[developers] Pet/ECL memory problems

[Francis Bond]

G'day,

> do you still have this memory problems? If so, this may be the
> consequence of a change in the chunk-allocator mmap code. I tried to
> circumvent the problem that cheap dies because of trying to mmap at
> hardcoded addresses and failing to do so.

Yes, we still have the problems.

> It now moves the address down (or up, depending on the machine
> architecture) by some amount and then retries. Unfortunately, this
> does not seem to work properly either.
>
> Since the addresses specified for the mmap are only <hints> for the
> underlying system, there is no guarantee at all that this scheme
> works. If you don't get to run cheap at all on some machine, disable
> the mmap feature during ./configure (at the cost of a bit of
> efficiency).

OK.  I will try that (that is, recompile with #define HAVE_MMAP 0)

> If i knew of a way to get rid of the mmap scheme without to much
> performance loss, i would immediately throw mmap out.
>
> Sorry about that,

Thanks for the possible solution.

--
Francis Bond  <www.kecl.ntt.co.jp/icl/mtg/members/bond/>
NTT Communication Science Laboratories | Machine Translation Research Group