Checked Exceptions do not exist on the JVM.
Reinier Zwitserloot
reinier at zwitserloot.com
Wed May 20 06:35:18 PDT 2009
Ruslan, you've got it all wrong. Though, you've just proved to this
list that java programmers don't really get this concept, which is one
of the reasons why I want to introduce it: Make it obvious to java
programmers that the JVM does *NOT* guarantee that checked exceptions
are declared.
Checked exceptions do not exist.
That's right. They don't. There's no such thing. It's all a figment of
javac. The JVM does not have checked exceptions. All it knows about is
'throws clauses', which are part of a method's signature and have ZERO
impact on how a JVM runs a class file. As far as the JVM is concerned,
throws clauses are just a comment. Utterly ignored by javac. You can
strip every single one out of every class file and run an app, and the
execution of it would be 100% the same. The only change would be in
java.lang.reflect's getExceptionTypes() method.
I repeat: The JVM *DOES* *NOT* know about checked exceptions. The JVM
assumes all exceptions are unchecked, and assumes that javac sorted it
all out. This is flawed, obviously, because javac isn't the only tool
out there that can make class files. This is similar to generics in
objects, which is also a figment of javac's imagination: On the JVM
level, erasure occurs - that info just isn't there. If you want, you
can name this 'checked exception erasure'. The notion that exceptions
are checked is erased during compilation.
The only application in existence that cares about this 'throws
clauses' comment in class files is the application called 'javac'.
Javac will refuse to compile a throws statement, or a call to a method
that has a throws clause on it, unless this statement occurs in either
a try block with the appropriate catch, or the method throws the
exception onwards.
So, let's revisit:
If I take this code:
public class Foo {
public void foo() throws IOException {
throw new IOException();
}
}
compile it with javac, then use a hex editor or ASM to remove the
'throws IOException' from the method signature, *nothing changes*. The
Class Verifier will not complain. The JVM will just throw that
exception; it won't get wrapped into an unchecked exception.
It gets worse: JRuby, groovy, Jython, and scala, they ALL feature
*ZERO* checked exception clauses on methods. They just generate class
bytecode with throw statements without caring about the throws clause
on methods. This is why, in java, anytime you interop with code not
compiled by javac, you can get checked exceptions in situations where
you didn't expect them. Millions of classes generated by those
compilers are out there right now throwing checked exceptions that
escape from methods that do NOT have a 'throws X' clause on them.
In other words: There is *NO* guarantee whatsoever in the JVM that
checked exceptions can only occur in situations where they are
explicitly declared. If your code relies on it, then your code is buggy.
Secondly, you mentioned that java wraps exceptions that occur in the
constructor with Class.newInstance() into an
ExceptionInInitializerError. That's not what happens at all; That
exception (ExceptionInInitializerError) is for static initializers and
object initializers, not constructors. A practical example will make
this clear. Take the following code, save it in Foo.java, compile it,
and run it:
import java.io.IOException;
public class Foo {
public static class Test {
public Test() throws IOException {
throw new IOException();
}
}
//Note how I didn't add 'throws IOException' in the next line!
public static void main(String[] args) throws
InstantiationException, IllegalAccessException {
Test x = Test.class.newInstance();
}
}
The result is an IOException. Not a ObjectInstantionException with an
IOException as cause; no - an IOException. You can doublecheck this
by wrapping the newInstance in a try/catch (Throwable t) block and
printing the class name of the resulting t.
Thirdly, you misunderstand what my ignoreslist proposal does: It does
not silently swallow the exception, that'd be a very stupid way of
handling exceptions that you don't -think- can occur. An ignoreslist
has zero effect on the execution of a method body. The only difference
is that, unlike a throws clause, you don't burden any of the methods
that call you to handle the listed exception. It's a signal to javac
to disable its rigid checking of making sure any checked exception
thrown in the body is also present in that method's 'throws' clause.
It's exactly analogous to adding the 'throws' clause, compiling the
code, then using a hex editor to remove the throws clause.
You're also mistaken in retrofitting Runnable with 'throws Exception':
That'll break many thousands of classes out there in real life. You
forget that people don't just make Runnables and hand them off to
java.lang.Thread; there's plenty of code out there that *receives*
Runnables, and that code presumes that its run() method does not throw
any checked exceptions. It would be backwards compatible (because, as
I said, the JVM doesn't know about these things), but it wouldn't be
source compatible; recompiling your code would all of a sudden produce
errors. e.g:
public void myMethod(Runnable r) {
r.run();
}
would no longer compile, because all of a sudden 'r.run();' can throw
Exception, which I need to catch, or declare as thrown (or, if this
proposal is accepted, declare as sneakyThrown/ignore).
--Reinier Zwitserloot
On May 20, 2009, at 07:01, Ruslan Shevchenko wrote:
>> Here's a trivially simple change to implement, though it has some far
>> reaching repercussions for java as a whole:
>>
>> change the 'rethrows' clause to an ignores clause: Any checked
>> exception in the ignores list may be thrown from the method body, but
>> is NOT part of the method's checked exception list. It is hence also
>> not part of the signature, just like sticking a 'synchronized'
>> keyword
>> on a method isn't part of its signature, but an implementation
>> detail.
>>
>>
>> example:
>>
>> Runnable r = new Runnable() {
>> public void run() ignores UnsupportedEncodingException {
>> String x = new String(inBytes, "UTF-8");
>> }
>> };
>>
>
> And how actually debug such statements ?
> //In this example, run() method do nothing, so look on slightly
> complex
> primer:
>
> class X
> {
> String x;
>
> public void setString(byte[] inBytes)
> ignores UnsupportedEncodingException
> {
> x = new String(inBytes, "UUTF-8");
> }
>
> public void setAndPrintString(byte[] inBytes)
> {
> setString(inBytes);
> System.out.println("received:"+x);
> }
>
>
> }
>
> Assume, that programmer wrote (by error) "UUTF-8" instead "UTF-8"
>
> Output of programm will be "received:null".
> Reasons for null x in complex programs can be millions. Debugging with
> ignoring exception will be extremally hard.
>
> I. e. wrapping checked exception to runtime is more or less normal;
> ignoring will become a very big paint-point.
>
> Alternative to ignore can be call default thread exception handler.
> (handle keyword ?)
>
>> This sounds like blowing up the entire point of checked exceptions,
>> but with java moving towards more interoperation with other JVM based
>> languages, which almost always throw checked exceptions without
>> actually declaring them (Jython, JRuby, Scala -- I can't think of a
>> single one other than java itself that has the notion of checked
>> exceptions in the first place). If the whole 'java is like the
>> assembler of the JVM' story is going to hold water, something like
>> this is needed.
>>
>
> Yes, this is interesting problem.
> In theory we have UndeclaredThowableException, but .. wrapping each
> invocation of foreign language into it is ineffecient.
>
> But InvokeDynamic can throws any throwable in declaration, so this
> problem
> can be solved (on other level)
>
>> For some further support: I doubt anyone seriously holds the position
>> that java's checked exception experiment was an unmitigated success.
>> There are pain points, here and there. The above example in fact
>> includes two of them: The fact that runnable throws no checked
>> exceptions is clearly a mistake in regards to threading, because
>> Thread objects clearly should handle ANY exception falling out of its
>> run() method (and it in fact does, via the unchecked exception
>> handler
>> mechanism), and there's a general mismatch between main(), which MAY
>> throw exceptions, and run(), which may not, even though they are both
>> 'thread start points'. Another mistake is
>
> Adding throw exception to signature of Runnable.run does not broke
> anything in existent code, so this can be nice library change.
>
>> UnsupportedEncodingExceptions, a checked exception, resulting from a
>> call to converting byte arrays to strings using an encoding that is
>> guaranteed by the JVM using a string literal, such as "UTF-8". The
>> appropriate exception would be UnsupportedEncodingError("This JVM is
>> broken. Charset UTF-8 missing. Reinstall it") - which is unchecked.
>>
>
> Or porgrammer can write UUTF-8 instead UTF-8. It's happens ;)
>
>> In other words, while I'm not ready to write off checked exceptions
>> as
>> a whole, I am more than ready to acknowledge that the programmer more
>> often knows better than the compiler, compared to how often the
>> programmer, in laziness or inexperience, abuses this freedom.
>> Especially considering that, on the JVM, you really DONT get checked
>> exceptions guarantees; any class compiled by anything other than
>> javac
>> is free to throw checked exceptions without declaring them, and even
>> in javac you can technically do this using 'sneaky throw' libraries,
>> which use various workarounds, one of which will NEVER go away*.
>> We're
>> not losing much here, in other words. In fact, the way javac works,
>> inexperienced java programmers may erroneously assume that checked
>> exceptions couldn't possibly happen unless the checked exception is
>> declared. This is not the case.
>>
>> *) The one that will never go away is java.lang.Class.newInstance(),
>> which is functionally defined to sneakily throw on any exceptions
>> thrown by the constructor, checked or not, and does NOT wrap them up
>> into a wrapper the way java.lang.reflect.Method.invoke() does. This
>
> Its wrapped to java.lang.ExceptionInInitializerError
>
>
>> cannot change without breaking backwards compatibility. Another one
>> that will never go away is constructing a class on the fly that
>> sneaky
>> throws, loading it with ClassLoader's load from bytes mechanism, and
>> executing it.
>>
>
> And what is JVM behaviour in such case ?
>
>>
>> Attempting to solve this issue with a library requires rather a lot
>> of
>> ugly boilerplate:
>>
>> import static Utils.sneakyThrow;
>>
>> Runnable r = new Runnable() {
>> public void run() {
>> String x;
>> try {
>> x = new String(inBytes, "UTF-8");
>> } catch ( UnsupportedEncodingException e ) {
>> sneakyThrow(e);
>> }
>> }
>> };
>>
>>
>> NB: For sanity purposes, the restriction on javac that the try block
>> MUST contain at least 1 statement that could throw a checked
>> exception
>> type that is listed on one of the accompanying catch blocks (other
>> than Exception, which you can always catch), should go away; this has
>> already been proposed before, and in fact is a backwards
>> compatibility
>> painpoint for the coin proposal that allows you to rethrow a final
>> Exception as if its type is the intersection of all checked types
>> thrown by the statements in the try body.
>>
>> --Reinier Zwitserloot
>> Like it? Tip it!
>> http://tipit.to
>>
>>
>>
>> On May 20, 2009, at 02:22, Mark Mahieu wrote:
>>
>>> Yeah, I suppose I did make it rather more complicated than it needed
>>> to be.
>>>
>>> Thanks for the feedback.
>>>
>>> Mark
>>>
>>>
>>> On 19 May 2009, at 23:59, Joe Darcy wrote:
>>>
>>>> Hello.
>>>>
>>>> Hmm; this strikes me as a bit involved for the potential benefits
>>>> of the feature.
>>>>
>>>> -Joe
>>>>
>>>> On 03/30/09 05:04 PM, Mark Mahieu wrote:
>>>>> Forgot my JLS references in the first one...
>>>>>
>>>>>
>>>>>
>>>>> HTML version + prototype available at:
>>>>>
>>>>> http://slm888.com
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> Rethrows Clause
>>>>> v0.1.1
>>>>>
>>>>>
>>>>> AUTHOR(S):
>>>>>
>>>>> Mark Mahieu
>>>>>
>>>>>
>>>>> OVERVIEW
>>>>>
>>>>> FEATURE SUMMARY: Should be suitable as a summary in a language
>>>>> tutorial.
>>>>>
>>>>> A new clause on method declarations which allows exception
>>>>> translations (wrapping and rethrowing as a different type) to be
>>>>> cleanly defined separately from the body of the method. In many
>>>>> cases, checked exception type names do not then need to be
>>>>> repeated in a method's throws clause and in a throw statement in
>>>>> the method body.
>>>>>
>>>>>
>>>>> MAJOR ADVANTAGE: What makes the proposal a favorable change?
>>>>>
>>>>> The proposal adds direct support for a common idiom in daily use
>>>>> by Java programmers worldwide, allowing them to express their
>>>>> intentions with greater clarity and ease. In comparison with
>>>>> some proposals, this is an attempt to make dealing with checked
>>>>> exceptions easier by increasing the expressiveness of exception
>>>>> handling code in general, rather than by attempting to deprecate
>>>>> checked exceptions in favour of unchecked exceptions.
>>>>>
>>>>>
>>>>> MAJOR BENEFIT: Why is the platform better if the proposal is
>>>>> adopted?
>>>>>
>>>>> There is a reduction in the amount of boilerplate Java
>>>>> programmers have to read and write for code dealing with checked
>>>>> exceptions. Declarations specifying both thrown and rethrown
>>>>> (wrapped) exceptions are kept together, aiding comprehension of
>>>>> the code.
>>>>>
>>>>>
>>>>> MAJOR DISADVANTAGE: There is always a cost.
>>>>>
>>>>> As with any syntax sugar which enables an alternative way of
>>>>> expressing an existing idiom, programmers may be tempted to use
>>>>> it even when the existing idiom would be more appropriate.
>>>>>
>>>>>
>>>>> ALTERNATIVES: Can the benefits and advantages be had some way
>>>>> without a language change?
>>>>>
>>>>> No.
>>>>>
>>>>>
>>>>> EXAMPLES
>>>>>
>>>>> SIMPLE EXAMPLE: Show the simplest possible program utilizing the
>>>>> new feature.
>>>>>
>>>>>
>>>>> Before:
>>>>>
>>>>> void before() throws ConfigException {
>>>>> try {
>>>>> Class.forName("where.is.the.Code");
>>>>> }
>>>>> catch (ClassNotFoundException e) {
>>>>> throw new ConfigException(e);
>>>>> }
>>>>> }
>>>>>
>>>>> After:
>>>>>
>>>>> void after()
>>>>> catch ClassNotFoundException throw ConfigException {
>>>>>
>>>>> Class.forName("here.is.the.Code");
>>>>> }
>>>>>
>>>>>
>>>>> ADVANCED EXAMPLE: Show advanced usage(s) of the feature.
>>>>>
>>>>> Before:
>>>>>
>>>>> void suspendAccount()
>>>>> throws AuthorizationException,
>>>>> PersistenceException {
>>>>> try {
>>>>> checkMyAuthoritah();
>>>>> db.update(/*...*/);
>>>>> log.recordInfo(/*...*/);
>>>>> }
>>>>> catch (InfernalDBException e) {
>>>>> throw new PersistenceException(e);
>>>>> }
>>>>> catch (InfernalLogException e) {
>>>>> throw new RuntimeException(e);
>>>>> }
>>>>> }
>>>>>
>>>>> After:
>>>>>
>>>>> void suspendAccount()
>>>>> throws AuthorizationException
>>>>> catch InfernalDBException throw PersistenceException,
>>>>> InfernalLogException throw RuntimeException {
>>>>>
>>>>> checkMyAuthoritah();
>>>>> db.update(/*...*/);
>>>>> log.recordInfo(/*...*/);
>>>>> }
>>>>>
>>>>>
>>>>> DETAILS
>>>>>
>>>>> SPECIFICATION: Describe how the proposal affects the grammar,
>>>>> type system, and meaning of expressions and statements in the
>>>>> Java Programming Language as well as any other known impacts.
>>>>>
>>>>> The syntactic grammar is modified to allow an optional rethrows
>>>>> clause immediately prior to a MethodBody:
>>>>>
>>>>> MethodDeclaratorRest:
>>>>> FormalParameters {[]} [throws
>>>>> QualifiedIdentifierList] ( ( [catch ExceptionTranslationList]
>>>>> MethodBody ) | ; )
>>>>>
>>>>> ExceptionTranslationList:
>>>>> QualifiedIdentifier throw QualifiedIdentifier
>>>>> { , ExceptionTranslationList }
>>>>>
>>>>>
>>>>>
>>>>> JLSv3 ї8.4.6 : A rethrows clause lists one or more exception
>>>>> translations, each translation consisting of a caught type C and
>>>>> a translated type T for which all of the following must hold:
>>>>> * C <: java.lang.Exception
>>>>> * T < java.lang.Throwable
>>>>> * Neither C nor T is a type variable.
>>>>> * T has an accessible constructor suitable for rethrowing a
>>>>> value of type C (see below).
>>>>> * T is not the same type as C.
>>>>>
>>>>> Any exceptions thrown by the method body which are a subtype of a
>>>>> caught exception type in the rethrows clause, are rethrown as the
>>>>> corresponding translated exception type.
>>>>>
>>>>> For a given translated type T with corresponding caught type C, if
>>>>> T has an accessible constructor accepting a value of type C, then
>>>>> the translation is equivalent to the following:
>>>>>
>>>>> catch (C e) {
>>>>> throw new T(e);
>>>>> }
>>>>>
>>>>> Otherwise it must have an accessible no argument constructor, and
>>>>> the translation is equivalent to:
>>>>>
>>>>> catch (C e) {
>>>>> throw new T().initCause(e);
>>>>> }
>>>>>
>>>>>
>>>>> A rethrows clause does not restrict which types may appear in a
>>>>> throws clause for the same method. In particular, for a given
>>>>> caught type C in the rethrows clause, it is permitted for some
>>>>> type C1 :> C to also be listed in the throws clause.
>>>>>
>>>>>
>>>>> JLSv3 ї8.2 : The set of exception types declared to be thrown by
>>>>> a method is the union of:
>>>>> * the types in the throws clause
>>>>> * the translated types in the rethrow clause
>>>>> * the types thrown by the translated types' selected
>>>>> constructors
>>>>>
>>>>> JLSv3 ї11.2.2 : For the purposes of exception analysis, the set
>>>>> of checked exception types which may be thrown by the method's
>>>>> body is the union of:
>>>>> * the types in the throws clause
>>>>> * the caught types in the rethrows clause
>>>>>
>>>>> JLSv3 ї11.2.3 : It is a compile-time error if a rethrows clause
>>>>> contains a translation from a checked exception type C but there
>>>>> exists no checked exception type E such that all of the following
>>>>> hold:
>>>>> * E <: C
>>>>> * The method body can throw E
>>>>> * No preceding translation in the rethrow clause catches E or
>>>>> a supertype of E
>>>>> unless C is the class java.lang.Exception.
>>>>>
>>>>> JLSv3 ї13.4.21 : Changes to the rethrows clause of methods or
>>>>> constructors do not break compatibility with existing binaries;
>>>>> these clauses are checked only at compile time.
>>>>>
>>>>>
>>>>>
>>>>> COMPILATION: How would the feature be compiled to class files?
>>>>> Show how the simple and advanced examples would be compiled.
>>>>> Compilation can be expressed as at least one of a desugaring to
>>>>> existing source constructs and a translation down to bytecode. If
>>>>> a new bytecode is used or the semantics of an existing bytecode
>>>>> are changed, describe those changes, including how they impact
>>>>> verification. Also discuss any new class file attributes that are
>>>>> introduced. Note that there are many downstream tools that
>>>>> consume class files and that they may to be updated to support
>>>>> the proposal!
>>>>>
>>>>> A simple desugaring could consist of enclosing the method body's
>>>>> statements in a try statement, with catch clauses for each
>>>>> translated exception type. For example, the following method:
>>>>>
>>>>> Method findMethod()
>>>>> catch ClassNotFoundException throw ConfigException,
>>>>> NoSuchMethodException throw ConfigException {
>>>>>
>>>>> Class<?> c = Class.forName("some.Thing");
>>>>> return c.getDeclaredMethod("execute", null);
>>>>> }
>>>>>
>>>>> would be desugared to:
>>>>>
>>>>> Method findMethod()
>>>>> throws ConfigException {
>>>>> try {
>>>>> Class<?> c = Class.forName("some.Thing");
>>>>> return c.getDeclaredMethod("execute", null);
>>>>> }
>>>>> catch (ClassNotFoundException e) {
>>>>> throw new ConfigException(e);
>>>>> }
>>>>> catch (MethodNotFoundException e) {
>>>>> throw new ConfigException(e);
>>>>> }
>>>>> }
>>>>>
>>>>> No changes to the classfile format are required.
>>>>>
>>>>>
>>>>> TESTING: How can the feature be tested?
>>>>>
>>>>> An initial set of jtreg tests is included in the prototype.
>>>>>
>>>>>
>>>>> LIBRARY SUPPORT: Are any supporting libraries needed for the
>>>>> feature?
>>>>>
>>>>> No
>>>>>
>>>>>
>>>>> REFLECTIVE APIS: Do any of the various and sundry reflection APIs
>>>>> need to be updated? This list of reflective APIs includes but is
>>>>> not limited to core reflection (java.lang.Class and
>>>>> java.lang.reflect.*), javax.lang.model.*, the doclet API, and
>>>>> JPDA.
>>>>>
>>>>> com.sun.source.tree.MethodTree would require updates to access
>>>>> the rethrows clause's caught and translated types.
>>>>>
>>>>>
>>>>> OTHER CHANGES: Do any other parts of the platform need be updated
>>>>> too? Possibilities include but are not limited to JNI,
>>>>> serialization, and output of the javadoc tool.
>>>>>
>>>>> No
>>>>>
>>>>>
>>>>> MIGRATION: Sketch how a code base could be converted, manually or
>>>>> automatically, to use the new feature.
>>>>>
>>>>> Catch clauses which simply wrap and rethrow an exception as
>>>>> another exception type not caught in an enclosing scope, can be
>>>>> trivially replaced with a rethrows clause, either manually or
>>>>> automatically.
>>>>>
>>>>> It should be possible for tools to offer bidirectional
>>>>> conversions such that an exception translation may be moved back
>>>>> into the method body if it is subsequently decided that
>>>>> additional logic is required.
>>>>>
>>>>>
>>>>> COMPATIBILITY
>>>>>
>>>>> BREAKING CHANGES: Are any previously valid programs now invalid?
>>>>> If so, list one.
>>>>>
>>>>> No
>>>>>
>>>>>
>>>>> EXISTING PROGRAMS: How do source and class files of earlier
>>>>> platform versions interact with the feature? Can any new
>>>>> overloadings occur? Can any new overriding occur?
>>>>>
>>>>> The semantics of existing class files and legal source files are
>>>>> unchanged by this feature.
>>>>>
>>>>>
>>>>> REFERENCES
>>>>>
>>>>> EXISTING BUGS: Please include a list of any existing Sun bug ids
>>>>> related to this proposal.
>>>>>
>>>>> http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=6534270
>>>>> (similar, but emphasizes unchecked exceptions)
>>>>>
>>>>>
>>>>> URL FOR PROTOTYPE (optional):
>>>>>
>>>>> http://slm888.com/javac
>>>>>
>>>>>
>>>>>
>>>>
>>>
>>>
>>
>>
>>
>
>
More information about the coin-dev
mailing list