已检查与未检查异常统计

Question

After getting tired of convincing my peers to abandon checked exception in the argumentative way, I am looking for any statistics that will say how many programmers prefer checked exception, and how many prefer not to use them.
Is there such a survey / voting / statistics on the web?

Answer 1

How about pointing them to the Java Tutorial:

http://download.oracle.com/javase/tutorial/essential/exceptions/runtime.html

which includes a final line which summarizes the key point clearly:

Here's the bottom line guideline: If a client can reasonably be expected to recover from an exception, make it a checked exception. If a client cannot do anything to recover from the exception, make it an unchecked exception.

Answer 2

One problem with many languages and frameworks which derive their general exception-handling concepts from C++ is that way too many things are effectively encoded in the type of an exception. Checked exceptions would be a good concept, if the checked-versus-unchecked distinction were made at the throw- and catch sites, rather than in the declaration of the exception , and if there were a way to concisely specify that particular checked exceptions should get rethrown as unchecked. Unfortunately, Java requires that any exception of a given type always be thrown as checked, or always as unchecked, depending upon the type, and provides no convenient or concise way to map checked exceptions to unchecked ones. Instead, it makes it most convenient to have checked exceptions percolate up as checked exceptions, which increases the "nuisance" factor while reducing their usefulness.

If checked exceptions were implemented at the throw site, and if the normal way of handing unwanted checked exceptions was to convert them to unchecked, then one could be assured that a method which indicates a particular fault condition by throwing a particular checked exception would only throw that checked exception when it "wanted" to report that particular fault; if the same type of exception occurred, unanticipated, during the execution of a method, it would get converted to an unchecked exception. Such a "checked exceptions" model could be useful, and could also lead to more efficient code than normal exceptions (since checked exceptions would have to be handled in the immediate caller of the routine throwing them, much of the stack unwinding and stack tracing required for normal exception handling could be skipped). Unfortunately, Java's "checked exceptions" don't work that way.

Answer 3

If there was a survey like this wouldn't your cow-orkers just argue there are more bad programmers than good programmers on the web, and they would expect more bad programmers voted for unchecked exceptions? So it doesn't really solve your problem.

Two data points are convincing to me:

• No other language implementer anywhere afaik has copied checked exceptions from Java. It is a total flop as far as catching on in other languages. If there was anything good about the feature somebody would have appropriated it for their new language.

• There has been a broad movement among framework writers, from Spring to Hibernate to Java EE, to replace many checked exceptions with unchecked ones.

Besides that you could try reviewing your projects' code and finding examples of how checked exceptions lead to problems with exceptions getting eaten or useless "throws Exception" clauses everywhere.