… the gcroot bug in the llvm compiler
… cases. # first test round with nofib on variant 8 gives the following results: all working tests trigger serveral garbage collections and a lot of memory is cleaned after GC Always works: - TAK Works: - digits-of-e1, works with args up to 250 - wheel-sieve2, works with args up to 4000 with larger args a malloc panic is triggered (?) LLVM-Bug: the other test cases (primes, queens) trigger a bug in the LLVM compiler. the LLVM compiler crashes on an failed assertion caused by removing the GC intrinsics. according to llvm devs its a bug and they will try to fix it.
…as root. - GC is working now for my test case! but needs some more testing with nofib suite.
…e to update pointer during a GC
Now all nofib programs survive the new garbage collector (C backend). The new GC is always faster than Boehm, for some programs even much faster. More programs now survive the largest input (exp3_8 and gen-regexps). Only the primes program still crashes on the largest input. (Also, double is still not implemented, so "integrate", "paraffins" and "fib-double" do not compile).