I would say that how good the programmer is (how smart he or she is regarding choice of data structures and algorithms) and generally how much thought is put into the system design will have a much, much greater impact on the performance of the application than choosing Java or C++. Personally, for me it's Java, not so much due to the language, but due to the Eclipse IDE. I've found Eclipse with Java to be much more productive than C# or C++ in Visual Studio, some may disagree. In any case, I'd start with Java and move to C++ if you really need more power. Also, if you go with C#, then you've just locked your system into Windows so better find a good virus scanner and an extra core or two to run it.
Depends on what you're trying to achieve. Among the OO languages,
0. If you're trying to write crap, use Python.
1. If you're trying to write crappy GUIs, use Java. As has been pointed out, the direction of Java is bad as it is led by Oracle, and there are too many protocols for doing things.
2. If you're trying to spend weeks rewriting makefiles and figuring out dependencies for libraries, use C++.
3. If you're trying to spend weeks writing bad code to trick the JIT and using poor libraries or interop bindings rather than native APIs for larger libraries, use C#.
Lastly, when you're picking a good developer, don't ask him how many years of development experience he has. It's not the number of years and lines of code that he has written, but the contributions to the projects that he has made. I can confess that I've rejected someone with 15 years of development experience, several recent years maintaining the HPC stack at Yahoo Finance using relevant technologies, Tibco etc. (they're surprisingly developed - I think Yahoo owns the 3rd most servers after Google and Microsoft). I've also accepted someone with 4 years of development experience because they wrote the entire test and staging environment for a major product at Google, and I made the right choice because the person made me take interest in decisions that I was never interested about. So,
0. Find someone who will make you take interest in things you've never considered, not someone who will just listen and create what you described to them, even if it's crap. (This is for a web development stack, but imagine the same scenario with whatever you're developing: How a Web Design Goes Straight to Hell - The Oatmeal)
1. Generally, I find it easier to filter out bad developers based on the combination of languages that they've learned. A person who is experienced with 4 different OO languages is often not as good as the guy who learned Prolog and Scheme. I don't mind hiring the latter so he spends some time learning the syntax of a language that we use for inhouse development, but there are troves of the former.
The following 3 users say Thank You to artemiso for this post:
( I would especially agree with the end of your message. I am not a professional programmer, but I am rather fluent in Java, and progressing in C++. I indeed practiced a lot Scheme and Prolog... for my pleasure, and am still doing so. This complete shift of paradigm induced by the 2 languages helped me to have more ideas in... OO languages. )
The following user says Thank You to Nicolas11 for this post:
I agree with all your other points, but I find that I spend more time focusing on other aspects of the trading system besides the look and feel of the GUI. The Swing framework could be improved, but it's sufficient for what I am doing. And I have yet to find a better open source charting framework than JFreeChart for any language.
Java development is mostly driven by the Java Community Process. Oracle just steps in when they think they can sue other companies for patent infringement. I would rather they not do that as I think they are shooting themselves in the foot, but they have open sourced most if not all of the JDK through the OpenJDK effort.
The following user says Thank You to Crow for this post:
1. GUI can be very important. A majority of the time you might spend as a quant in the sell-side is simply writing and maintaining a GUI for some valuation model that someone smarter than you had written, so I have come to acquire respect for good GUI. Another thing is that I come from a physics background, and most physicists have Feynman-envy; meaning, we care a lot about elegant visualization because it avails better intuition of the problem. Finally, if you have a client-facing job, a demonstration of the GUI is definitely several times more powerful than a thick PPM stack.
2. For me, the problem with Java/C# is that the memory management is not deterministic, and this an incurable flaw by design of its advertised benefit. This is not necessarily a problem, depending on what you're up to.
3. C++ isn't the only alternative and I recommend people choose their development language based on their familiarity and business needs, not to pick what is most popular. If you have to write to a specific API or platform, or count a lot on the typical time series or statistical stuff, then I suppose you have little choice, and might even have to resort to polyglotism. My team had some flexibility as we do very little statistical modeling, and even if we do, we preferred geometric analogs - so we had to rely very little on the traditional charting libraries. (We treat functors and categories as first class citizens - not time series, so we've become huge fans of MetaPost for our visualizations.) And so, we dropped everything in favor of mostly Fortran, and we're extremely happy with the move.
The following user says Thank You to artemiso for this post:
You may choose from c++, java, python - to name a few, but understand that domain knowledge is infinitely more important than language of implementation. If your concern is price of hire - imho, you haven't done your homework thoroughly enough. Someone who's able to deliver such a platform will command top dollars no matter what the language of implementation is. They should understand real-time programming, scalability issues, concurrency issues, lifecycle management of different types of orders; depending on what you want to achieve they also may need to be good at math (on the advanced side of higher education). There are many other points that I won't cover. Point is, if you factor them all in, your question – "what's the best language?" - is naive.
There is but one exception I can think of. If you plan to scalp exclusively, C++ is the way to go, along with having a datacenter right next to exchange. It has nothing to do with price point though, just a technical requirement.
Everything I said applies to one-two persons shop. The situation is different if you are trying to make a business case and build a team. But then you can't ask your question in such a generic form; you'd need to put together specific requirements and break it down into multiple questions that would speak in terms of man-hours.