From: "Greg Stein" <[EMAIL PROTECTED]> Sent: Friday, October 19, 2001 7:18 AM
> I brought this up once before. I think it was Aaron that suggested an > "optimization" which changed a pcalloc to a palloc. I noted that doing > things like that are troublesome for long term maintenance. > > Bam. Empirical evidence here. Yup. > Changing pcalloc to palloc should only be done when we have specific > information that it is *really* helpful. Agreed... but there is a simple (safe!) variant; merged = apr_palloc() memcpy (merged, conf) ...fill in .new members... Everything is assuredly set to a legal value ;) Would someone take 5 minutes to look at the mmap code in apr_pools.c, and see if the apr_pool_lock can be easily ported to the mmap protection? Unix mpm's have their own directives... we wouldn't trip on them on windows. Right now, we call apr_pool_lock on p->conf ... for the life of the server. Before we tear it down, we unlock. Any mismerge or other sorts of bogusness are caught by this lock. Bill