I'd also bring up another metric... which is also very important... how many lines of (readable documented!) code does it take to do something? How long does it take somebody to produce that code? (a language might be very powerful and terse, but really difficult to write, debug, or understand after you're away from it for a while). Do real world programs written the language have a higher rate of programming errors? (This is where things like strong typing can slow down the programmer, but getting an idea that something is not correct while still building the application, instead of when it is controlling a mission to Saturn... can be critical!). I've seen somewhere here already a plot of LOC vs. performance (IIRC), and Julia did very nicely...
Julia still needs work on debugging, and on letting a programmer maintain a separation between their abstraction and the implementation... but IMO, Julia has great bones, and compared to C/C++, as a performance geek, it really shines,as a language where I can get C/C++ performance writing *much* less and easier to understand code, and compared to Python, JavaScript, Lua, it shines because of it's performance. On Friday, May 15, 2015 at 8:24:39 AM UTC-4, Andreas Lobinger wrote: > > Hello colleague, > > On Friday, May 15, 2015 at 12:08:10 PM UTC+2, Páll Haraldsson wrote: >> >> >> Do I need to be worried? My guess is 0.4 might be faster and maybe >> already be faster in all the cases below (do you know?): >> > > i'm allowed to write that, because i already did my share in benchmarking > systems (20 years ago...): Benchmarks don't lie, but liars do benchmarks. > > The comparison of the runtime for the implementation of small synthetic > programming problems in julia to other languages is a nice exercise and > gives some numbers as indication. But the room for interpretation stops > with that. I am seriously doubting that you see the same proportion of > numbers if you go to non-synthetic benchmarks or real world programming > cases. In the real world the performance of a program is dependend on a lot > more things than choosing an implementation language (in my work 20 years > ago, i benchmarked different fortran77 compilers on exactly the same > system). Julia is a quite nice user friendly language, but the actual > performance is happening in the compiler and the optimization steps > provided by LLVM. > > >
