Front page | perl.perl5.porters |
Postings from January 2006
Nicholas Clark <nick@ccl4.org> writes: >On Wed, Jan 25, 2006 at 10:52:36PM +0200, Jarkko Hietaniemi wrote: >> nm -P perl (in Linux, -P being the "POSIX" option) tells me that >> (in x86) the PerlIO_fd_refcnt is 0x2000 or 8 kilobytes. >> >> perlio.c:2227:int PerlIO_fd_refcnt[PERLIO_MAX_REFCOUNTABLE_FD]; >> >> perlio.c:59:#define PERLIO_MAX_REFCOUNTABLE_FD 2048 >> >> Hmm. An array of 2048 ints, representing 2048 fds. Hmm. >> A smaller array for first, say, 256 fds, then a hash for >> the rest, or just a hash for all? I can't off-hand believe >> that refcounting fds can require blazing speeds. Or are hashes >> so big that our space savings just went south? (The array being >> a global makes me shiver, too.) > >As far as I can tell, the rest of the code won't notice if that array is >malloc()ed and resized as needed. From what I remember, that would add no >significant complexity to the code. It is global so that all threads and pseudo-forks can see it. It isn't malloc-ed just because the Win32/PerlHost "per-thread" malloc got in the way. It is possible that the global malloc pool could be used. > >Nicholas ClarkThread Previous | Thread Next