Written by Sarah Thompson - sarah@telergy.com
You can find the latest version of this documentation at http://sigslot.sourceforge.net/
For downloads, mailing lists, bug reports, help and support, or anything similar, go to http://sourceforge.net/projects/sigslot
The author's personal web site is at http://www.findatlantis.com/
Downloading the sigslot library
The C++ programming language is an amazingly capable beast. However, it is no panacea for all ills - it is just as possible to write spaghetti in this language as it is in any other (arguably easier than some).
C++ is usually described as an object-oriented language. Its claim to object orientation is generally appropriate, but unlike some other languages, C++ doesn't actually enforce good practice. Whether or not you think this is a good thing or a bad thing is up to you, I'm not interested in starting a religious war here, so I'll neatly sidestep the issue by changing the subject for a moment.
|
Flame me, flame me not |
|
Before you lock & load your rocket launchers, folks, I'm using the MFC by way of example only. The signal/slot library is ISO C++ compliant (at least where possible) and will work on pretty much anything. All you need is a reasonable C++ compiler that supports templates. You don't need partial template specialisation support, so VC6 and VC.NET are both fine. |
The Microsoft Foundation Classes (MFC) are a nightmarish mess. I'll say it again - the MFC is a programmer's nightmare. At one point it almost inspired me to give up programming and take up something less irritating, like becoming a professional 'nut the javelin' player or move to Outer Patagonia and become a cat wrangler. Nevertheless, the MFC is capable of supporting most things that you might want to do with the Windows platform, at the price of a little slice of one's sanity.
There are two reasons why the MFC is so painful. Firstly, it too-often seems to require an uphill struggle in order to get the simplest things to work. Secondly, it is very difficult and/or evilly messy trying to get disparate classes within an MFC application to talk to each other. The former problem is down to the design of the MFC itself - only deep experience using it can really get around this. The second problem is a consequence of the MFC being quite a thin wrapper around the underlying Windows SDK - getting one window to talk to another window often requires resorting to oldfashioned Windows 3.1 techniques like ::SendMessage() or ::PostMessage() with custom Windows messages. Yuck. That kind of code is difficult to write, difficult to debug, hard to maintain and generally speaking a Bad Idea.
Take a simple example. Imagine we want to create a dialog box, with a number displayed on it, and a couple of buttons which let you increment or decrement the number:
If you're going to program the MFC way, you'd slap in that dialog to the dialog editor in Visual Studio, then add message handlers for the 'Increment' and 'Decrement' buttons that use the MFC's DDX mechanism to update the number in the edit box. Easy enough you might think, but where's the reusability? I signed up to the card-carring object orientation movement on the basis that it was supposed to make it easier to reuse my code, or to use components written by other people with a minimum of pain. The MFC way of working might as well be Visual Basic, requiring cut-and-paste code reuse. Evil, nasty, bad.
What, in circumstances like this, I'd really like to be able to do is turn the numeric edit control and the increment/decrement buttons into genuinely reusable components in their own right. By component, I don't mean COM object or Active-X control, just a simple, lightwight, reusable C++ class.
The MFC way of doing this, reasonably enough, would be to inherit from the basic controls (CEdit and CButton in this example), then give those classes a common interface of some kind. You might, sensibly enough, add a couple of entry points to your numeric edit class, 'CMyEdit::Increment()' and 'CMyEdit::Decrement()', say, then in your CIncrementButton and CDecrementButton classes hold a pointer to a CMyEdit so that they can implement OnOK() handlers that call the relevant member functons in CMyEdit. This is fractionally better, but still not at all nice. What I'd like is to be able to create a general purpose button, that, er, well, 'clicks' when you click it, and an edit box that can be directed to do a variety of things remotely, such as increment, decrement, copy to clipboard, go back to zero, etc. The edit box shouldn't need to care what is calling it. Just as importantly, buttons shouldn't have to care what they are calling either, so the same, unmodified button class can be used to increment, decrement or clear the control this afternoon, and next week end up wired to my internet compliant remote garage door opener. Or whatever. But the important thing is that only the user of the classes should need to know (or care) how they are wired up - the classes themselves, to be genuinely reusable, should be above that kind of thing.
Speaking as an ex-hardware designer, I'd like software components to have pins around their edges, like ICs, so they can be 'wired up', but otherwise should be well behaved. To design a board with, say, an 8 bit microcontroller on it is made much easier by knowing that pin 34 is always Chip Enable (or something). I don't want to know how that pin is wired internally - I just want to know that it will work when I send a signal into it.
The Qt library (see http://www.troll.no/ for further information) was the first attempt I personally ever saw at extending C++ by adding a 'signal-slot' metaphor to the language's existing repertoire of programming techniques.
Qt was a revelation to me when I started using it, which must be something like three years ago at the time of writing. For once, I could write code relatively quickly, with a reasonable chance of it actually working the way I intended it without the days of pushing water uphill with a fork that was usually required to breathe life into MFC applications. The thing that impressed me most about Qt was its signal/slot metaphor. Qt uses a preprocessor, moc, to preprocess an extended C++ syntax. Put briefly, any Qt class can possess one or more signals, and one or more slots. A slot is very much like an ordinary member function. Indeed, slots can be called directly as member functions, with the only syntactic difference being the need for the slots keyword in the class header file. A signal in Qt is declared much like a member function, except that it has no local implementation - it just appears like a member function prototype in the header file. Signals can be connected, using the connect function, with any number of slots in any other C++ objects. When a signal is emitted, using the new keyword emit, all connected slots get called.
In principle, signals and slots are a bit like pointers, where a signal can be 'wired up' after the fact to the slots that need to be informed whenever it is emitted. Using pointers, be they function pointers, pointers to classes or pointers to class member functions, does have its own risks, of course. There are no guarantees, given such a pointer, that using it is safe - it is always necessary for the programmer to know, given the context of its use, that it is safe. Most programmers get this right nearly all of the time, of course. But when we get it wrong, our code goes horrendously bang, usually five minutes before a demo to the visiting CEO of your biggest client. I always tend to favour situations where the programming language picks up automatically on that kind of thing.
Signals and slots, in my opinion, have three major advantages over pointers:
Syntactically Neater. Signals & slots make for very readable code. You define your signals. You define your slots. You wire them up. That's it, and no surprises.
Inherently Robust. This is the neat part: when either end of a signal/slot connection is destroyed, the connection is automatically removed. It is therefore impossible to emit a signal and have it arrive at a class that has already been deleted. Best of all, you don't need to write any explicit cleanup code - if the caller goes out of scope, no problem. If the callee goes out of scope, no problem either.
Easier code reuse. Since the thing that has to type-agree is the signal and the slot, not the whole class interface, it is much easier to plug together disparate classes that had never initially been intended to work in that way. (Take it from me - I've done this in anger, and it really does make a difference)
I really like Qt. I think the people at TrollTech in Norway have done a brilliant job - their class library is far and away the best GUI library I've ever used (though the .NET framework actually comes remarkably close, but enough flame bait for one article).
When I was working at Trayport in London, a coworker there, James Slaughter, got me interested in the Boost libraries, and was largely responsible for reawakening my interest in C++ as a programming language with real promise. Thank you James, if you're reading this. (As an aside, I was previously getting very interested in Objective CaML, but that's for another article at another time).
When I evangelised Qt within James' earshot, he opined that Qt was pretty good, but he wasn't keen on the moc preprocessor. Rightly enough, moc has a rather bad rep for mangling attempts to use templates in Qt-enabled classes, and it's always irritating to need to hack build scripts to preprocess code. James did mention that type safe signal/slot functionality was possible in C++. I'd kind of thought about it myself, but hadn't initially realised that it really was possible to do without something like moc.
I did a bit of a net search, and it seemed that a couple of people had put together C++ signal/slot libraries. One was in the pipeline for adoption by Boost (still is, at the time of writing). Another is due to the GTK bunch, as part of the GTK's C++ wrapper. Both were a little heavyweight for my needs (writing MFC code whilst staying at least a bit sane), and neither looked like good candidates for a straightforward Windows port.
I chose to write my own instead. The rest is history.
You can download sigslot from the project downloads page.
Detailed documentation for sigslot is currently in LaTeX format. I haven't yet decided whether to convert this to HTML and post it here, or to laboriously glomm it into the documentation system at SourceForge. For the moment, you can get the PDF version from the link below. If you want the LaTeX source, or are feeling kind and generous and would like to contribute by doing this conversion, please email me.
PDF Documentation (sigslot.pdf, 73kb)
So far, sigslot has been tested on VC++ 6.0 on Win32, VC++ 7.0 (Unmanaged) on Win32, gcc under Cygwin and Intel C++ on Win32. It has also been used in some real code without problems. However, it is relatively new, so tread carefully and please make sure to report any bugs, fixes or feature requests to the project team. By preference, please do this through the bug tracking system at SourceForge.
The sigslot library has been placed in the public domain. This means that you are free to use it however you like.
The author takes no responsibility or liability of any kind for any use that you may make of this library.
If you screw up, it's your fault.
If the library screws up, you got it for free, so you should have tested it better - it's still your responsibility.