A generic hash table [LWN.net]

A generic hash table

Posted Aug 17, 2012 9:02 UTC (Fri) by jzbiciak (✭ supporter ✭, #5246) [Link]

Actually... any use of that max that returns bool would have rather unusual semantics with that return type. Almost reminds me of PHP. ;-)

As far as the whole C / C++ debate goes: Up until a few years ago I was an ardent C proponent and avoided C++. Since then, I've really given C++ a chance, and I now tend to prefer it for new projects. I still use C for quick one-offs, but for anything larger I use C++.

For one thing, C++ compilers are much better now than when I first looked at them ~15 years ago. You can throw some seriously complicated template messes at a modern compiler and it'll sort it out and give you rather tight code more often than not. 5 to 10 years ago, good code in that circumstance was more the exception than the rule. Bringing STL into the language in the 90s, plus the rise of Boost and template metaprogramming really forced compilers to up their game here.

I remember around 10 years ago playing with the "Stepanov" benchmark. It's a benchmark that measures a compiler's ability to deal with abstraction in a C++ program by wrapping a simple addition with a double in increasing amounts of C++ abstraction. G++ performed rather poorly at the time (incurring a 10x - 15x slowdown at the worst, IIRC) and our own compiler at work showing a 50x slowdown. Now I believe both compilers show no slowdown. They are able to flatten away the abstractions. It's beautiful.

For the problems I'm tackling in my day job, I really appreciate the compositional style templates offer me. Too often I find myself needing "an X with detail Y". In C, I'd have to code up a separate function that combined the two, or a macro that instantiated X subbing in a Y where necessary. In C++, it's just X<Y>, if X is a suitable template. I get type safety and an awful lot of compile-time computation and help.

This can be really wild but useful at the limit. I have a truly compelling example I'd love to share, but since it's my day job and it's proprietary technology, I don't know just how much I can share. I will say that the code I generate relies on a ton of compile-time computation to generate code that's sufficiently efficient that I don't mind running on design simulations of our chips that run at speeds on the order of 100Hz, plus/minus an order of magnitude. You read that right: anywhere from 10 to 1000 CPU cycles per second.

(Note: I've completely ignored error messages. C error messages are rarely obtuse and usually easily grokked. C++ with multiple nested templates is another ball of yarn altogether... They generally suck, and I'm with you if you think they're an abomination.)