It looks that you see many statuses in this period for a reason: I had a fairly light workload at work and I had time to think for some fine things to add to CR. In future I will not guarantee the same frequency and the same speed. The best way to make CR to develop fast, is to contribute ;)
So, what is done and note-worthy:
- type-id support and a trivial/crashing Virtual method calls. People wanting to have virtual method implementation, can go into the implementation and fix the remaining things (or still wait): make possible to run virtual functions with more than one parameter, to take in account of escape analysis, and things like this
- type closure is more robust: types are sorted for dependencies, and methods and types are discovered in the same time. It is possible right now to have more entry methods at least on the code level. This can make possible in future to support DLL/lib code generation or to make sure that statically are added all methods needed
- added state of escapeness of variables of unused, the variables in the past could be just: stack, pointer and smart-pointer. This is very important as it can detect (in future) that a function argument is not used, so the compiler can remove the usages of that call and making the call fast(er). More work will be definitely in this area.
- make an unified way to keys to identify methods: this is as important as type-id for me: every time when a function is called, a strange code was written to look if the method is a runtime method, or is a generic method, etc. Even worse was the comparison between functions as overloads. I will expect with this change that generics will be returned with the original prototype (like will be a clone for every static specialization, and no C++ templates are used).
Basically: this status means that there is a big base to be improved in CR. And the more fixes there are, the better you can expect your program to work. So try to look into the code and make simple fixes.
What I expect in future to work at:
- try to make a program wide optimization kind. In the past and all optimization passes of CR are just accessing one function as input, but some cases are better handled if the visibility of the optimizations is global. This optimization could remove a virtual call if it knows that there is just one implementation globally.
- look into the Qt's Core runtime. I worked in my past with Qt 4.x (x being a small number) and my experience was very good. Qt could be a great base to offer a good implementation for some operations like: directory navigation, string encodings, and things like that. It would be really ugly to try to implement them in C++: there is no standard multiplatform implementation I'm aware to handle them so well
No comments:
Post a Comment