I'm maintaining a Java class that's 40K li

问题:

This may be a subjective question leading to deletion but I would really like some feedback.

Recently, I moved to another very large enterprise project where I work as a developer. I was aghast to find most classes in the project are anywhere from 8K to 50K lines long with methods that are 1K to 8K lines long. It's mostly business logic dealing with DB tables and data management, full of conditional statements to handle the use cases.

Are classes this large common in large enterprise systems? I realize without looking at the code it's hard to make a determination, but have you ever worked on a system with classes this large?

回答1:

Here are the ten largest class in the JDK 6 by line count of 7209 .java files. These classes include significant amount of comments which could be longer than the code.

4495 ./javax/sql/rowset/BaseRowSet.java
4649 ./java/awt/Container.java
5025 ./javax/swing/text/JTextComponent.java
5246 ./java/util/regex/Pattern.java
5316 ./javax/swing/JTree.java
5469 ./java/lang/Character.java
5473 ./javax/swing/JComponent.java
9063 ./com/sun/corba/se/impl/logging/ORBUtilSystemException.java
9595 ./javax/swing/JTable.java
9982 ./java/awt/Component.java

I would agree one printed page is long enough for a method. There really should not be a need for classes over 10K lines long IMHO.

回答2:

This is definitely not right. A method should not contain more code than sufficient for a single unit of work. A class should not contain more methods than the ones related to the state of the class' instance.

This is too much like God Object anti-pattern. I would personally drop the project and look for another.

回答3:

Without looking at the code, it actually remains quite easy to make a determination. Never should a class be 40K lines, and never should a method be even 1K. Typically, if I can't print out a method on a piece of paper and see both the beginning and end brackets, I find a way to split it up.

Might I ask, are they using OOP principles at all, or are they trying to use Java more as a functional or procedural language? I can't imagine a truly OOP project having a 40K line class.

回答4:

Oh, I think this a terrible sign, and I don't have to look at the code to say so. Sounds like a massive refactoring effort is needed.

Let me guess - you have no unit tests for the system as written, either. You have my sympathy.

回答5:

In addition to software maintenance issues described in other answers, be careful of the technical limit that a compiled Java method may not exceed 64k bytes. (How many lines of code that will correspond to will depend on the lines themselves.)

http://www.databasesandlife.com/java-method-64k-limit/

回答6:

In 12 years of Java development I can honestly say that this is unusual.

In fact; I have never come across files or classes of that size in any language in over 25 years of development.

Crack out the refactoring tools!

回答7:

A small thing to pay attention to is the difference between lines, lines of code, and statements. If you analyze your project with e.g. Sonar you can easily see the difference between those.

Nevertheless, whatever the exact measure, 40k lines of business code is hideous.

In the business module of an enterprise application I develop, the highest number is 444 lines of code. This is for a rather large Service. Most Service classes are between 200 and 100 lines of code. Entities (model objects) are in our situation mostly between 40 and 100 loc.

In another part of this same application we have one class that is 1224 lines of code (2477 lines total, 706 statements). This class is almost universally hated within the team because of its size. It's perceived as bloated, complicated and doing way too much.

Now if an entire team thinks this about a class that's only 2477 lines total, this may give you some perspective about what kind of abomination a 40k lines class is.

回答8:

An alarm bell started going off when I read through this :

It's mostly business logic dealing with DB tables and data management, full of conditional statements to handle the use cases.

If this code isn't in the data layer and there is no abstraction with respect to how the database is accessed, something is wrong. I also have the feeling that some of these methods aren't directly related to the classes where they are found. The comment about conditional statements and use cases doesn't sound right either. I'll echo duffymo's comment that some refactoring would be needed.

回答9:

As a young programmer I still remember my teacher telling us to break up big functions and work on a good OO design before writing code.

So unless there is a REALLY good reason in your design to impose 40k lines (which I strongly doubt) then you already have your answer : your class is too big.

I will quote my wife (who is chemist and doesn't program) : "40k lines of code, there is something really wrong!"

I've had friends take up projects in their companies that were really old, tossed from one programmer to the over and what we all agreed on is that a class that size simply means:

-patch and fix : people had to do minor changes here and there and didn't want to/ didn't have the time to do it corretly.

at runtime there might not be any problems with that code, everything works, but usually problems occure when you want to do any form of modifications:

• it takes ages to find anything

• when there is a bug you can't pin point it easily

...

In conclusion I would sit down rethink the oo design of your project and restructure (at least into 1k ~ 5k lines classes to start with), I know its annoying to do bu usually on the long run its better

回答10:

50K lines of code? I thought KLOC was a metric of project size, not file size. That's like our whole codebase (including tests).

I'm working with JavaScript, so it's not directly comparable, but we only have few files that are over 500 lines long - and these are the highly problematic ones.

回答11:

I know some section of code needs breaking up when I have problems trying figure out what it does and how. Or if its larger than a screen-shot.

Typically I refactor it if it makes me feel bad. I don't like feeling bad :-(

I suspect the code got out of control because the management just wanted bugfixes/features done and nothing else, and bit by bit it slowly got worse.

Refactoring the code to make the programmers job easier was probably not very high on their list of priorities. The management always want their bugfixes/features yesterday :-(

Also a working program is obviously better than one thats broke, and breaking up code that large without unit-tests would end in disaster. So all the tests would have to be made before touching the code. Yet another reason management wouldn't allow it.

回答12:

I think your aghastness is qualified :) I can't imagine that the program is properly OOPified. Classes are a bit tougher to classify, but methods are easy: 1 behavior per method (that's not a rule, but it should be). Behaviors can't possibly be even close 1k lines of code. At least, as far as my imagination will take me.

Classes, on the other hand, can represent many things but they should represent something. If it is difficult to tell what the class represents, then you have a problem.

Now, I half imagine that you are well aware of these concepts and I'm preaching to the choir. So, I'll just pretend like I didn't go off on a tangent there and answer your question directly:

Yes. Very unfortunately, it is remarkably common for large enterprise projects to have code that lazy. I have worked on projects almost as large (yours shoots anything I've seen out of the water) and my first tendency is to start breaking things up into logical components, particularly in places where I intend on making changes. I can't handle that kind spaghetti, it's too irritating.