🚀 go-pugleaf

>>>>> "Rickard" == Rickard Westman <ricwe@ida.liu.se> writes:
    Rickard> The sad truth is that this program, compiled with GCC
    Rickard> 2.7.0 (-O3), blows the stack when tested with a number
    Rickard> large enough.  Tail calls within a single function seems
    Rickard> to be optimized, though, but this limitation is really
    Rickard> too restrictive for people who want to program in a
    Rickard> functional style (or translate functional programs to C).

Hmm - I never bothered to try mutually (tail)recursive functions.  I'm
not really surprised that Gcc doesn't handle the calls well.

    Rickard> On the other hand, Suns unbundled C compiler (v3.0.1)
    Rickard> seems to handle this example well, as well as mutual
    Rickard> recursion between different compilation units.  The DEC
    Rickard> OSF/1 compiler (v3.11 on an Alpha) also does OK, but only
    Rickard> within a single compilation unit.  More data points?

I tried this test with both the HP C compiler (the development
compiler, with most patches) and GCC 2.6.3.  Both seemed to run out of
gas at about the same rate.  So I think that HP (and FSF) might stand
to improve their handling of mutual (tail) recursion.

    Rickard> All in all, I don't think it's wise to assume that tail
    Rickard> call optimization is done by the typical C compiler.

I stand corrected.  So why haven't more C compiler folks added this
useful optimization?

Kelly