Talk:Use Modular Architecture

From APIDesign

---

Current revision

Done: 74d64dca0394

The example of the JDK as non-modular may be out of date by the time you go to print! :)

• I am afraid we publish few years sooner

I'd like some discussion of how you modularize a project that already has a large, non-modular, full-of-spaghetti code base. I think this would resonate with a lot of programmers.

• I may blog about it or describe that in wiki, but for the book it would be maybe too detailed. Just remind me.

Also, I'd like you to address non-runtime modularization. That is, just enforcing module dependencies at compile time, in the ant script. It's perfectly valid, requires no additional software like OSGI, but has some pitfalls (Class.forName, etc). For some organizations that are mistrustful of runtime modularization's performance impact (warranted or not), it's better than nothing.

• I have contributed description of Ant Javac task parameters, search for sourcepath=""

--RichUnger 22:40, 10 April 2008 (UTC)

1. Page 99: "Modular programming is here to make it happen." This implies that modular programming encourages the degradation of the architecture rather than preventing it, so you probably meant "Modular programming is here to help prevent it."

--TomWheeler Sat Apr 12 19:13:37 CDT 2008

Done: 134c6f9cad88

Page 103, para 1: "using some sort of Hashtable" -- DriverManager uses Vector's (but sure, you can view it as a dictionary with O(N) complexity). The registration is static, but there is a deregisterDriver method (sad, but it is only today that I learned about it). ConnectionPool should probably be DataSource. "J2EE" should probably be "Java EE".

--AndreiBadea 09:59, 7 April 2008 (UTC)

I'm not sure what "missing method exceptions" are. There are various kinds of LinkageError, but these are errors, not exceptions.

URLHandlerFactory: you mean URLStreamHandlerFactory.

Crimson is so obsolete no one will know what you are talking about. On the other hand, I cannot actually find any implementations of DOM in Java other than Xerces-J.

"I hear that..." is not a confidence-inspiring statement in a book pretending to authority. Research it, and if true, state it definitively.

--JesseGlick 01:36, 8 April 2008 (UTC)

Page 102: "There is a need for some sort of 'indirect dependency' declaration..." Who needs this and why? An example here would help explain the need in less theoretical terms.

• you need to read the whole sentence, it says to ...

--TomWheeler Sat Apr 12 19:13:37 CDT 2008

Done: dd09dc43619c - I am currently rewriting this part to reflect your comments. As the topic is quite alive, I have also created a dedicated online page for this called Injection.

Page 104, para -1: "The you hope that the framework will" sounds a bit emotional to me for a technical writing. "In Spring you define setters" -- you are not restricted to setters, you can also use constructor parameters.

Page 107, para 2: "As the JDK already contains the ServiceLoader API": isn't "Java platform" more appropriate than "JDK"?

LookupListener example: the parameter of resultChanged should have a sane name, not arg0, even if it is not used.

--AndreiBadea 10:10, 7 April 2008 (UTC)

This section seems to me to diverge from the goal of talking about how to write APIs, to cover the details of one (relatively unpopular) component assembly framework. I would instead expect a general discussion of the purpose of such frameworks, followed by a list of the major ones (ServiceLoader, Lookup, Spring, Spring with annotations, the various OSGi service discovery APIs, Eclipse extension points), with a brief explanation of the advantages of each. The defensive tone of the sidebars is also unbecoming; this is not the right place to defend your personal architecture from critics.

--JesseGlick 01:44, 8 April 2008 (UTC)

p.38: meta-inf/services/foo.Service isn't a configuration file?

--RichUnger 22:40, 10 April 2008 (UTC)

Page 103: "However, we are not unaware" - Avoid the double negative. Page 108: Again, I do not understand the image in Figure 8-1 nor its relevance of teleinterfaces. I don't think many readers would be able to explain the concept of a teleinterface at this point in the book.

--TomWheeler Sat Apr 12 19:13:37 CDT 2008

I find much of the text in this section to be too NetBeans-specific. I agree with Jesse that it might make sense to talk a bit less about NetBeans details and a bit more about other popular component frameworks.

--AdamDingle 00:43, 13 April 2008 (UTC)

Done: bc774fd94cce

TipOfTheDay example: sayHello doesn't seem a good name for the method. How about getTip?

Page 109, para -2,-4: extra "s" in TipsOfTheDay.

Page 109, para -2: "TipsOfTheDayProvider" should probably be "TipOfTheDay" provider.

Page 110, para 4: extra "s" in TipsOfTheDay.

--AndreiBadea 11:39, 7 April 2008 (UTC)

"Each module that would like to register its TipsOfTheDayProvider needs to have to core module on classpath" - I don't understand this sentence. There's either a typo or a grammar mistake here, but I'm confused enough that I don't know what you're saying, so I can't suggest a fix.

--RichUnger 22:40, 10 April 2008 (UTC)

Done: b058ea24096e

Page 113, code example: encode should be encrypt. And probably it should return a byte[]. OK, you claim the example is a bit artificial, so consider this a nit pick.

Page 112: image uses Encoder instead of Encryptor.

Page 112, code example: why do you use a plain assert in the encrypt method? In production builds, when assertions are turned off, this will fail with a meaningless NullPointerException.

--AndreiBadea 13:06, 8 April 2008 (UTC)

Your discussion seems to be specific to NetBeans. Are you suggesting that every component system should disallow cyclic dependencies? If so, you should say that explicitly. Do other component systems (e.g. Spring?) allow cycles between components?

--AdamDingle 00:46, 13 April 2008 (UTC)

Done: d0b6efbc74d8

Page 115: only seasoned NetBeans developers will know what ic in ic.add(new ModifiedImpl(this)) stands for. I think you can just drop the example.

Page 116: code example: CatQueryImplementation should probably be an interface. You claimed in a previous section that interfaces are to be preferred to abstract classes. Moreover, see Jesse's comment in Are Abstract Classes Useful? about protected methods in abstract classes.

I would also argue that the example is too artificial. FileEncodingQuery would pehaps be a better example.

--AndreiBadea 13:20, 8 April 2008 (UTC)