On Sat, May 31, 2008 at 7:10 PM, Waldek Hebisch
> AFAICS Fedora problem is triggered by security measures: when
> I did (as a root):
> echo 0 > /proc/sys/kernel/randomize_va_space
> echo 0 > /proc/sys/kernel/exec-shield
> then build went fine.
I am a bit puzzled. Are you using the latest sources? I committed a
patch today (explained in another email) that removes the call to
si_gc() at the price of _not_ using the wrapper around dlopen() which
is shipped with the Boehm-Weiser garbage collector. This way, the
garbage collector is not deactivated when loading binary files and it
should not run out of memory.
Now, as for the two flags that you set up, if the problem is related
to memory fragmentation, then only the first line should be needed.
exec-shield, is related to making memory nonexecutable. If that was
causing the problem, then we would see some other signals coming out,
such as SIGBUS, I suppose.
Incidentally, I checked and the memory randomization feature is active
in my server (Ubuntu/64), but I must admit I have not yet completed a
full build of FriCAS.
I believe that memory randomization should not affect the garbage
collector. Unlike other systems, we do not allocate memory at a fixed
position of memory. The garbage collector only gives a hint, typically
a low address, and then waits for the operating system to allocate the
memory at the right position. Same goes for subsequent calls shrinking
or enlarging the allocation memory: they may well end up receiving
chunks of memory in other positions.
Facultad de Fisicas, Universidad Complutense,
Ciudad Universitaria s/n Madrid 28040 (Spain)