Sorry for being overly vague with the term "good".
I guess I would break it down to 3 categories.
1. Solid understanding of CS concepts. You don't have to build a Red-Black tree in the interview (ugh) but you should know properties of self-balancing trees and when to use them.
2. Effective use of C++ . Not looking for C++ trivia here (or someone who memorized the C++ lite FAQ). But things like good practices when it comes to memory management, using templates, virtual destructors, functors (and operating / passing functions).
3. Good idea how things work under the hood. So concepts like atomic variable (why and when to use it), bit-packing data, avoiding branching in hot code paths, and allocation strategies (memory pools).
Am I asking too much? I've found out that folks who have a solid grasp of CS, language and how things work under the hood can reason more complex things they don't know (yet).
We're using C++ to build a relational database for storing lots of multi-dimensional time series data. We're leveraging Google's supersonic for execution (https://code.google.com/p/supersonic/). The other important pieces (like the IO subsystem, indexing, query planing, sharing & routing) we've built ourselves.
The problem is that there is a wide range of different coding styles, between "C with sanity" and "Boost.Spirit". You seem to lean more towards the C side, but that's not clear from the "C++ expert" label alone.
I guess I would break it down to 3 categories. 1. Solid understanding of CS concepts. You don't have to build a Red-Black tree in the interview (ugh) but you should know properties of self-balancing trees and when to use them.
2. Effective use of C++ . Not looking for C++ trivia here (or someone who memorized the C++ lite FAQ). But things like good practices when it comes to memory management, using templates, virtual destructors, functors (and operating / passing functions).
3. Good idea how things work under the hood. So concepts like atomic variable (why and when to use it), bit-packing data, avoiding branching in hot code paths, and allocation strategies (memory pools).
Am I asking too much? I've found out that folks who have a solid grasp of CS, language and how things work under the hood can reason more complex things they don't know (yet).
We're using C++ to build a relational database for storing lots of multi-dimensional time series data. We're leveraging Google's supersonic for execution (https://code.google.com/p/supersonic/). The other important pieces (like the IO subsystem, indexing, query planing, sharing & routing) we've built ourselves.