On Wednesday 30 April 2008 01:04:26 John Femiani wrote:
* When I'm testing different compiler settings trying to get another few percent of performance, I don't have to compile another Boost version for each compiler setting.
I think you have to rebuild the compile_inplace.cpp file each time you want to test a different compiler setting. I don't find an advantage to what you advocate here.
I think that his approach would be optional, in addition to the existing bjam setup, right?
Yes.
Why is branching necessary though? Isn't the proposal just to add a couple new cpp files? will it require modifying the existing ones?
There are a few modifications necessary, in particular to get win32 import/export and autolinking right and also some symbol clashes via anonymous namespaces. I already made prototype implementations (it's actually rather easy) both for win32 and Linux, but I haven't tested the exact same code under both OSs. Now, what I'm planning to do first is to branch the current stable version and implement it there. The simple reason is that if something breaks, I can be pretty sure that I broke it. Further, it allows others to check out just those modifications without otherwise downloading the current development trunk. Once I get that ready, I'd merge (or let someone else merge) the changes into the trunk. That's all - the only reason is to keep development a bit separate in order not to break things. Uli