On Fri, 2006-07-07 at 08:26 +0000, Philip Van Hoof wrote: > It's also that I recently learned about posix_madvise for my camel- > folder-summary.c mmap() idea. The only way, afaics, to make it > possible > to at the right time invoke the right posix_madvise on the addr > pointer > of the mmap, is to expose a new API to the CamelFolder: > > -> the behaviour of the program in relation to the addr > pointer of the > mmap() would change from sequential to random (this is what > posix_madvise allows you to tweak) when for example sorting happens. > It's, however, ui code that knows when sorting is happening or is > going > to happen. Every little change to a message, like marking it read/unread/important/flagging etc will result in mmap'ed write, which in turn has overhead on performance, may it be sorting or other message-list operations.
mmap() in camel-folder-summary.c is not acceptable. A proposal that EXTENDS the current architecture to support small-devices ALSO, will be more interesting. The professional way of achieving this is to extend camel_folder_summary into say, camel_mmapped_folder_summary and camel_folders use either camel_folder_summary or camel_mmaped_folder_summary based on the options/flags set by the application. V. Varadhan Novell, Inc. Software for the Open Enterprise™ http://www.novell.com _______________________________________________ Evolution-hackers mailing list [email protected] http://mail.gnome.org/mailman/listinfo/evolution-hackers
