(C++11 is actually an interesting thing to analyze regarding this tradeoff, by the way: the new "common initializer" syntax is designed to provide as much of the benefits as possible of a simplified built-in data type syntax without taking on the semantic burden of having it; however, it also does not provide syntax that ends up being entirely devoid of the type of the container.)
You can't redefine haskell's syntax unless you're using Template Haskell (which is a language extension, not part of Haskell itself). It also has nothing to do with monads. Likewise with most MLs, or with Erlang. All of them have a literal list syntax.
And if containers have no reason to be special, why would strings be special? They're just sequences of unicode codepoints after all.
Continuing your argument into absurdity, why have literal syntax for most datatype at all really? You could just shove a bag of bytes into a constructor when you want integers or floats as well. Now you've got one literal syntax (which isn't even for a datatype): bunch of bytes.
As for strings, it is very seldom that you find interesting alternative implementations: the only one I can think of is a rope. Interestingly, C++11 now allows you to override string literals, so you can actually do this.
Sure but it's not syntax redefinition.
> Interestingly, C++11 now allows you to override string literals, so you can actually do this.
You still have a literal string notation. Literal notations don't have to impede multiple implemetations, and the truth is there is generally a primary representation used for the vast majority of cases (even if that representation is a cluster class and flexible under the interface).
In Cocoa, the primary sequence and maps are NSArray and NSDictionary, what would be the issue with making those literal? And one of your objections is
> When this suddenly matters, you are in the situation where what you want to be able to do is to make a very small modification to areas of your code where you need to select a different algorithm, in order to get the different result; you don't want to be forced to rewrite half your code to use a different syntax just because it was slow
But that makes no sense: as long as all equivalent containers implement the same interface (which they do, or you can't swap them anyway) that creating an object be done with a literal or with a constructor and a bunch of messages has no influence on the rest of the code, the only thing you need to change is the initialization code in both cases.
Hell, a smart enough editor can even swap between the literal and the "constructor" versions of a given collection (IntelliJ can do that for Python dicts, for instance). Not to mention in many cases the non-literal can just take the literal as a parameter, if the collection with a literal syntax has been well chosen, that way you get your cake eat it.
http://gcc.gnu.org/onlinedocs/gcc/Constant-string-objects.ht...
I will also, though, point out that I write almost as much (if not more) Python as I do Objective-C++ these days: I therefore can be said to certainly not consider Python to be for "stupid people", without including myself in that set. ;P
Languages that some look at as "real programming languages", in comparison, tend to not have syntax like this, and the reason why is that you often, either now, or at some point later, are going to care whether the data structure you just allocated is a red-black tree, a hash map, a patricia trie, or even an AVL tree (which I include mostly to make a point: there actually are situations where it is preferred to a red-black tree).
When this suddenly matters, you are in the situation where what you want to be able to do is to make a very small modification to areas of your code where you need to select a different algorithm, in order to get the different result; you don't want to be forced to rewrite half your code to use a different syntax just because it was slow (I mean, if you wanted to do that, you'd have written it in Ruby and then recoded it in C).
Therefore, you find that it is normally the case in languages like Java, C++, and Objective-C, that there are no "built-in container types", as you will never find a container type that is actually correct to use in an even fractional majority of the cases; in fact, most of the time, there isn't even a single obvious choice in these languages for what class to use: you find default implementations of multiple algorithms.
Objective-C, here, is no different from this concept: NSDictionary is just an interface, and can be implemented by numerous backends. Apple has a rather good implementation backing the default version, and even attempts to switch between algorithms as the data structure grows, but your code is always just a few identifiers away from choosing a different subclass in that collection hierarchy.