memory.profile() messes up the vector heap on Alpha/Linux?

Peter Dalgaard BSA
26 Nov 1999 22:27:54 +0100

Naoki Takebayashi <> writes:

> Hello,
> I have been trying to debug a problem with R-0.90.0 (this bug was in
> 0.65.1, too). The following code results in seg faults.  It doesn't
> seg-fault on Linux/intel.
> > memory.profile()
> > gc()
> As long as I don't execute memory.profile(), there is no problem with
> garbage collection.  So I think that memory.profile() screws up the
> heap in some way.
> When it seg-faults, it dies in the function compactPhase() in memory.c,
>     while (vfrom < R_VTop) {
>         s = BACKPOINTER(*vfrom);
>         switch (TYPEOF(s)) {    /* get size in bytes */  /* <- dies here */
> it segfaults becuase s becomes a pointer to 0x0, when the while loop is
> almost finishing, i.e., vfrom is close to R_VTop (vfrom = 0x20001840ef8
> and R_VTop = 0x20001841198).
> I tried to see what's going on in do_memoryprofile() in memory.c by
> stepping with gdb, but I couldn't see what was going wrong (well, to be
> honest, I didn't fully understand what was going on).
> I thought someone may have some insights about the problem or strategy to
> debug this problem.  I wanted to set some kind of watch point, so the 
> gdb breaks when the address 0x20001840ef8 is changed.  But I'm not
> really good at gdb yet, so I'm wondering if someone can help.

The gdb command would be 'watch *0x20001840ef8' (it's important to use
the address as a number so that one gets a hardware watchpoint), but
are you sure that's the right address to watch? I'd try the address of
s itself ('p &s' to print it). Also, what happens if you turn
gctorture() on just before the call to memory.profile? Usually slows
things down horribly, but causes a crash much closer to the actual

Happy hunting...

   O__  ---- Peter Dalgaard             Blegdamsvej 3  
  c/ /'_ --- Dept. of Biostatistics     2200 Cph. N   
 (*) \(*) -- University of Copenhagen   Denmark      Ph: (+45) 35327918
~~~~~~~~~~ - (             FAX: (+45) 35327907
r-devel mailing list -- Read
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: