The C++11 thread library (http://en.cppreference.com/w/cpp/thread) is at least not insane, and at best you might even like it. In my view it's certainly a better bet than pthreads, because it runs on Windows too.
Thanks. Why don't I use the C++11 thread library? 1, I wrote this code on Linux and it's only used on Linux, so I don't consider portability issues. 2, actually, I dislike some classes name in the C++ thread library, so....
Note that this is not the same as "using namespace std;" since you are only exporting the namespace std::string.
This should almost always be ok since no sane programmer should ever name their class string in the namespace of std.
As a developer, I wouldn't feel too bad about causing namespace conflicts in this manner because hopefully it will suggest to the maintainer that their class name is a poor choice.
Actually, it tells the compiler that every time it comes across a class called string that isn't fully qualified, it should use std::string. And since he's put it in the global namespace, it will apply everywhere this header is visible.
He really shouldn't do that in a header, but if he at least moved it down a couple of lines so it's inside his private namespace, it at least wouldn't affect anyone else. There is no good reason to do it globally.
But that begs the question why do you have two classes named string to begin with? The ambiguous name choice of string for an alternative to the standard is a terrible idea and all the problems associated with it should be the responsibility of the person making that decision. Putting using std::string in the header would at the very least make whoever is using your class strongly consider the necessity of 2 classes named string. They would be forced to use the fully qualified name for their type instead of ambiguously referring to their alternative string class as string. This prevents future maintainers from getting confused as to why there are 2 versions of the class string.
But "string" in std isn't ambiguous, precisely because it's in the std namespace.
As to why a program would have two classes called string... well, there's a funny thing there, because it didn't, at least not until it started using this library that does "using std::string" in its header...
(Anyway, why is it the library author's concern? They're writing a library! Not only is this the tail wagging the dog, but they're actually making it harder to use, and easier to say no to. Exactly the wrong thing.)
I think we're viewing the issue from 2 different stand points. You are correct that from a library authors perspective they shouldn't impose undue restrictions. I was thinking more along the lines of code maintenance within the same library module where it could be helpful to signal to others that come along that the stl string is the expected default. I think most people would agree that (using...) in the header is a bad idea if the header is intended to be apart of the Public API for the library but I think its use can be justified in a few select instances of internal headers in order to impose cohesive coding convention.
If its my library and I don't want other maintainers adding additional string classes internally, I can at the very least force them to use fully qualified names for their alternative string classes to avoid ambiguity. I hate working on code that has 5 different versions of the same freaking thing because each developer decided to implement their own version when the standard way would have sufficed. That was the point I was trying to make. Just like putting const and override on functions, it's a way to help maintainers avoid doing dumb things.
It kind of sounds like you're looking for technical means by which to "signal to others". In that case, perhaps some documentation (for other contributors/maintainers) could be of use?
Beyond that I'm not sure what to suggest. The problem is that even if you have a Really Good Reason for what might otherwise appear to be a bit of gratuitous incompatibility, it's rather unlikely that others will figure out what that reason is without some explanation. IME anyway.
Documentation is only good if people read it... which most developers don't. If someone tried to make a class called string in my module, they would receive immediate feed back from the compiler that there are now multiple definitions of string which should tell them that they are creating another copy of something that already exists.
Nevertheless, I'll concede that its not the best way to go about things. But it is something you can do and I wanted to highlight a case of why a "Never do This" feature is even allowed in the first place.
Another simpleapproach to exception handling (beyond calling abort) is to allow an exception callback to be registered. This can allow appropriate diagnostic logging, and/or compensation
the c++11 library std::async provides a built in mechanism for spawning asynchronous tasks. Under the hood, std::async uses a thread pool. So I'm asking why use this library when the stl already provides a thread-pool mechanism?
I think The C++ thread library is overdesigned so it lost some function. What's more, on Linux it's chicken ribs. So I encapsulate POSIX Threads using C++, to take advantage of RAII.
Good explanation, and frankly I agree. From looking through it, I like the control your lib allows. If its performance is superior on Linux as you say, nice work!
Thanks! BTW, the encapsulation is not finished yet, such as mutex and condition variable.
I'll go on with the work, and welcome to contribute if you're interested!
28 comments
[ 2.4 ms ] story [ 70.0 ms ] threadTalks in-depth about the thread library and also other C++11 features that made it possible. Highly recommended.
using std::string;
It's not cool to impose that decision on clients of the library.
This should almost always be ok since no sane programmer should ever name their class string in the namespace of std.
As a developer, I wouldn't feel too bad about causing namespace conflicts in this manner because hopefully it will suggest to the maintainer that their class name is a poor choice.
He really shouldn't do that in a header, but if he at least moved it down a couple of lines so it's inside his private namespace, it at least wouldn't affect anyone else. There is no good reason to do it globally.
As to why a program would have two classes called string... well, there's a funny thing there, because it didn't, at least not until it started using this library that does "using std::string" in its header...
(Anyway, why is it the library author's concern? They're writing a library! Not only is this the tail wagging the dog, but they're actually making it harder to use, and easier to say no to. Exactly the wrong thing.)
If its my library and I don't want other maintainers adding additional string classes internally, I can at the very least force them to use fully qualified names for their alternative string classes to avoid ambiguity. I hate working on code that has 5 different versions of the same freaking thing because each developer decided to implement their own version when the standard way would have sufficed. That was the point I was trying to make. Just like putting const and override on functions, it's a way to help maintainers avoid doing dumb things.
Beyond that I'm not sure what to suggest. The problem is that even if you have a Really Good Reason for what might otherwise appear to be a bit of gratuitous incompatibility, it's rather unlikely that others will figure out what that reason is without some explanation. IME anyway.
Nevertheless, I'll concede that its not the best way to go about things. But it is something you can do and I wanted to highlight a case of why a "Never do This" feature is even allowed in the first place.