Jeff Johnson
Fri, 08 Jun 2007 15:09:13 -0700
On Jun 8, 2007, at 6:01 PM, Michael Jennings wrote:
On Friday, 08 June 2007, at 17:55:04 (-0400), Jeff Johnson wrote:You mean like dpkg does?I don't know how dpkg does its stuff. :)Nothing. But performance, not locking, is the key issue for choosing a representation of the installed package set.What about as an alternative representation, kinda like what you did, only on a grander (and crazier scale)? Imagine rpmd, a user-space daemon which maintains a pseudo-filesystem (a la /proc or /sys) interface to the rpmdb. Client programs could use rpmdb via rpmlib or the pseudofilesystem, either way. The former for performance, the latter for portability and ease of development/rapid prototyping.
Direct access to an rpmdb requires locking, and shared read locks are a DoS for exclusive write locks, and vice versa, no matter what. So custom automagic exports avoid the locking issues associated with direct access to an rpmdb. Note that other locking issues are possibly introduced because the export/update may have issues with whatever some application desires, but at least the locking is finer grained, with less contention, than having every application competing for rpmdb access.
And you thought your del.icio.us idea was nutty. ;-)
;-) So what's nutty about using del.icio.us instead of rpmfind.net? 73 de Jeff ______________________________________________________________________ RPM Package Manager http://rpm5.org Developer Communication List rpm-devel@rpm5.org