最近Stack Overflow上有一群讨厌perl的人,所以我想我应该把我的“关于你最喜欢的语言你讨厌的五件事”的问题带到Stack Overflow上。拿你最喜欢的语言来说,告诉我你讨厌它的五件事。这些可能只是让你烦恼的事情,承认的设计缺陷,公认的性能问题,或任何其他类别。你只需要讨厌它,它必须是你最喜欢的语言。
不要拿它和其他语言比较,也不要谈论你已经讨厌的语言。不要用你最喜欢的语言谈论你喜欢的事情。我只是想听到你讨厌但能容忍的东西,这样你就可以使用所有其他的东西,我想听到你希望别人使用的语言。
每当有人试图把他们最喜欢的语言强加给我时,我就会问这个问题,有时是面试问题。如果有人找不出他最喜欢的工具的5个缺点,那他对它还不够了解,不能提倡它,也不能利用它赚大钱。他还没有在足够多的不同情况下使用它来充分探索它。他把它作为一种文化或宗教来倡导,这意味着如果我不选择他最喜欢的技术,我就错了。
我不在乎你用什么语言。不想使用特定的语言?那就不要。你通过尽职调查做出了明智的选择,但仍然没有使用它?好吧。有时正确的答案是“你有一个强大的编程团队,有良好的实践和丰富的Bar经验。改成Foo是愚蠢的。”
This is a good question for code reviews too. People who really know a codebase will have all sorts of suggestions for it, and those who don't know it so well have non-specific complaints. I ask things like "If you could start over on this project, what would you do differently?" In this fantasy land, users and programmers get to complain about anything and everything they don't like. "I want a better interface", "I want to separate the model from the view", "I'd use this module instead of this other one", "I'd rename this set of methods", or whatever they really don't like about the current situation. That's how I get a handle on how much a particular developer knows about the codebase. It's also a clue about how much of the programmer's ego is tied up in what he's telling me.
憎恨并不是衡量人们了解多少的唯一尺度,但我发现它是一个相当不错的尺度。他们讨厌的事情也让我知道他们对这个话题的思考有多好。
JavaScript:
The Object prototype can be modified. Every single object in your program gets new properties, and something probably breaks.
All objects are hash maps, but it's difficult to safely use them as such. In particular, if one of your keys happens to be __proto__, you're in trouble.
No object closure at function reference time. In fact, no object closure at all -- instead, this is set whenever a function is called with object notation or the new operator. Results in much confusion, particularly when creating event callbacks, because this isn't set to what the programmer expects.
Corollary: calling a function without object notation or the new operator results in this being set equal to the global object, resulting in much breakage.
Addition operator overloaded to also perform string concatenation, despite the two operations being fundamentally different. Results in pain when a value you expect to be a number is in fact a string.
== and != operators perform type coercion. Comparisons between different types involve a list of rules that no mortal can remember in full. This is mitigated by the existence of === and !== operators.
Both null and undefined exist, with subtly different, yet redundant meanings. Why?
Weird syntax for setting up prototype chains.
parseInt(s) expects a C-style number, so treats values with leading zeroes as octal, etc. You can at least parseInt(s, 10) but the default behaviour is confusing.
No block scope.
Can declare the same variable more than once.
Can use a variable without declaring it, in which case it's global and probably breaks your program.
with { }.
Really difficult to document with JavaDoc like tools.
C++
Strings.
They are not interoperable with platform strings, so you end up using std::vector half of the time. The copy policy (copy on write or deep copy) is not defined, so performance guarantees can not be given for straightforward syntax. Sometimes they rely on STL algorithms that are not very intuitive to use. Too many libraries roll their own which are unfortunately much more comfortable to use. Unless you have to combine them.
Variety of string representations
Now, this is a little bit of a platform problem - but I still hope it would have been better when a less obstinate standard string class would have been available earlier. The following string representations I use frequently:
generic LPCTSTR,
LPC(W)STR allocated by CoTaskMemAlloc,
BSTR, _bstr _t
(w)string,
CString,
std::vector
a roll-my-own class (sigh) that adds range checking and basic operations to a (w)char * buffer of known length
Build model.
I am sick to death of all the time spent muddling around with who-includes-what, forward declarations, optimizing precompiled headers and includes to keep at least incremental build times bearable, etc. It was great in the eighties, but now? There are so many hurdles to packing up a piece of code so it can be reused that even moms dog gets bored listening to me.
Hard to parse
This makes external tools especially hard to write, and get right. And today, we C++ guys are lacking mostly in the tool chain. I love my C# reflection and delegates but I can live without them. Without great refactoring, I can't.
Threading is too hard
Language doesn't even recognize it (by now), and the freedoms of the compiler - while great - are to painful.
Static and on-demand initialization
Technically, I cheat here: this is another puzzle piece in the "wrap up code for reuse": It's a nightmare to get something initialized only when it is needed. The best solution to all other redist problems is throwing everything into headers, this problem says "neeener - you cannot".
诚然,其中许多内容超出了严格的语言范围,但在我看来,整个工具链都需要进行判断和发展。
Python:
No standard GUI toolkit (the community goes round and round about this but never seems to settle on anything).
The evolution of tools and methods to distribute and install Python apps and libraries has been, well, rocky. (Although lately this seems to be moving closer to getting fixed.)
CPython is still slow as interpreters go (although PyPy is looking pretty good these days, if it becomes the "standard" Python this problem goes away).
You can't subclass built-in classes (e.g., list and dict) without overriding a lot of methods, even if all you want to do is a simple hook into an event (e.g., to hook into an item being added to or removed from the list, you need to override delitem, append, extend, insert, pop, and remove--there's no subclassable "change" event notification, nor any "protected" methods that factor out common code used by all the above methods).
Up until virtualenv was invented, keeping separate Python environments for different purposes on one machine was a real pain.
Python:
1) It's a scripting language and not a fully compiled one (I'd prefer to be able to compile binaries—I don't care about bytecode). This is very annoying if I have to use very many libraries (i.e. everyone who uses my program has to install all the libraries, and this basically means no normal people will be able to, or have the patience to, properly set it up—unless I do a ton of work that should be unnecessary). I know ways to make binaries, but they don't always work, and I'm guessing they bundle the interpreter in the binaries anyhow (and I don't want that). Now, if I could get a bytecode compiler that would include copies of all the files that I imported (and only those) to be placed in my program's folder, that might be a suitable compromise (then no one would have to download extra libraries and such). It would also be nice if the compiled python files could be compressed into a single file with one specified as the file to run the program before this is done.
2)有时看起来有点bug;有几次,应该工作的代码根本没有工作(没有程序员错误),特别是与“from moduleX import *”和其他导入相关的问题有关的代码,以及一些与全局和局部变量有关的问题。
3)最大递归深度可以更高。至少有一次,我觉得我需要它去更高的地方。
4)没有switch语句(更不用说允许数字、字符串和范围的语句)
5)新版本的Python似乎取消了很多有用的字符串操作,而且似乎没有简单的文档说明如何在没有它们的情况下做同样的事情。
6)强制自动垃圾收集(我希望能够手动执行,尽管不一定强制执行)。
7)没有预先制作的定时器类没有使用GUI(好吧,可能有一个,但在我所做的所有搜索之后,它肯定不方便找到!我确实找到了一些东西,但当我尝试时,它根本不起作用。)所谓计时器,我指的是每隔x秒执行一个指定函数的排序,并能在需要时关闭它,等等。
8)社区里举例的人很少告诉我们他们导入了哪些模块,以及他们是如何导入的。
9)与Lua集成的支持并不多。
10)似乎没有办法向一个类的特定实例(而不是整个类)添加一个额外的函数,除非你动态地向该类添加一个对象变量,该对象具有所需的函数(但仍然,你必须为此创建另一个类)。