Hi, under the Linux version the compiler optimizes the toy example correctly; unfortunately, it fails to optimize the real code that depends on tail merging. I will, of course, unroll tail recursion manually -- this is tedious but doable. But it's a pity to get stumbled on an issue like this: all other lisp implementations available to me (CMUCL, Allegro CL, OpenMCL, CLISP, SBCL) optimize this code into a loop without a problem; and that's what I would expect to happen. David
On Mon, 12 Dec 2005 22:59:05 +0400, David Tolpin <dvd@davidashen.net> wrote: > under the Linux version the compiler optimizes the toy example > correctly; unfortunately, it fails to optimize the real code that > depends on tail merging. I will, of course, unroll tail recursion > manually -- this is tedious but doable. > > But it's a pity to get stumbled on an issue like this: all other > lisp implementations available to me (CMUCL, Allegro CL, OpenMCL, > CLISP, SBCL) optimize this code into a loop without a problem; and > that's what I would expect to happen. I suggest that you send a bug report to LispWorks support. As long as you don't show the code to anyone (which is understandable for proprietary stuff) this is hard to comment on. Maybe it's a bug? Maybe /you/ made a mistake? Who knows?
On Tue, 13 Dec 2005 00:19:24 +0400, David Tolpin <dvd@davidashen.net> wrote: > the code is not secret. It just takes time to make it suitable for a > bug report. I expect the code below to be optimized into a chain of > tail calls. It meets the expectation with CMUCL, Allegro Common > Lisp, OpenMCL, and CLISP. However, on both Linux and Mac OS X, > Lispworks 4.4.6, it causes stack overflow with any compiler > settings. You can try it. Run (test-runes 10000) to see the effect. Yes, I can confirm that I also see the stack overflow on LWW 4.4.6 and I don't see it with other Lisp implementations I have here. > If you know how to make Lispworks compile it, I would be grateful. I was just playing around and I noticed that replacing the second LABELS with FLET did the trick for me - now the stack overflow is away, even without declarations. I'd still say you should file this as a bug report. Cheers, Edi.