c#(好吧，我猜一部分是VisualStudio IDE):

No covariance (yet), like Class<D> cannot be used in place of Class<B> even though type D derives from type B. Graphic designers don't support generic based inheritance (or inheritance from abstract classes), even though the inheritance itself works just fine if you work around the designer problems by adding extra inheritance levels just so designers always see concrete non-generic variants of your code. No constructor inheritance No constructors in where clauses of generic type parameters VisualStudio seems to have a tendency to mysteriously check out files (like project files and/or unit test definitions) when opening a solution, even though the files do not seem to actually get altered.

如果你明天再问我，可能会有不同的名单。即使协方差和设计师的麻烦将在我的前5名，直到他们被解决(随着方差添加到c# 4.0，这似乎已经发生了至少一个…)

F#

Type inference is limited. It propagates forward only. F# won't try to infer an object type based on the methods and properties used: you'll get "lookup of indeterminate object type" errors when it doesn't have a clue. One cannot mix floats and ints: 1 + 2.3 is a type error. It's a little awkward to have to create a builder object in order to define a monad or computation expression. In Haskell or Scala, you can define the monad operations directly on the monadic object. Though the #light syntax is preferred, the indentation rules are sometimes not very intuitive or become cumbersome.

Scala是我最喜欢的语言。五件讨厌的事?容易:

Takes a long time to learn properly. I know you can write Scala as a 'better java'. That is what we used to say about C++ and C too. I agree this is an inevitable consequence of the deep ideas in the language. But still ... Methods vs. Functions: def f(x: Int) = x*x defines a method f, not a function f. Methods are not functions despite a lot of early Scala tutorial material blurring the distinction. The language tries to blur it too because if you supply a method in some places where a function is expected it is accepted. Do we have to have both methods and functions? Yes it is fundamental. But it was initially confusing to me. Composing classes or objects from mixins in the 'cake' pattern is prone to NPE's. e.g. trait X { val host: String; val url = "http://" + host } is a mixin that will NPE on instantiation, or not, depending on its position in the class declaration. The compiler could tell you if it will fail but doesn't. (In 2.7 anyway.) It is hard to diagnose the problem in complex inheritance graphs. Arrays in 2.8 rely on implicits to mesh with the main scala collection types. But implicits are not applied everywhere. An Array can be supplied where a Seq is expected. But an Option[Array] cannot be supplied where an Option[Seq] is expected. I know there are no completely 'right' ways to handle java Arrays. Type erasure. Enough said.

C

No parametric polymorphism (i.e. C++ templates). It makes writing reusable data structures and algorithms a pain (and there's hardly any static checking). See for instance the comparator argument to qsort and bsearch: the comparator takes void pointers :( No library of data structures. I really hate writing my own hash table. I also really hate scouring the web for a library of reusable data structures. Especially if it turns out to be incomplete. Strings. Inefficient representation, unwieldy if you make it sane, too hard to safely input a string. No standard for snprintf. Too hard to create a format string with sprintf, then use that to create a string with sprintf again, in a safe way. Only lexical macros. If different compilers expects function annotation in different places, I have to put the same HAS_NO_SIDE_EFFECTS in different places. Why can't I just grab the function, switch over the compiler type, and then insert it at the right place by a macro call? No portable libraries for common functionality. For sockets and threading, I use SDL---a frigging game library. For .ini-style parsers, the only library I could find which was packaged for ubuntu, I posted on the daily wtf (it calculates an array of hash values, then does a linear scan through it...)

C++

Template syntax is heavy and unweildy. Let's see, for(map<string, int>::const_iterator it = mymap.begin(); it != mymap.end(); ++it). Design errors in the STL. Should changing allocation strategy for your vector really change its type? Overly complex type system. Type T1 has a convert-to-T2 method, and T2 has an implicit from-T1 constructor. Which is called? How does overloading, overriding and multiple inheritance interact? Poorly, I guess... Incredibly long and unwieldy error messages from templates. You know what I mean... References means you can't see output parameters at call sites. In C, you can guess what foo(bar, &baz) can and can't modify.

VB6

仅Windows。 不再支持。 数组可以从任何数字开始，而不是全部归一化为0。 已编译的应用程序依赖于许多dll才能正常运行。 许多复杂的控件(如浏览器控件或复杂的代码段)在运行未编译的代码时往往会破坏IDE，但在编译后却可以正常工作。

C#

1)缺乏为值类型编写泛型的实际能力。例如，任何白痴(好吧，大多数白痴)都可以编写一个例程，用c++计算int, float, double等列表的标准偏差，它写起来很简单，易于阅读，并作为快速的非泛型代码执行。我认为，如果你能用c#写一些东西，接近于达到这三个中的任何一个，而在其他两个上又不荒谬，你就是一个真正伟大的程序员。

2)协方差和反方差，尽管这被添加到4。

3)非常糟糕的LINQ文档(好吧，并不是语言的一部分)。

4)尝试使用foreach/迭代器，当我每次都想做同样的事情，除了上次略有不同(如连接一串字符串与逗号之间的单词和最后两个)。如果我用一个IEnumerable来写它，它很难写和读，而用一个for (int I =0 I <…)它并没有好到哪里去，而且效率更低。

5)我知道我会收到抱怨，但是缺少受控的例外。这并不需要像在java中那样实现(框架开发人员确实提出了一些很好的观点，为什么他们没有这样做)，但我很乐意看到编译器警告不喜欢受控异常的用户可以关闭。