Re: Fix for "agressive" memory segmentation

On Thu, 2006-07-06 at 21:32 +0200, Philip Van Hoof wrote:
> On Thu, 2006-07-06 at 15:18 -0400, Jeffrey Stedfast wrote:
> > For some strange reason I thought the pstring stuff already did that,
> > oops. I guess I was thinking of similar code I wrote a few years back
> > for another project...
> > 
> > This patch does it the way I had done it in another project of mine
> 
> Yours looks a little bit more clean in naming and stuff like that. It
> probably does more or less the same? So I'd say commit one of the two?

yea, it's basically the same as what you did

> 
> Mine made some significant differences when running it with valgrind. It
> was also a little bit faster in cachegrind (probably because there's
> less malloc()'s and free()'s happening).
> 
> I'm still committed to the mmap() idea. Although I don't know for sure
> keeping folder-count amount of file descriptors open is a very good
> idea. I know the disk-summary branch would help a lot here. Regretfully
> that one isn't yet finished ;-).
> 
> However. In tinymail I open and close folders much more quickly (each
> time a folder becomes inactive, I close it). Therefore I'm almost
> certain that for tinmails case the mmap() idea is going to significantly
> improve the situation for larger folders. I might introduce a compil-
> ation switch at the configure script.
> 
> I wonder, would such a patch (in case it's clean and doesn't harm
> Evolution more than that Evolution would gain from it yadi yada) get
> upstream? The patch would definitely bump the version number of the
> summary file from 13 to 14. It would add a '\0' at the end of each
> string in the file, and it will add one to the length-bytes in front of
> the strings .. also in the file (which is forward compatible, so going
> back to an older Evolution version will work with the same summary
> files).

it's not that invasive, I guess... so I don't think I'd mind.

> 
> 
-- 
Jeffrey Stedfast
Evolution Hacker - Novell, Inc.
[email protected]  - www.novell.com