I've been playing with heavily templated C++ code (Boost MPL) over the last weeks and am getting various compiler ICEs. While some of these may be actual llvm bugs I think a number are inherited from the somewhat outdated C++ frontend (I've personally reported a number of g++ bugs in the gcc bug database which are fixed in the meantime). While grepping over the llvm-gcc sources it seems that not too many files are actually touched by llvm, and when they are they are usually nicely tagged, so I'm asking if a re-merge with the current 3.4 CVS branch seems resonable. I'm quite aware that such a merge is boring, cumbersome and error prone, but with some fresh energy and some modern merge tools it should not be too hard. The llvm test suite should catch most merge problems, and the reward would be a huge number of gcc fixes (estimating several hundred of bugs, though llvm is not affected by all of them). Comments ? Thanks, Markus
That would be wonderful, patches gladly accepted!! -Chris
I'm changing this to be a meta bug. All problems that are blocked by our current llvm-gcc are going to be marked as depending on this bug. -Chris
llvm-gcc3 is dead