{
e.printStackTrace();
} catch (IllegalArgumentException e)
{
e.printStackTrace();
} catch (InvocationTargetException e)
{
e.printStackTrace();
}
}
}
I already tested this solution and it works really very smooth, but please if anyone has a better solution, you will be more than welcome.
I will finalize the code and make it something like a tool, and I will post it soon, in order to make everybody's life easier debugging the online judges test cases.
Update:
You can do the same for the static methods, just remove .newInstance() from calledMethod1.invoke(calledClass.newInstance(), param[i], param2[i]); to something like calledMethod1.invoke(calledClass, param[i], param2[i]);
I try to remove all PMD warnings in my project and have some difficulties. I have a method that returns a Container object. This was my original method (simplified):
try {
return setResult();
} catch (Exception ex) {
}
return null;
PMD warned that there are 2 return statements. So I tried this:
Container result = null;
try {
result = setResult();
} catch (Exception ex) {
}
return result;
=> PMD warned about initializing with null. So I tried this:
Container result;
try {
result = setResult();
} catch (Exception ex) {
}
return result;
=> Eclipse did not compile this, suggesting to use the "= null" variation.
What am I missing here? How do I have to write this method without PMD warnings?
I'd use the second approach you show:
Container result = null;
try {
result = setResult();
} catch (Exception ex) {
}
return result;
Because, as PMD points out, it is important to avoid unnecessary return statements for the purpose of reducing complexity.
The PMD documentation classifies the NullAssignment rule as "controversial". It's usefulness very much depends on the context. Here is the sample code they provide:
public void bar() {
Object x = null; // this is OK
x = new Object();
// big, complex piece of code here
x = null; // this is not required
// big, complex piece of code here
}
So, unless you can assign the Container variable with a meaningful value beforehand - you should probably ignore this rule for this code fragment.
If you want to completely avoid the null assignment, you could use Guava's Optional. In this case it has the following advantages:
It forces you to think about the absence of a value
It makes the code more understandable
It makes PMD happy
I'm very new to java and the idea of try catch blocks to handle exceptions.
This roughly what I'm ending up with, and there simply has to be a better way:
try {
JSONObject jsonObject = new JSONObject(jsonString);
int aCount = jsonObject.getInt("acount");
String devTok = jsonObject.getString("dt");
String qURL = jsonObject.getString("qu");
try {
DatastoreService datastore = DatastoreServiceFactory.getDatastoreService();
Key qKey = KeyFactory.createKey("qu", qURL);
int dsACount = (Integer) datastore.get(qKey).getProperty(kLastKnownANumber);
//..etc.. more try catch blocks needed
} catch (EntityNotFoundException e) {
e.printStackTrace();
}
} catch (com.google.appengine.repackaged.org.json.JSONException e) {
e.printStackTrace();
}
There are more try catch blocks embedded in the same way, so that at the end, there is just a lump of catch blocks. How else should exceptions be handled, Eclipse keeps asking my to either surround with a try catch block or "Add throws declaration".
Sometimes I want to catch certain exceptions, for example if it can't find an entity, I want to print something like "Entity not found", and if the JSON string can't be parsed into an object, I want to print something like "Can't parse JSON".
(I'm used to objective-c where the are delegate methods for failure, or the method returns null and you have passed a pointer to [a pointer to] an NSError object, which will have been "filled", Is there somewhere to learn about try-catch?)
If all you're doing is catching them and printing the stack trace regardless of the exception type, you can just wrap the code in one large try/catch block. To save many "catches", you can catch java.lang.Throwable which is the interface that all exceptions implement. If not, you can have a catch for every type of checked exception the code you're calling throws, and handle them specifically.
Eclipse keeps asking you to do so because Java code will not compile if the checked exceptions are not caught, or declared to be thrown by the caller.
+Adding this comment to the answer (Thanks, Paul Tomblin):
In production quality apps you'd be logging the trace, adding some logic where you're handling the exception in a right way, taking an alternate flow, and/or re-wrapping it in another exception and throwing it, etc. It all depends on the particular problem you're trying to solve.
The idea of exception handling is that you can handle errors at points in your program flow where you can deal with them meaningfully. Rather than checking every function's return value like in C, where most of the time you can't do anything sensible other than passing the error further up, you install a try/catch block at sensible points in your program:
Basically, whenever there is a point where you can react meaningfully to an error, then catch that error, and pass everything else on. That way, error handling is only invoked when there is a plausible recovery from the error.
For example, worst case if any error stops your program from executing meaningfully, then you might almost not catch anything at all and just let the OS handle the situation (well, perhaps one single try/catch to produce a friendly error message).
Example (in C++, sorry, I'm can't type Java blind):
int main()
{
try {
while (masterloop()) { }
catch (...) {
LOG("Fatal program error, terminating!"); // nothing else we can do!
}
}
/* lots of program logic */
void process_image()
{
try {
Image im = load_image_from_disk();
/* ... */
}
catch (const OutOfMemoryExc & e) {
LOG("Not enough memory to process the image.");
return;
}
catch (const DataErrorExc & e) {
LOG("Could not read the image data.");
return;
}
catch (...) {
throw; // pass everything else along
}
}
In this example, we may try to process an image and fail for some anticipable reasons (out of memory, or failure to read the image). In that case we just return without doing work and let the program continue gracefully. All other errors are propagated up to a higher point. Most importantly, we do not need to litter the actual image processing function with error checks and responses all the time, it suffices for any code there to throw one of our two good exceptions and not worry any further.
Moral: If you have try/catch blocks absolutely everywhere, you're doing it wrong.
I know there's a lot of answers here, and they do a good job of covering how to structure the try/catch blocks. However, I'm thinking one of the things bothering you is the significant... indentation and code growth (... because I know it's not the indentation or amount of code, but the implied complexity by wrapping it and shifting it over and growing longer and longer between the opening try and enclosing catch, and I can't put a word to that apprehension).
The way to get around this is to refactor into functions the distinct bits in the code. I know it's a simplistic answer, but it's a good way to isolate individual tasks and keep the error handling fairly local to the code that requires it without padding things out vertically and horizontally with nested try/catch blocks.
You can make these methods private as they are intended for internal use only, presumably.
private Integer getDatastoreACount() {
try {
DatastoreService datastore = DatastoreServiceFactory.getDatastoreService();
Key qKey = KeyFactory.createKey("qu", qURL);
return (Integer) datastore.get(qKey).getProperty(kLastKnownANumber);
//..etc.. more try catch blocks needed
} catch (EntityNotFoundException e) {
// expects an Integer return, so need to deal with this
// but for simplicity I'm just simply recycling 'e'
throw e;
}
}
public void parseJSON(String jsonString) {
try {
JSONObject jsonObject = new JSONObject(jsonString);
int aCount = jsonObject.getInt("acount");
String devTok = jsonObject.getString("dt");
String qURL = jsonObject.getString("qu");
Integer dsACount = getDatastoreACount();
//etc etc
} catch (com.google.appengine.repackaged.org.json.JSONException e) {
e.printStackTrace();
}
}
You can catch multiple exceptions in the same try e.g.
try{
xyz;
}catch(NullPointerException npx){
npx.getMessage();
}catch(ArrayOutOfBoundsException ax){
ax.getMessage();
}
Also, by declaring the Exception as throws in your method signatures you can pass the Exception up the stack.
If you're just doing something like this:
try {
do smth
try {
do smth more
...
} catch (Exception1 e1) {reaction to e1}
} catch (Exception2 e2) {reaction to e2}
You can do everything in one try-block:
try {
do smth
do smth more
...
}
catch (Exception1 e1) {reaction to e1}
catch (Exception2 e2) {reaction to e2}
You can also break this down to one catch block if you're just printing the exception:
try {
do smth
do smth more
...
}
catch (Exception e) {e.printStackTrace();}
But this doesn't if you want to do somthing more, even if e1 is thrown, like:
try {
do smth
try {
do smth more
...
} catch (Exception1 e1) {reaction to e1}
do smth even if e1 was thrown
} catch (Exception2 e2) {reaction to e2}
The last example can't be written shorter.
If you have a block of code in which more than one type of exception may be thrown, you can declare two separate catch blocks:
try {
JSONObject jsonObject = new JSONObject(jsonString);
int aCount = jsonObject.getInt("acount");
String devTok = jsonObject.getString("dt");
String qURL = jsonObject.getString("qu");
DatastoreService datastore = DatastoreServiceFactory.getDatastoreService();
Key qKey = KeyFactory.createKey("qu", qURL);
int dsACount = (Integer) datastore.get(qKey).getProperty(kLastKnownANumber);
} catch (EntityNotFoundException e) {
e.printStackTrace();
} catch (com.google.appengine.repackaged.org.json.JSONException e) {
e.printStackTrace();
}
//..etc.. as many catch blocks as needed
Alternatively, if you don't care about the exact type of the exception, you can have onyl one catch block which catches Exception (or maybe Throwable; I can't remember exactly what the superclass of exceptions is in Java).
Another point I will make now is that you might not have the most modular code. Remember, code that does one thing well makes for good, modular code. If you find that you have many nested black (whether try/catch blocks, if/else blocks, etc.) you may want to check if some of the code can be extracted into its own method. This may also make your code look better when many exceptions must be handled.
First, from a design perspective, catching and printing exceptions is not a good thing. Something went wrong, and your code just keeps going in the same way as if it went right. That is not usually correct. So: perhaps your method needs to throw these exceptions instead of catching them. Perhaps only the caller is in a position to decide what happens if something fails like this.
But otherwise, the only advice I can offer to clean up how the code looks, syntactically, is to tell you that you can write:
try {
...
} catch (...) {
...
} catch (...) {
...
}
You can also catch for a broader exception class like Exception and just write one catch block but this is bad design. In Java 7, you will be able to catch for several exception types in one block.
You should use try/catch blocks if you have a way to recover from the exception, for example if you want to check if a string is a valid integer, you can write a method (this is a lame method, but just to show the idea):
public boolean isInteger(String str) {
try {
new Integer(str);
}
catch(NumberFormatException e) {
return false;
}
return true;
}
If you don't have a way to recover from the exception and all you do is to print the stack trace, it is suggested to add throws declaration (as eclipse suggest) to the method, and let the caller handle the exception (or throw it to its caller).
If you want to handle some exceptions and throw some other, you can do it as well.
I like to box up the call behind a static method, just to keep it tidier. For example, here's my reduced Set Json Value call.
private static boolean setJsonValue(JSONObject j,String key,Object value)
{
try
{
if(value instanceof Integer)
{
// numbers are special. We want them unquoted.
int valueI = (Integer)value;
j.put(key,valueI);
}
else
j.put(key,value);
return true;
}
catch (JSONException e)
{
// do nothing, it turns out
return false;
}
}
...and then I ignore the return values, because I am bad.
Somewhere or other I have a similar Get method, that returns null if it fails. You get the idea.
You have two basic code-style choices here (that don't involve changing method signatures)
Method1: Put everything in the one try catch and have multiple catch blocks, like this:
try {
JSONObject jsonObject = new JSONObject(jsonString);
int aCount = jsonObject.getInt("acount");
String devTok = jsonObject.getString("dt");
String qURL = jsonObject.getString("qu");
DatastoreService datastore = DatastoreServiceFactory.getDatastoreService();
Key qKey = KeyFactory.createKey("qu", qURL);
int dsACount = (Integer) datastore.get(qKey).getProperty(kLastKnownANumber);
//..etc.. more try catch blocks needed
} catch (EntityNotFoundException e) {
e.printStackTrace();
} catch (com.google.appengine.repackaged.org.json.JSONException e) {
e.printStackTrace();
}
Method 2: Break up your code into sections that each have one catch, like this:
String qURL = null;
try {
JSONObject jsonObject = new JSONObject(jsonString);
int aCount = jsonObject.getInt("acount");
String devTok = jsonObject.getString("dt");
String qURL = jsonObject.getString("qu");
} catch (EntityNotFoundException e) {
e.printStackTrace();
}
try {
DatastoreService datastore = DatastoreServiceFactory.getDatastoreService();
Key qKey = KeyFactory.createKey("qu", qURL);
int dsACount = (Integer) datastore.get(qKey).getProperty(kLastKnownANumber);
} catch (EntityNotFoundException e) {
e.printStackTrace();
}
Method 2 is the recommended one, as it makes it obvious which lines are throwing which exceptions and generally segments the code into natural processing blocks.
Create another exception and put it below or above of the other exception. Depends on the context of your application.