c# / .net:

Classes should be sealed by default There should be no lock statement - instead, you should have specific locking objects, and there should be methods such as Acquire which return disposable lock tokens. Corollary: there shouldn't be a monitor for every object. GetHashCode() and Equals() shouldn't be in System.Object - not everything's suitable for hashing. Instead, have an IdentityComparer which does the same thing, and keep the IComparer<T>, IComparable<T>, IEqualityComparer<T> and IEquatable<T> interfaces for custom comparisons. Poor support for immutability Poor way of discovering extension methods - it should be a much more conscious decision than just the fact that I'm using a namespace.

这些都是我想出来的，明天问我，我会想出一个不同的5个:)

ActionScript / AS3

没有抽象类 没有私有构造函数(所以单例是一种hack) FP10之前没有类型化数组 Flash IDE中的编译/发布时间慢得离谱 内置函数(例如Math)的性能较慢

除此之外，它实际上是一种很好的语言——与流行的观点相反，它比JavaScript好得多，比PHP之类的语言好一百万倍。

C#

I wish I could switch() on any type, and that case could be any expression. Can't use object initializer syntax with 'readonly' fields / private set autoprops. Generally, I want language help with making immutable types. Use of {} for namespace and class and method and property/indexer blocks and multi-statement blocks and array initializers. Makes it hard to figure out where you are when they're far apart or mismatched. I hate writing (from x in y ... select).Z(). I don't want to have to fall back to method call syntax because the query syntax is missing something. I want a do clause on query syntax, which is like foreach. But it's not really a query then.

我真的到达这里了。我认为c#非常棒，而且很难发现它有什么缺陷。

Lua

If you do foo.bar(1,2) then 'self' is nil inside the bar method. You must remember to do foo:bar(1,2) instead. I'd rather have that switched ('self' should be defined by default unless you use the ':' operator, or you call a function that isn't a method). Variables are global by default. I'd rather ditch the 'local' keyword and have a 'global' one instead. Undeclared variables are assigned the nil. I'd rather receive an error message. You can sidestep this by manipulating the global env's metatable, but I'd rather have it implemented by default and be able to deactivate it. Multiple returned values on parameters are not handled very nicely. Say you have a function foo() that returns 1,2,3 (three values) and bar() returns 4,5 (two values). If you do print(foo(),bar()) you will get "1,4,5" ... only the "last" tuple is expanded on calls. The # (table length) operator only works in tables indexed with continuous integers. If your table isn't like that and you want to know how many elements does it have, you need to either parse it with a loop, or update a counter each time you insert/remove an element from it.

Objective Caml

Non-concurrent garbage collector. I can write multi-threaded programs all day long, but they're only ever going to get one of my eight cores at a time. This makes me sad. No type classes (or their moral equivalent). There's Furuse-san's GCaml, but it's A) not quite as good as type classes, and B) not in the INRIA distribution. Badly in need of a Cocoa bridge. Seriously. If I wrote more code with actual interfaces to DNA-based life forms, then I'd probably break down and write the damned thing myself. Why hasn't anybody else done this yet? Functors are abominable. Seriously, modules ought to be first-class values. There should be only one kind of function. Read Montagu and Rémy before you flame me for this. Should use LLVM for its back-end. Who do I have to murder to get OCaml to compile for my stupid little ARM6 core?

是的，我有一些问题。我仍然非常喜欢这门语言。这太棒了。

TCL

这是我最喜欢的语言，几乎可以做任何事情。多年来，它已经(慢慢地，非常缓慢地)演变为解决大多数让我烦恼的事情。而且这门语言非常灵活，很容易实现语法来覆盖那些仍然困扰我的东西。但是语言中有一些东西是不能轻易改变的，只是打破了它的禅意:

Arrays (of the associative kind, what Perl calls hash) don't have proper value semantics. This makes them awkward to pass to and return from functions. Also, this means that they can't be nested. For this reason dicts (dictionaries) were invented but too late, the nice array access syntax: $array($foo) is now forever taken by stupid arrays for backwards compatibility. We're now stuck with: dict get $dict $foo which is much more verbose and to me feels less readable. No real closures. Though it can be emulated somewhat by globals or namespaces but that defeats the reason for closures in the first place. Although, I can't really see for now how closures can be implemented in a pure value semantics system. Teacup is hard to use and is not at all intuitive compared to all other repository tool out there. This is more ActiveState's fault than tcl-core and doesn't really break tcl's Zen when coding but it is still very annoying.