On Thu, 3 Oct 2002, Leopold Toetsch wrote:
> [1] a test looks for availability of mallinfo, and assumes then, that
> the underlying libc uses ptmalloc (which is derived from Lea malloc
> 2.6.4). For this case, no malloc.c is linked to parrot, saving ~80 KB.
> (This test could be omitted, to use any system's malloc, with probably
> horrible benchmark timings - or not - who knows)
In SVR3 (and probably in SVR4 and derivatives, and certainly in
Solaris) mallinfo() is often available through a special -lmalloc library.
I could also imagine some System-V-derived systems having it available
in the C library.
It's not specific to Lea's malloc.
But that's good! It means you can plug in your patch and test with an
even wider variety of different malloc()'s. Malloc()'s that have
mallinfo() also usually have mallopt(), which allows you to tune various
allocation parameters. Knowing typical allocation patterns in parrot, you
can possibly use those parameters to tune performance even better.
Similarly, trying out perl5's malloc (the Kingsley malloc, very heavily
modified by Ilya Z to better match perl's typical allocation patterns, and
quite tunable) might be a good idea.
--
Andy Dougherty [EMAIL PROTECTED]
