UmlGraph vs APIViz for Maven javadoc generation - java

I'm wondering if there are any clear reasons to choose UmlGraph over APIViz for javadoc UML diagram generation in a Maven2 build. Are there any integration or features that one has over the other, they seem pretty similar?

There is an interesting thread here about UMLGraph vs apiviz (which are mentioned in this question on SO too) and my understanding is the following:
UMLGraph is older but is really nice (dixit Fowler which has more weight than me).
apiviz is a rewrite of UMLGraph but it's not really clear why the rewrote it.
The "big" advantage of apiviz is that the build won't fail if Graphviz isn't installed (this doesn't justify a whole rewrite though in my opinion).
apiviz doesn't generate graphs automatically, you have to add tags versus UMLGraph do it automatically by default (which can be convenient) and the totally automatic results is fine and easy for all (UMLGraph can also model composition relation with custom javadocs tags).
UMLGraph seems to do more things ("it looks like apiviz does nothing more than provide a view of generalizations and interface utilization" or "the class usage shown by UMLGraph actually helps to understand how a class works") but I didn't check myself and this might be not true anymore or outdated.
So, if you project is public, the fact that apiviz gracefully works without Graphviz is nice. If you have lots of classes, having to add tags in javadoc everywhere is a big pain. If you're looking for special features, UMLGraph might offer more. Actually, I think you should experiment both as you obviously have a better knowledge of your needs and constraints for the final choice.

It is not true that apiviz requires annotations for the generation of graphs, as its documentation clearly states:
APIviz automatically discovers the relationship between packages and classes and their stereotype by default. Additionally, you can use the following doclet tags to add more relationships or to make the generated diagram look cleaner.

Related

Is Eclipse's ecj compiler extensible?

I am interested in modifying Java syntax and some implicit paradigms. Since I develop with Eclipse which provides it's own compiler, which can also be used standalone, I was wondering if it wasn't possible to extend ecj to respect additional grammar rules (and correctly handle them).
My syntactical changes are all resolvable by removing elements from the AST and creating some new ones, so I assume that what I want to do is possible without diving into bytecode.
Essentially, what I want to do could be done by 'virtually' modifying the source code before the actual compilation. However I suspect that doing so would mess up the source mapping, which would make debugging a hell.
On a sidenote: I am aware of project Lombok, which extends and alters class compilation, however Lombok uses annotations only, and does not modify syntax, strictly speaking. So what I want to do is more invasive to the language specs.
As Object Teams has been mentioned in comments:
(1) Object Teams itself extends JDT for its own language OT/J which is an extension of Java. This is done in a dual strategy:
We maintain a fork of org.eclipse.jdt.core. While this is quite heavy lifting it successfully demonstrates that the JDT architecture is suitable for modification.
We use our own concepts of role objects to non-invasively adapt the behavior of other parts of the IDE (notably org.eclipse.jdt.ui) to reflect the semantics of OT/J
(2) I have a few (oldish) blog posts that demonstrate how OT/J can be used for creating non-invasive variants of JDT including support for extended syntax:
IDE for your own language embedded in Java? (part 1)
IDE for your own language embedded in Java? (part 2)
Get for free what Coin doesn’t buy you
Disclaimer: I am author of OT/J and lead of its implementation, and later became a committer on Eclipse JDT.
For further questions, there's a forum.

Should annotations in jar305.jar be preferred over similar annotations in annotation.jar for FindBugs?

In the FindBugs distribution, annotations.jar is not a subset of jsr305.jar. However, several annotations seem to be duplicated (either exactly, or very closely). Should I prefer an annotation in jsr305.jar if I have a choice?
Note that I'm not just interested in knowing that it would be "better" to use annotations from jsr305.jar simply because they represent a standard. Rather, I want to know whether the FindBugs tool will perform the same (or better) analysis if I prefer the jsr305.jar version of a particular annotation. It could be the case that some jsr305.jar annotations should be preferred, but others should not.
I'm using FindBugs 1.3.9 which is the most recent version as of this writing. With this version, I see the following choices (please update this table if there are others):
edu.umd.cs.findbugs.annotations.CheckForNull → javax.annotation.CheckForNull
edu.umd.cs.findbugs.annotations.CheckReturnValue → javax.annotation.CheckReturnValue
edu.umd.cs.findbugs.annotations.NonNull → javax.annotation.Nonnull (NB capitalization)
edu.umd.cs.findbugs.annotations.Nullable → javax.annotation.Nullable
edu.umd.cs.findbugs.annotations.When → javax.annotation.meta.When
Also, all of the JCIP annotations are duplicated:
net.jcip.annotations.GuardedBy → javax.annotation.concurrent.GuardedBy
net.jcip.annotations.Immutable → javax.annotation.concurrent.Immutable
net.jcip.annotations.NotThreadSafe → javax.annotation.concurrent.NotThreadSafe
net.jcip.annotations.ThreadSafe → javax.annotation.concurrent.ThreadSafe
Yes, you should prefer the JSR305 annotation if possible. Why? Because the JSR305 is a standard and it makes total sense to stick to the standards where possible. While porting my own applications I did not notice any problems or changes in behavior. Moreover, you can even define your own #NotNull annotation and findbugs will pick it up (as long as you name it NotNull) see this blog entry for more details.
As far as I can see by looking at the sources, findbugs is using the same analysis methods internally. The dispatching is done only based on the annotations name. As mentioned in the blog post linked above, have a look at the edu.umd.cs.findbugs.ba.NullnessAnnotation and NullnessAnnotationDatabase classes to get an initial view how this is done internally. Have a look at this package and you'll find similar classes for the other annotations like the jcip ones.
So from an implementations point of view it really doesn't matter. For everybody still not sure which annotations to use I would consider using the standard annotations or self defined ones to avoid their code being dependent on the findbugs library.

Is it safe to use Project Lombok? [closed]

Closed. This question is opinion-based. It is not currently accepting answers.
Want to improve this question? Update the question so it can be answered with facts and citations by editing this post.
Closed 4 years ago.
Improve this question
In case you don't know Project Lombok helps with some of the annoyances of Java with stuff like generating getters and setters with annotations and even simple JavaBean like generation with #Data. It could really help me, especially in 50 different event objects where you have up to 7 different fields that need to be constructed and hidden with getters. I could remove almost a thousand lines of code with this.
However, I'm worried that in the long run, it will be a regretful decision. Flamewars will erupt in the ##Java Freenode channel when I mention it, providing code snippets will confuse possible helpers, people will complain about missing JavaDoc, and future commiters might just remove it all anyway. I would really enjoy the positive, but I'm worried about the negative.
So: Is it safe to use Lombok on any project, small or large? Are the positive effects worth the negatives?
TL; DR:
Yes, it's pretty safe to use and I'd recommend using it. (May 2022)
Original Answer
Just started using Lombok today. So far I like it, but one drawback I didn't see mentioned was refactoring support.
If you have a class annotated with #Data, it will generate the getters and setters for you based on the field names. If you use one of those getters in another class, then decide the field is poorly named, it will not find usages of those getters and setters and replace the old name with the new name.
I would imagine this would have to be done via an IDE plug-in and not via Lombok.
UPDATE (Jan 22 '13)
After using Lombok for 3 months, I still recommend it for most projects. I did, however, find another drawback that is similar to the one listed above.
If you have a class, say MyCompoundObject.java that has 2 members, both annotated with #Delegate, say myWidgets and myGadgets, when you call myCompoundObject.getThingies() from another class, it's impossible to know if it's delegating to the Widget or Gadget because you can no longer jump to source within the IDE.
Using the Eclipse "Generate Delegate Methods..." provides you with the same functionality, is just as quick and provides source jumping. The downside is it clutters your source with boilerplate code that take the focus off the important stuff.
UPDATE 2 (Feb 26 '13)
After 5 months, we're still using Lombok, but I have some other annoyances. The lack of a declared getter & setter can get annoying at times when you are trying to familiarize yourself with new code.
For example, if I see a method called getDynamicCols() but I don't know what it's about, I have some extra hurdles to jump to determine the purpose of this method. Some of the hurdles are Lombok, some are the lack of a Lombok smart plugin. Hurdles include:
Lack of JavaDocs. If I javadoc the field, I would hope the getter and setter would inherit that javadoc through the Lombok compilation step.
Jump to method definition jumps me to the class, but not the property that generated the getter. This is a plugin issue.
Obviously you are not able to set a breakpoint in a getter/setter unless you generate or code the method.
NOTE: This Reference Search is not an issue as I first thought it was. You do need to be using a perspective that enables the Outline view though. Not a problem for most developers. My problem was I am using Mylyn which was filtering my Outline view, so I didn't see the methods. Lack of References search. If I want to see who's calling getDynamicCols(args...), I have to generate or code the setter to be able to search for references.
UPDATE 3 (Mar 7 '13)
Learning to use the various ways of doing things in Eclipse I guess. You can actually set a conditional breakpoint (BP) on a Lombok generated method. Using the Outline view, you can right-click the method to Toggle Method Breakpoint. Then when you hit the BP, you can use the debugging Variables view to see what the generated method named the parameters (usually the same as the field name) and finally, use the Breakpoints view to right-click the BP and select Breakpoint Properties... to add a condition. Nice.
UPDATE 4 (Aug 16 '13)
Netbeans doesn't like it when you update your Lombok dependencies in your Maven pom. The project still compiles, but files get flagged for having compilation errors because it can't see the methods Lombok is creating. Clearing the Netbeans cache resolves the issue. Not sure if there is a "Clean Project" option like there is in Eclipse. Minor issue, but wanted to make it known.
UPDATE 5 (Jan 17 '14)
Lombok doesn't always play nice with Groovy, or at least the groovy-eclipse-compiler. You might have to downgrade your version of the compiler.
Maven Groovy and Java + Lombok
UPDATE 6 (Jun 26 '14)
A word of warning. Lombok is slightly addictive and if you work on a project where you can't use it for some reason, it will annoy the piss out of you. You may be better off just never using it at all.
UPDATE 7 (Jul 23 '14)
This is a bit of an interesting update because it directly addresses the safety of adopting Lombok that the OP asked about.
As of v1.14, the #Delegate annotation has been demoted to an Experimental status. The details are documented on their site (Lombok Delegate Docs).
The thing is, if you were using this feature, your backout options are limited. I see the options as:
Manually remove #Delegate annotations and generate/handcode the delegate code. This is a little harder if you were using attributes within the annotation.
Delombok the files that have the #Delegate annotation and maybe add back in the annotations that you do want.
Never update Lombok or maintain a fork (or live with using experiential features).
Delombok your entire project and stop using Lombok.
As far as I can tell, Delombok doesn't have an option to remove a subset of annotations; it's all or nothing at least for the context of a single file. I opened a ticket to request this feature with Delombok flags, but I wouldn't expect that in the near future.
UPDATE 8 (Oct 20 '14)
If it's an option for you, Groovy offers most of the same benefits of Lombok, plus a boat load of other features, including #Delegate. If you think you'll have a hard time selling the idea to the powers that be, take a look at the #CompileStatic or #TypeChecked annotation to see if that can help your cause. In fact, the primary focus of the Groovy 2.0 release was static safety.
UPDATE 9 (Sep 1 '15)
Lombok is still being actively maintained and enhanced, which bodes well to the safety level of adoption. The #Builder annotations is one of my favorite new features.
UPDATE 10 (Nov 17 '15)
This may not seem directly related to the OP's question, but worth sharing. If you're looking for tools to help you reduce the amount of boilerplate code you write, you can also check out Google Auto - in particular AutoValue. If you look at their slide deck, the list Lombok as a possible solution to the problem they are trying to solve. The cons they list for Lombok are:
The inserted code is invisible (you can't "see" the the methods it generates) [ed note - actually you can, but it just requires a decompiler]
The compiler hacks are non-standard and fragile
"In our view, your code is no longer really Java"
I'm not sure how much I agree with their evaluation. And given the cons of AutoValue that are documented in the slides, I'll be sticking with Lombok (if Groovy is not an option).
UPDATE 11 (Feb 8 '16)
I found out Spring Roo has some similar annotations. I was a little surprised to find out Roo is still a thing and finding documentation for the annotations is a bit rough. Removal also doesn't look as easy as de-lombok. Lombok seems like the safer choice.
UPDATE 12 (Feb 17 '16)
While trying to come up with justifications for why it's safe to bring in Lombok for the project I'm currently working on, I found a piece of gold that was added with v1.14 - The Configuration System! This is means you can configure a project to dis-allow certain features that your team deems unsafe or undesirable. Better yet, it can also create directory specific config with different settings. This is AWESOME.
UPDATE 13 (Oct 4 '16)
If this kind of thing matters to you, Oliver Gierke felt it was safe to add Lombok to Spring Data Rest.
UPDATE 14 (Sep 26 '17)
As pointed out by #gavenkoa in the comments on the OPs question, JDK9 compiler support isn't yet available (Issue #985). It also sounds like it's not going to be an easy fix for the Lombok team to get around.
UPDATE 15 (Mar 26 '18)
The Lombok changelog indicates as of v1.16.20 "Compiling lombok on JDK1.9 is now possible" even though #985 is still open.
Changes to accommodate JDK9, however, necessitated some breaking changes; all isolated to changes in config defaults. It's a little concerning that they introduced breaking changes, but the version only bumped the "Incremental" version number (going from v1.16.18 to v1.16.20). Since this post was about the safety, if you had a yarn/npm like build system that automatically upgraded to the latest incremental version, you might be in for a rude awakening.
UPDATE 16 (Jan 9 '19)
It seems the JDK9 issues have been resolved and Lombok works with JDK10, and even JDK11 as far as I can tell.
One thing I noticed though that was concerning from a safety aspect is the fact that the change log going from v1.18.2 to v1.18.4 lists two items as BREAKING CHANGE!? I'm not sure how a breaking change happens in a semver "patch" update. Could be an issue if you use a tool that auto-updates patch versions.
UPDATE 17 (Mar 17 '21)
There is some drama unfolding between the Lombok developers and an OpenJDK developer around JDK 16. The JDK developers argue that Lombok is taking advantage of unpublished JDK internals via loopholes the JDK team would like to close, but have intentionally left open for various reasons.
They have stated their concern (about the safety of Lombok) as such:
All access to internals will remain available as before, provided that
the client application explicitly allows it, acknowledging that it is
knowingly taking on any maintenance (or security) issue this might
entail.
While Lombok might think they're deceiving OpenJDK, all they're doing
is announcing that it is their intention to deceive their own users.
There may come a day soon where Lombok will not be able to find any more creative solutions around the JDK's security restrictions. Even if they do, the safety of using Lombok in your project may be in question.
UPDATE 18 (May 11 '22)
A recent comment asked for a summary, so I put that at the top.
The short answer is it's perfectly safe to use and I'd highly recommend using it if we're writing Java code.
Given that support for JDK 17 has been out for awhile and was released less than a month after the JDK was officially released, the safety of Lombok sticking around is high. And you can always de-lombok if you need to.
As a consultant, I get to see how a lot of different companies write code. Every client I've had for the past 5 years has used Lombok. These have all been Fortune 1000 companies. It speeds development and makes it less error prone.
That said, you still need to keep up on the latest features of the JDK. Consider using the Java record keyword to make your objects immutable instead of some Lombok feature. Use Lombok where it makes sense. Use the Lombok config options to prevent usage of it in ways you don't agree with.
So unless something major happens, this will probably be my last update to this answer. Thanks for all the votes. I'm glad it helps.
It sounds like you've already decided that Project Lombok gives you significant technical advantages for your proposed new project. (To be clear from the start, I have no particular views on Project Lombok, one way or the other.)
Before you use Project Lombok (or any other game-changing technology) in some project (open source or other wise), you need to make sure that the project stake holders agree to this. This includes the developers, and any important users (e.g. formal or informal sponsors).
You mention these potential issues:
Flamewars will erupt in the ##Java Freenode channel when I mention it,
Easy. Ignore / don't participate in the flamewars, or simply refrain from mentioning Lombok.
providing code snippets will confuse possible helpers,
If the project strategy is to use Lombok, then the possible helpers will need to get used to it.
people will complain about missing JavaDoc,
That is their problem. Nobody in their right mind tries to rigidly apply their organization's source code / documentation rules to third-party open source software. The project team should be free to set project source code / documentation standards that are appropriate to the technology being used.
(FOLLOWUP - The Lombok developers recognize that not generating javadoc comments for synthesized getter and setter methods is an issue. If this is a major problem for your project(s), then one alternative is to create and submit a Lombok patch to address this.)
and future commiters might just remove it all anyway.
That's not on! If the agreed project strategy is to use Lombok, then commiters who gratuitously de-Lombok the code should be chastised, and if necessary have their commit rights withdrawn.
Of course, this assumes that you've got buy-in from the stakeholders ... including the developers. And it assumes that you are prepared to argue your cause, and appropriately handle the inevitable dissenting voices.
Go ahead and use Lombok, you can if necessary "delombok" your code afterwards http://projectlombok.org/features/delombok.html
Personally (and therefore subjectively) I've found that using Lombok makes my code more expressive about what I'm trying to achieve when compared to IDE/own implementations of intricate methods such as hashcode & equals.
When using
#EqualsAndHashCode(callSuper = false, of = { "field1", "field2", "field3" })
it's much easier to keep Equals & HashCode consistent and keep track of which fields are evaluated, than any IDE/own implementation. This is especially true when you're still adding / removing fields regularly.
The same goes for the #ToString annotation and its parameters which clearly communicate the desired behavior regarding included / excluded fields, usage of getters or field access and wether or not to call super.toString().
And again by annotating an entire class with #Getter or #Setter(AccessLevel.NONE) (and optionally overriding any divergent methods) it's immediately clear what methods will be available for the fields.
The benefits go on and on..
In my mind it's not about reducing code, but about clearly communicating what you desire to achieve, rather than having to figure it out from Javadoc or implementations. The reduced code just makes it easier to spot any divergent-method implementations.
I read some opinions about the Lombok and actually I'm using it in some projects.
Well, in the first contact with Lombok I had a bad impression. After some weeks, I started to like it. But after some months I figure out a lot of tiny problems using it. So, my final impression about Lombok is not so positive.
My reasons to think in this way:
IDE plugin dependency. The IDE support for Lombok is through plugins. Even working good in most part of the time, you are always a hostage from this plugins to be maintained in the future releases of the IDEs and even the language version (Java 10+ will accelerate the development of the language). For example, I tried to update from Intellij IDEA 2017.3 to 2018.1 and I couldn't do that because there was some problem on the actual lombok plugin version and I needed to wait the plugin be updated... This also is a problem if you would like to use a more alternative IDE that don't have any Lombok plugin support.
'Find usages' problem.. Using Lombok you don't see the generated getter, setter, constructor, builder methods and etc. So, if you are planning to find out where these methods are being used in your project by your IDE, you can't do this only looking for the class that owns this hidden methods.
So easy that the developers don't care to break the encapsulation. I know that it's not really a problem from Lombok. But I saw a bigger tendency from the developers to not control anymore what methods needs to be visible or not. So, many times they are just copying and pasting #Getter #Setter #Builder #AllArgsConstructor #NoArgsConstructor annotations block without thinking what methods the class really need to be exposed.
Builder Obssession ©. I invented this name (get off, Martin Fowler). Jokes apart, a Builder is so easy to create that even when a class have only two parameters the developers prefer to use #Builder instead of constructor or a static constructor method. Sometimes they even try to create a Builder inside the lombok Builder, creating weird situations like MyClass.builder().name("Name").build().create().
Barriers when refactoring. If you are using, for example, a #AllArgsConstructor and need to add one more parameter on the constructor, the IDE can't help you to add this extra parameter in all places (mostly, tests) that are instantiating the class.
Mixing Lombok with concrete methods. You can't use Lombok in all scenarios to create a getter/setter/etc. So, you will see these two approaches mixed in your code. You get used to this after some time, but feels like a hack on the language.
Like another answer said, if you are angry about the Java verbosity and use Lombok to deal with it, try Kotlin.
Lombok is great, but...
Lombok breaks the rules of annotation processing, in that it doesn't generate new source files. This means it cant be used with another annotation processors if they expect the getters/setters or whatever else to exist.
Annotation processing runs in a series of rounds. In each round, each one gets a turn to run. If any new java files are found after the round is completed, another round begins. In this way, the order of annotation processors doesn't matter if they only generate new files. Since lombok doesn't generate any new files, no new rounds are started so some AP that relies on lombok code don't run as expected. This was a huge source of pain for me while using mapstruct, and delombok-ing isn't a useful option since it destroys your line numbers in logs.
I eventually hacked a build script to work with both lombok and mapstruct. But I want to drop lombok due to how hacky it is -- in this project at least. I use lombok all the time in other stuff.
Update to specifically mapstruct+lombok: the two libraries work out of the box with each other nowadays. The problem will still exist for other annotation processors though.
There are long-term maintenance risks as well. First, I'd recommend reading about how Lombok actually works, e.g. some answers from its developers here.
The official site also contains a list of downsides, including this quote from Reinier Zwitserloot:
It's a total hack. Using non-public API. Presumptuous casting (knowing
that an annotation processor running in javac will get an instance of
JavacAnnotationProcessor, which is the internal implementation of
AnnotationProcessor (an interface), which so happens to have a couple
of extra methods that are used to get at the live AST).
On eclipse, it's arguably worse (and yet more robust) - a java agent
is used to inject code into the eclipse grammar and parser class,
which is of course entirely non-public API and totally off limits.
If you could do what lombok does with standard API, I would have done
it that way, but you can't. Still, for what its worth, I developed the
eclipse plugin for eclipse v3.5 running on java 1.6, and without
making any changes it worked on eclipse v3.4 running on java 1.5 as
well, so it's not completely fragile.
As a summary, while Lombok may save you some development time, if there is a non-backwards compatible javac update (e.g. a vulnerability mitigation) Lombok might get you stuck with an old version of Java while the developers scramble to update their usage of those internal APIs. Whether this is a serious risk obviously depends on the project.
I know I'm late, but I can't resist the temptation: anybody liking Lombok should also have a look at Scala. Many good ideas that you find in Lombok are part of the Scala language.
On your question: it's definitely easier to get your developers trying Lombok than Scala. Give it a try and if they like it, try Scala.
Just as a disclaimer: I like Java, too!
I have used Lombok in almost all my projects for one year but unfortunately removed it. In the beginning it was a very clean way of development but setting up the development environment for new team members is not very easy and straightforward. When it became a headache I just removed it. But it is a good work and needs some more simplicity to setting up.
My take on Lombok is that it merely provides shortcuts for writing bolilerplate Java code.
When it comes to using shortcuts for writing bolilerplate Java code, I would rely on such features provided by IDE -- like in Eclipse, we can go to menu Source > Generate Getters and Setters for generating getters and setters.
I would not rely on a library like Lombok for this:
It pollutes your code with an indirection layer of alternative syntax (read #Getter, #Setter, etc. annotations). Rather than learning an alternative syntax for Java, I would switch to any other language that natively provides Lombok like syntax.
Lombok requires the use of a Lombok supported IDE to work with your code. This dependency introduces a considerable risk for any non-trivial project. Does the open source Lombok project have enough resources to keep providing support for different versions of a wide range of Java IDE's available?
Does the open source Lombok project have enough resources to keep providing support for newer versions of Java that will be coming in future?
I also feel nervous that Lombok may introduce compatibility issues with widely used frameworks/libraries (like Spring, Hibernate, Jackson, JUnit, Mockito) that work with your byte code at runtime.
All in all I would not prefer to "spice up" my Java with Lombok.
When I showed the project to my team the enthusiasm was high, so I think you should not be afraid of team response.
As far as ROI, it is a snap to integrate, and requires no code change in its basic form. (just adding a single annotation to your class)
And last, if you change your mind, you can run the unlombok, or let your IDE create these setters, getters, and ctors, (which I think no one will ask for once they see how clear your pojo becomes)
Wanted to use lombok's #ToString but soon faced random compile errors on project rebuild in Intellij IDEA. Had to hit compile several times before incremental compilation could complete with success.
Tried both lombok 1.12.2 and 0.9.3 with Intellij IDEA 12.1.6 and 13.0 without any lombok plugin under jdk 1.6.0_39 and 1.6.0_45.
Had to manually copy generated methods from delomboked source and put lombok on hold until better times.
Update
The problem happens only with parallel compile enabled.
Filed an issue:
https://github.com/rzwitserloot/lombok/issues/648
Update
mplushnikov commented on 30 Jan 2016:
Newer version of Intellij
doesn't have such issues anymore. I think it can be closed here.
Update
I would highly recommend to switch from Java+Lombok to Kotlin if possible.
As it has resolved from the ground up all Java issues that Lombok tries to work around.
I have encountered a problem with Lombok and Jackson CSV, when I marshalized my object (java bean) to a CSV file, columns where duplicated, then I removed Lombok's #Data annotation and marshalizing worked fine.
I haven't tried using Lombok yet - it is/was next on my list, but it sounds as if Java 8 has caused significant problems for it, and remedial work was still in progress as of a week ago. My source for that is https://code.google.com/p/projectlombok/issues/detail?id=451 .

Maintainability of Java annotations?

My project is slowly implementing Java annotations. Half of the developers - myself included - find that doing anything complex with annotations seems to add to our overall maintenance burden. The other half of the team thinks they're the bee's knees.
What's your real-world experience with teams of developers being able to maintain annotated code?
My personal experience is that, on average, dealing with annotations is far easier for most developers than dealing with your standard Java XML Configuration hell. For things like JPA and Spring testing they are absolute life-savers.
The good thing about annotations is that they make configuration on your classes self-documenting. Now, instead of having to search through a huge XML file to try and figure out how a framework is using your class, your class tells you.
Usually the issue with changes like this is that getting used to them simply takes time. Most people, including developers, resist change. I remember when I started working with Spring. For the first few weeks I wondered why anyone would put up with the headaches associated with it. Then, a few weeks later, I wondered how I'd ever lived without it.
I feel it breaks into two uses of annotations - annotations to provide a 'description' of a class vs. annotations to provide a 'dependency' of the class.
I'm fine with a 'description' use of annotations on the class - that's something that belongs on the class and the annotation helps to make a shorthand version of that - JPA annotations fall under this.
However, I don't really like the 'dependency' annotations - if you're putting the dependency directly on the class - even if it's determined at runtime from an annotation rather than at compile time in the class - isn't that breaking dependency injection? (perhaps in spirit rather than in rule...)
It may be personal preference, but I like the one big XML file that contains all the dependency information of my application - I view this as 'application configuration' rather than 'class configuration'. I'd rather search through the one known location than searching through all the classes in the app.
It depends highly on IDE support. I feel that annotations should be kept in sync with the code via checks in the IDE, but that support for this is somewhat lacking.
E.g. the older version of IDEA would warn if you overrode a function without #Override, but wouldn't remove the #Override tag if you changed the method signature (or the superclass signature, for that matter) and broke the relation.
Without support I find them a cumbersome way to add metadata to code.
I absolutely love annotations. I use them from Hibernate/JPA, Seam, JAXB....anything that I can. IMO there's nothing worse than having to open up an XML file just to find out how a class is handled.
To my eye annotations allow a class to speak for itself. Also annotations are (hopefully) part of your IDEs content assist, whereas with XML config you are usually on your own.
However, it may come down to how the XML configs and Annotations are actually used by any particular library (as most offer both), and what sort of annotation is used. I can imagine that annotations that define something that is build-specific (eg. file/url paths) may actually be easier as XML config.
i personally feel that the the specific use case you mentioned (auto-generate web forms) is a great use case for annotations. any sort of "framework" scenario where you can write simplified code and let the framework do the heavy (often repetitive) lifting based on a few suggestions (aka annotations) is, i think, the ideal use case for annotations.
i'm curious why you don't like annotations in this situation, and what you consider to be the "maintenance burden"? (and, i'm not trying to insult your position, just understand it).

Cross-class-capable extendable static analysis tool for java?

I'm trying to write rules for detecting some errors in annotated multi-threaded java programs. As a toy example, I'd like to detect if any method annotated with #ThreadSafe calls a method without such an annotation, without synchronization. I'm looking for a tool that would allow me to write such a test.
I've looked at source analyzers, like CheckStyle and PMD, and they don't really have cross-class analysis capabilities. Bytecode analysers, like FindBugs and JLint seem rather difficult to extend.
I'd settle for a solution to something even simpler, but posing the same difficulty: writing a custom rule that checks whether each overriden method is annotated with #Override.
Have you tried FindBugs? It actually supports a set of annotations for thread safety (the same as those used in Java Concurrency in Practice). Also, you can write your own custom rules. I'm not sure whether you can do cross-class analysis, but I believe so.
Peter Ventjeer has a concurrency checking tool (that uses ASM) to detect stuff like this. I'm not sure if he's released it publicly but he might able to help you.
And I believe Coverity's static/dynamic analysis tools for thread safety do checking like this.
You can do cross-class analysis in PMD (though I've never used it for this specific purpose). I think it's possible using this visitor pattern that they document, though I'll leave the specifics to you.
A simple tool to checkup on annotations is apt (http://java.sun.com/j2se/1.5.0/docs/guide/apt/ also part of Java 6 api in javax.annotation.processing) however this only has type information (ie I couldn't find a quick way to get at the inheritance hierarchy using the javax.lang.model api, however if you can load the class you can get that information using reflection).
Try javap + regexes (eg. Perl)

Categories