[haiku-development] Re: Haiku R1/alpha updating (was decisions)

On 2008-08-07 at 22:25:26 [+0200], Axel Dörfler <axeld@xxxxxxxxxxxxxxxx> 
wrote:
> Bruno Albuquerque <bga@xxxxxxxxxxxxx> wrote:
> > Axel Dörfler wrote:
> > > Maybe we should have a mechanism that prevents the system from
> > > reloading add-ons in the update case. At least we should test this
> > > before releasing any version of Haiku :-)
> > Yep, we need that. Specially because, at least right now, it tends to
> > crash the system. Try a jam install-haiku to the same partition where
> > you are running it from and you will see what I mean. :)
> 
> What could also cause this is if we are just rewriting the same file
> instead of replacing it. "cp" for example just does that - in this
> case, code that isn't currently paged in will see the new binary
> whenever it gets there. Linux has the same problem, btw.
> So we could either change our build system to properly replace the
> files (ie. by passing the --remove-destination option to it, in case it
> supports that), or let the VM handle this case gracefully (see MAP_COPY
> for more info: 
> http://www.ussg.iu.edu/hypermail/linux/kernel/0110.1/1506.html
> - in case you want to have a good read). Despite Linus' slight aversion
> against MAP_COPY, I think this would actually be the preferred method
> (but for executables; I agree it doesn't really make sense as a
> replacement for read()).

Agreed. I'd even say that MAP_COPY is superfluous, since MAP_PRIVATE should 
imply that behavior -- at least I think it's pretty inconsistent ATM.

CU, Ingo

Other related posts: