On Fri, 2004-05-28 at 10:28, Mark Phippard wrote:
> " As for the pool parameter, the apr_poll() function on the
> iSeries has been enhanced to allow for more descriptors than the
> normal default of 200 defined with FD_SET size. In order to
> support this, the readset, writeset, and exceptset passed to the
> sockets select() function need to be allocated before calling
> select(). The best way to get the space allocated is to allocate
> it from a pool. Using malloc() is a performance problem.
What impressively twisted logic. I wonder if they bothered to measure
this "performance problem" before sacrificing interface compatibility to
the God of cycle-counting. (Even if they could measure a performance
problem, they should have created a new, differently-named variant with
a pool parameter, and modified httpd to call it instead of apr_poll().)
> The only place that this function is used within Subversion is in file
> marshal.c in libsvn_ra_svn. Here is my current patch, but I think clearly
> based on IBM's description this is not the correct pool to be passing:
Nah, I think it should be fine. If there are multiple threads using the
marshal.c routines, they will not be passing the same pools.
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Fri May 28 17:34:13 2004