> And I certainly don't think the Trolltech implementation will outperform > either gcc's or VC++'s implementation. Performance is not exactly a > highly visible target in Qt development...
I don't think it will outperform, but it should be on par with. It's really not rocket science, at least that's what I gather by looking at libstdc++ 3.3.2 code. It's mainly about playing ball with that particular compiler version's optimizations (i.e. taking max advantage of them). Surely Qt probably won't be as version-specific as stdlibc++ is, but still we should expect reasonable performance out of it. > > And Tulip implements some concepts that stl does not (e.g. Java-style > > iterators) -- those concepts are easier to use for those who are not very > > comfortable with STL, and even people intimately familiar with STL might > > find them more to their liking. > > Which could be eaily implemented as add-on to standard containers. > Completely non-intrusive. But hey, Trolls already did that. And the fact that it's for their containers shouldn't be of much difference unless someone does actual profiling and shows that it really performs badly. > > Whether it's STL or Tulip, it comes preassembled for you :) I.e. you, as > > the end user, does not have to reinvent the wheel. There's nothing great > > about STL apart from the fact that you don't have to code it. Most widely > > used implementations of STL are quite mediocre IIRC, performance-wise at > > least. > > That's five year old wisdom, I am afraid. Think about how many people still use VC6, don't have a clue that SP6 for it even exists, and who didn't hear of dinkumware's updates to VC6 STL either. [That's people who can't google either, but that's a side issue] There are many big software projects that are being maintained on VC6 out there. For some projects switching compilers is not a lighthearted decision -- in any nontrivial project, a new compiler version is very likely to expose bugs in the code, or even introduce bugs due to various creepy reasons that you'd rather not know about. > I just think, most tulip work is a waste of resources. Do you remember how bad it was at one point in time when Lars started "porting" LyX to hardcore C++? All the compiler problems and STL mis/under-implementations, etc? Code bloat due to under-optimizations and compiler bugs? And so on? Now, think that Qt has to support some really odd/old compilers, and they use their own containers mainly to level the playing field across all systems. I.e. so that fix-that-compiler-bug code is limited to their containers, and not strewn across everywhere. > And I actually > wonder how Trolltech will meet its 4.0 deadline given the amount of meat > that's missing in 4.0tp2 Did they even specify any hard deadlines? Cheers, Kuba Ober
