Visualizing the .NET Framework

eldavojohn writes "If you're a Web developer, you should check out a quick post about the number of types, methods, & fields in the .NET framework. This was done using NDepend. The numbers are quite large — e.g. 39,509 types. The blogger went on to generate tree maps and a dependency matrix."

Wow, that's a big fat ASS^H^HPI

Seeing that I have no personal experience with .Net, and seeing that this is Slashdot, I feel totally qualified to poke fun at its stupendous complexity with a quote:

Any third-rate engineer or researcher can increase complexity; but it takes a certain flair of real insight to make things simple. -E. F. Schumacher

Re:Wow, that's a big fat ASS^H^HPI

Any third-rate engineer or researcher can increase complexity; but it takes a certain flair of real insight to make things simple.
-E. F. Schumacher

Rube Goldberg is alive and working for Microsoft.

Re:Wow, that's a big fat ASS^H^HPI

I'm not saying that .NET isn't too complex, but having a large number of types does not necessarily increase complexity. In fact having less types often leads to more complexity.

Creating a new type abstracts away complexity and makes code easier to read. For example, you will often find that business software does a lot of comparing of dates. e.g. Checking whether a date is within a given range. More often than not you will find that programmers have just written things like...

If (EnteredDate >= StartDate) and (EnteredDate <= EndDate)
{
//Do stuff
}
else
{
//Tell the user they have entered an invalid date.
}

The logic of checking that dates are in a range is repeated all over the place and the more you have to type, the more likely it is that you will make mistakes. This is where adding a new type, a "Range" type will help. With a Range type you can just say something like...

If (AllowedRange.Contains(EnteredDate))
//blah blah

So adding a Range class actually reduces complexity rather than increasing it. There are plenty of examples of this sort of thing. Imagine writing a car ordering system without having a Car class to abstract away details about cars. You could do it, but the code would probably be a lot more sprawling and complex.

Re:Wow, that's a big fat ASS^H^HPI

It is this level of function (mis)use that makes me cringe.

So instead of being able to see both the variable AND the range it is being tested against IN THE SAME LINE, I now have to go trawling back through the code looking for the place where you created the Range object to find the low and high boundaries of it.

So yet more jumping all over the place hunting for stuff, when the original version was completely fit for purpose, clear, and most importantly, IN ONE BLOODY PLACE.

Of course, it get's even "better" ... not all range checking will use the same ranges ... so then some bright spark will create Range2, Range3, Range4 objects with different ranges in each one. You see how this function does nothing for either readability or speed of debugging, but simply hides information that a programmer NEEDS to know in the context of the line he is looking at ?

You can keep your Range object thanks.

Re:Wow, that's a big fat ASS^H^HPI

So instead of being able to see both the variable AND the range it is being tested against IN THE SAME LINE, I now have to go trawling back through the code looking for the place where you created the Range object to find the low and high boundaries of it.

You seem to be assuming that there would be a hard coded range. The allowable range may be defined in a database or elsewhere. Checking against the same range may occur in many different places, so you certainly would not want to have the range hard-coded in every routine you need to do such checking.

Imagine that the Date range object was intended to check the date of birth of new employees (e.g. You want prevent mistakes like they were born 200 years ago or 50 years in the future). If you are smart you will have created some kind of Employee class, and this Date Range checking object could just be a static variable of the class itself. It would be pretty easy to see where it was set.

So yet more jumping all over the place hunting for stuff, when the original version was completely fit for purpose, clear, and most importantly, IN ONE BLOODY PLACE.

The whole point is to reduce the unnecessary repetition of logic. Imagine if you wanted to do something more complex like check if one date range was contained within another. Suddenly you start repeating quite a lot of logic without a Range object.

Of course, it get's even "better" ... not all range checking will use the same ranges ... so then some bright spark will create Range2, Range3, Range4 objects with different ranges in each one.

Of course there will be different ranges. What does that have to do with anything? If anyone names variables "Range1, Range2" etc, they need some quick re-education.

You see how this function does nothing for either readability or speed of debugging, but simply hides information that a programmer NEEDS to know in the context of the line he is looking at ?

It's interesting that you think information is being hidden. This would only be the case if you compare it to the situation of hard coding things everywhere, which is generally a very bad practice. The information in a Range object is no more hidden that the case where your limits are two separate variables called "StartDate" and "EndDate" (Variables which might be initialized when the application first starts). What is really being hidden is logic. That's what object oriented programming is really all about, i.e. trying to abstract away complexity into new types.

You can keep your Range object thanks.

It's not really mine. Martin Fowler wrote about it in Analysis Patterns, although I'm quite sure it was being used long before it occurred to him.

Re:Wow, that's a big fat ASS^H^HPI

Yeah, let's compare .Net to PHP. .Net has a very extensive API. PHP also has a very extensive framework. The .Net framework was very well thought out and is very well organized. The PHP framework is cobbled together piece-meal, with everything in the same namespace (Yes, I'm aware that it still doesn't have namespaces, and won't until PHP6 is out, but that's yet another disadvantage of PHP), and separate functions for each database they support, where all are very similarly named, but not exactly the same.

Re:Wow, that's a big fat ASS^H^HPI

Monorails are efficient and sparkly shiny but are almost universally inappropriate except for certain very limited transportation scenarios (airport people-movers, theme parks, etc). They require alot of service and very expensive infrastructure. The old honda just needs a road, works on your schedule and is comparatively inexpensive to maintain on a per capita basis.

Just sayin.

Re:Wow, that's a big fat ASS^H^HPI

And a stripped-down non-existent API is a way to make things simple?

Nope. Einstein said it best, I think:

"Everything should be made as simple as possible, but not simpler."

It's a given that new features will increase complexity. (Which, surprisingly, is not always true. But stick with me for a moment.) The key is to increase the complexity only as much as necessary. If you increase the complexity any farther, you have failed.

I can't count how many times I have seen code that is overly complex. e.g. Someone piled up layer upon redundant layer of code, hoping to get simplicity out of each one. Instead, they create more maintenance, more points of failure, and more bugs. Probably one of the most egregious examples was a company that wanted me to use a go-between piece of software on another server to make a real-time request for XML. They had a lot of weak excuses for this decision, not one of which held water. After peeling back the layers of nonsense, I found out that the reason why they wanted it used is because they had already sunk $10,000 into it.

While not all examples I've seen are motivated by money or CYA, I have certainly seen a lot of examples that were motivated by blind adherence to company standards or existing technologies. Never mind that this particular module doesn't need any of the features of Struts, the company policy says use it, so we use it. Never mind that ADO isn't a good API. We have it, so we should wrap it. Never mind that we could just plug in a lib to do the secure transfer directly, we need to Rube Goldberg it through 5 different machines, protocols, and scripts written in 7 different languages!

There's a reason why engineers harp on the KISS principle. KISS is all about engineering a solution that will last for a long time to come. A solution that can be understood by others, maintained, is reliable, and exceeds the specs wherever reasonable. As Einstein said, make it as simple as possible and no simpler.

Re:The purpose of this complexity

I realize that you're one of those "Anything but Microsoft at any cost" people, but I have some news for you.

1) You don't have to memorize what is in every namespace in *any* language. You just have to have a good enough overview of them to know where to look effectively for something you need. Spending all of your time trying to memorize everything in libraries is a waste of time and usually only done by undergads who think that they know it all.

2) These are just toys to keep you occupied
I hate to break it to you, but there is a great deal of business that is done on Microsoft software. There are a number of reasons for that (admittedly, not all of them are good reasons). One of them is that MS has done something Linux and the others haven't - they make software that makes it fairly simple for most businesses and people to do 80% of what they need to easily. In addition, needs which are not met by MS itself, and a number of needs which are, are covered by various 3rd parties.

As for "toys to keep you occupied", I'm one heck of a lot more productive using C# in Visual Studio for most of the things that I have to do than I was using C or C++ in xemacs. C and C++ are fine tools for doing things closer to the metal. Most applications don't need that.

In addition, dealing with deprecation in .NET isn't really any worse than when I was dealing with it in C, C++, or any other language that I've used.

You need to learn to use the right tools for the job. Instead, you want to bash tools on an idealogical basis. It's a sign that you really need to grow up.

Re:Wow, that's a big fat ASS^H^HPI

Have a chip on your shoulder much? Most of what you're saying is simply incorrect. e.g. Java does not have half-a-dozen containers. Yes, the switch from the STL-inspired Vector to the more Java-ish ArrayList was annoying. Same with HashTable to HashMap. But beyond that, all those different containers you think you see are actually interfaces for wiring up complex functionality. Either that or completely different data structures with different performance characteristics. (Remember your CompSci courses?) The Java Collections package (which seems to be the only thing in Java you're remotely familiar with) provides enough functionality to write a complete database engine. Which, as a matter of fact, has [sourceforge.net] been [apache.org] done [sleepycat.com] quite [tigris.org] a few [mckoi.com] times [db4objects.com]. (Sorry [ozone-db.org], ran [objectweb.org] out [quadcap.com] of words [jepstone.net] to link [sourceforge.net]. Doh! [h2database.com] Still [mcobject.com] more [daffodildb.com]. Ah [smallsql.de], to hell [myoodb.org] with [metanotion.net] it [java-source.net].)

The rest of the Java API is also not bloat. There are libraries for printing, crytography, sound, graphics, DOM, file I/O, text parsing, text formatting, text display, mathematics, directory interfaces (e.g. LDAP), distributed object systems, reflection, security, SQL database interface, logging, cross-platform preferences, regular expressions, ZIP/GZip support, accessibility, networking, the compiler, scripting engines, etc., etc., etc. Very little of the core API is redundant, with most of the (few!) redundancies being a result of the early days of Java before they moved away from the C++ style objects.

Nearly all of the post-1.0 APIs were done correctly the first time. Which means that the core Java API is actually quite slim for the amount of functionality it provides. And even then, there is a HUGE number of official expansion APIs for mail, multimedia codecs, network request/response handlers (e.g. servlets), 3D graphics, 3D sound, text-to-speech, speech recognition, telephony, SOAP, REST, USB, Bluetooth, scientific units, cross-platform desktop integration, Instant Messaging, P2P, and quite a bit more. And that's just the official JSR-approved expansions! The OSS and (bleh) commercial worlds are full of unofficial libraries to deal with nearly any problem you can come up with.

If you want bloat, stop looking at Java. Try compiling a few Linux apps sometime and tell me how many redundant libraries you come across. If you know what they all do (which is a miracle in of itself), compiling just ONE of those programs is enough to make a person blush with embarrassment. Not to mention that when a platform IS solidified (e.g. GNOME), it suffers from versionitis. (i.e. The constant need to upgrade your version of the libraries because this latest program no longer targets the version you just compiled. Or even worse, it requires a specific minor release, thus requiring you to have multiple minor releases of the library compiled and installed.) I won't even go into Microsoft's practice of inventing a new API for the same technology over, and over, and over again. (ODBC, DAO, ADO, JET, anyone?)

Now I happen to think that a lot of the choice that Linux offers is good. But don't point fingers at other platforms when there are more than enough examples of far worse situations close to home.

If more types is bad, then Java is equally bad.

Before you impugn Microsoft with your short-sided emotional appeals against the sheer number of classes, use a hint of logic and realize that since MS copied Java shamelessly and ruthlessly (improving on some debacles in the Java classes, such as the crap IO classes that had to be redone from scratch), you'd be blasting Java, KDE, Python, and most any other class library as well.

Look, in a class library that purports to help most everyone, there's going to be an awful lot of code. Class library implies that classes are used to organize the abstractions provided by the library. Proper OO design favors designing more types with smaller number of features rather than God-objects that do many things. Fine-grained objects are simpler to unit test and are much easier to reuse. The downside is the propagation of types and the verbosity level of the code generally goes up. But that is a fair trade-off in my opinion, since the most important work on the code happens in the maintenance phase, when someone else can come along and at least get a vague idea of what is going on.

I've used the class libraries in Java, KDE, MFC, and Python, and the .NET class library beats them all easily. It is obvious that the designers did their homework and stole from each library what worked well, while dropping what didn't. If they were smart they'd take Java's excellent concurrency constructs such as the BlockingQueue and put them in (they may already have, I don't program much in .NET lately). Most of my beefs with the class library are the fact that it is huge (footprint size), and I don't agree with some of the modeling. But that is minor.

There's a reason Miguel wanted to make this happen on Linux. It is close to making programming fun again, instead of squinting at hyper-abbreviated function names like sprintf and mucking around with idiotic string representations such as C's.

Correction

If you're a Web developer, you should ignore .NET and use something much less bloated.

There, fixed that for you.

Compare it to the Human Genome

The comparison to the human genome is interesting. The genome contains about 3 billion base pairs and 30,000 coding genes. As best I can see, .NET is quite a bit bigger: The closest thing to a gene is a method (an object that can be used, or not used, and which does something). The genome has 30,000 and .NET has 384,000. So even if it takes 10 methods to do what one gene does, they are equivalent.

It takes 3 base pairs to code for a single protein, perhaps the closest we can come to an instruction. Each gene has an average of 3,000 base pairs, equivalent to 1,000 instructions. So we are looking at 30,000 genes x 1,000 instructions/gene or about 30,000,000 instructions in the genome. .NET has 8,000,000 instructions. Given the roughness of the comparison, this is pretty close.

The point here is that we are creating programs that are roughly equal in complexity to the human genome. If we were better programers, then perhaps we'd have come up with intelligent design.

Finally, it's worth noting that the functions are unknown for over 50% of discovered genes. It may be about the same for .NET :))

Re:Compare it to the Human Genome

The difference between genetic code and any human-invented language is the following: have you ever tried to debug a genome? It's friggin' ridiculous. No documentation at all.

-G

Re:.NET is OOP gone stupid.

Although they supposedly give more flexibility, something as essential as reading from and writing to a file becomes a hassle with .NET or Java. It's easy to get lost in whether we need a FileInputStream, or whether we should wrap a FileInputReader in a TextInputBuffer, and so forth. Give me fopen() any day.

The first time I saw what you (supposedly) have to do to read from a file in Java, it pegged my OMGWTF meter. I'm sure there's totally valid reasons for making such simple (and common) tasks so complicated, but apparently I'm not smart enough to understand them. IMHO it's one thing to have the complexity available if it's needed, but it's another to make me endure all that complexity if I don't need it.

Re:.NET is OOP gone stupid.

The first time I saw what you (supposedly) have to do to read from a file in Java, it pegged my OMGWTF meter.

The idea is that you could encapsulate all that complexity inside a method inside a class--instantiate that class inside a class that has a "main()" and then put the whole thing in a module. You call all of that method with the correct parameters in an instance of another class created and instantiated the same way. You then jar it up as bytecode and then run it on the JVM--making sure your users are running the right versions of the JVM.

On second thought, OMGWTF?

Re:.NET is OOP gone stupid.

I kinda disagree.
To talk about your example: fopen( ) might be nice and simple, but the capabilities provided by .net and Java are much bigger in scope.

You can use them to read and write files with different encodings, you can treat a lot of other things as files, and combined with formatters you could serialize your data to binary files or XML almost without writing code.

Even more, the different classes are orthogonal, so you can mix and match different encodings, formattings, and file operations without the combinatoric explosion of having a separate function for every possible operation. It's an elegant design in my opinion.

Furthermore, the libraries of Java and .net provide standard interfaces and hooks to link your own code. Want arrays of your new data type to have automatic sorting capabilities? just implement IComparable. A little bit of work would let your new collection class bind automatically to Winforms' data grid control. And many more examples.

If you remove .net's huge libraries, you get a situation like C++ where there are half a dozen pseudo-standard libraries for encryption, networking, GUI and stuff. You have projects with incompatible dependencies and a lot of wasted effort writing, debugging and maintaining all those libraries. Microsoft may have a lot of problems with their products but .net is one of the most well designed things they've produced.

To be fair, .net inherited a lot of this from Java, but they improved on it. Java, in turn, adapted/improved the Smalltalk libraries that have helped pave the way for the "language with everything included" paradigm.

Re:.NET is OOP gone stupid.

My experiences with Java were painful, but they are out of date, so I'm only going to talk about .NET, which I actually use in my day-to-day work because it's actually the best tool for many of the jobs I do.

.NET and Java are both prime examples of object-oriented programming gone stupid. Their class libraries have become so utterly huge that it becomes damn near impossible for an individual developer to suitably grasp anything more than a small portion of them.

Interestingly enough, an individual developer does not need to grasp anything more than a small portion of them. An individual developer needs to know the basics of the core class library and whatever else he needs to get his job done. The vastness of the ASP.NET (or whatever) libraries is not an impediment to one who does not use them.

Also, there is documentation, and Intellisense (freely available, now), and a naming convention that actually makes sense after a while. F1 isn't that hard to press.

Although they supposedly give more flexibility, something as essential as reading from and writing to a file becomes a hassle with .NET or Java. It's easy to get lost in whether we need a FileInputStream, or whether we should wrap a FileInputReader in a TextInputBuffer, and so forth. Give me fopen() any day

Seriously? You actually found string[] lines = File.ReadAllLines( string path ); to be difficult? Or are you just talking out your ass?

Of course, there are more complicated examples, but that's usually because they're either years out of date (.NET 1.0), or just plain doing more.

OO was supposed to solve the problems of writing applications in languages like C, Pascal and Fortran. All it has done is brought in a new level of complexity that results in monstrosities like the Java and .NET standard class libraries. Meanwhile, the POSIX API offers just as much flexibility, but is far easier to work with. Not to mention that programs using it are far more efficient.

Yeah? I find typing File[dot] and hunting through the fairly short list of methods easier than remembering what the valid values were for fopen's mode parameter are, and whether there was a platform-specific one I should be using to get the file-locking behavior I want. And file.Read( ... ) is a lot neater than fread( ..., file ). You can go on all you want about how I'm being lazy, but I have more important things to commit to memory than API entry points and the quirks of their parameters (like, say, the overall structure of my app, or the problem it's being written to solve). YMMV, of course.

As for plain C applications being more efficient, well, what exactly does that have to do with what methods are named and what namespaces they do (or do not) reside in? Second, that's not the point. Getting a quick GUI app up and running in a hurry is more what you'd use .NET for, something you can't even begin to do in C until you've sat around for a while thinking about fun things like memory management for shared resources.

Yes, C is valuable and it's still pretty much the best choice for writing tight, high-performance loops that do lots of pointer-manipulating, bit-twiddling evil - that's what I and every other sane programmer I know uses it for. But it's also a damn waste of my time to be using it to write Win32 GUIs for art tools. My time is more valuable than a few CPU cycles.

Re:.NET is OOP gone stupid.

In .NET:

byte[] fileContents = System.IO.File.ReadAllBytes("myBinary.blah");
string fileText = System.IO.File.ReadAllText("myText.txt");

That's if you want to read it all in as quickly as possible (no buffering). What's tough about that?

Obviously if you need buffering you have to do some REALLY complex work:

while (s.Position < s.Length)
{ //process your stream... read one byte at a time out of it
    int oneLittleByte = s.ReadByte(); //or chunk 50 bytes out of it, don't hardcode like this kids!
    byte[] someBytes = new byte[50];
    int bytesRead = s.Read(someBytes, s.Position, someBytes.Length);
}

That is both tough and complex. I don't know how I can cope.

Re:This is News?

It doesn't suck because it's made by MS or ripped off from something. It sucks because the documentation is piss-poor. And there isn't a single working (i.e. cut-and-paste) example for a single API (someone told me they break them on purpose so that newbies don't cut-paste themselves into security holes without understanding exactly how the thing works, but hell!)

I noticed the same thing when Apple released their Cocoa framework (with over half of help pages saying "TODO: descrition, example"). Some Quicktime documentation is still that way today!

How anyone expects the undocumented API stuff to be useful is beyond me.

Re:The horror!

Hallelujah, brother!

What bugs the snot out of me is that a lot of that stuff is documented really well, but only on developers' blogs. What the hell kind of insulting documentation non-strategy is that? And of course, there's no cross-referencing between msdn and "the blogosphere." So you get to churn away at a search engine until you find a blog entry that's kind of addressing what you want to know.

That said, I do like a lot of stuff about C#. Delegates, for example rank high on that list. And C# 3.5 offers some pretty cool new stuff as well. I likey the lambda expressions, inferred typing, and LINQ.

But the documentation does make me cry at night, sometimes. Sometimes.

Re:The horror!

I have mod points, but this deserves about 10 of them so I reply in solidarity.
The msdn docs are not enough, there are too many useless empty API pages. If it weren't for Google I couldn't do any .net programming work. It's ironic that their worst enemy is essential to working with their system.

Often the most useful documentation and samples are scattered over a zillion forums (all requiring logins!), newsgroups, blogs, 3rd party books, etc.

This has always been the case with microsoft development libraries, but it's getting worse because early on (internet time) it was mostly just usenet.

I agree

The biggest plus to working in Java is that the documentation Sun provide is comprehensive (and once you figure your way around it) easy to access.

Microsoft is getting better but, for a framework that's quite clearly a java rip-off, they could have ripped its documentation style too.

Re:Answer: No Thanks

Let me get this straight.

1. You argue that .NET lacks "internal logic" and "simplicity" after pointing to the article. Never mind that the article merely only reported statistics about the .NET framework instead of asserting that it, does, in fact, support your arguments. So this claim remains baseless, unless you're trying to say lots of types somehow means it lacks internal logic and simplicity. We are waiting for your keen insight on this point.

2. Another user questions your assumption rather innocently.

3. You imply that they did not read the article (which is rather hilarious considering the previous point) and then, to add the icing to the cake, indicate you'd much rather work on Mono. Mono is a version of .NET that runs on non-MS platforms and is compatible at the bytecode level. What sort of examination have you done on the source code of it to determine that it has "simplicity" and "internal logic?" How does it meet those goals, yet have the same external API of .NET? How does it not suffer from "bloatedness" if it has at least as many publically available classes as .NET?

We await your answers, mighty Naughty Bob.

Re:Answer: No Thanks

Did you bother clicking the link to "Mono?" It links to this URL:
http://en.wikipedia.org/wiki/Infectious_mononucleosis [wikipedia.org]