On Sun, Mar 26, 2006 at 03:10:31PM +0200, Leopold Toetsch wrote: > > On Mar 25, 2006, at 5:11, Patrick R.Michaud (via RT) wrote: > > >I'm running a fresh checkout of r12014. I wrote a simple test > >program (t.pir attached) that uses "dumper.pir" to dump the > >contents of one of PGE's internal variables. On my 64-bit > >environment (SUSE 10 Professional Linux) I get very strange > >output and the strings appear to have been corrupted. On > >my 32-bit environment, the output is exactly as I would > >expect it to be. > > I've tried that with r12014 on an amd64/linux box. Dump is the same as > with 32bit. There must be more than just running it on 64-bit. You > could further test: > > $ ./parrot -G t.pir >d # does it work with DOD/GC off? > $ ./parrot --gc-debug t.pir >d # is it worse? > $ ./parrot -D40 t.pir >d # fill I/N regs with garbage
With "-G" I get the correct output -- the other two fail the same as when no options are present. > You should also verify that you don't have some installed old copies of > parrot in your path. I never run "make install" on my system, so it's unlikely there are old copies lying around. But just to be sure, I removed all copies of parrot from my system and re-ran the tests and obtained the same results. I agree it looks like a GC bug, and not anything x86_64 specific. It appears my x86_64 system is just good at flushing those out. :-) Pm