[prev in list] [next in list] [prev in thread] [next in thread] 

List:       kde-devel
Subject:    Re: KApplicationServer?
From:       "Friedrich W. H. Kossebau" <Friedrich.W.H () Kossebau ! de>
Date:       2002-08-25 22:10:41
[Download RAW message or body]

Tim Jansen wrote:
> 
> On Sunday 25 August 2002 18:52, Friedrich W. H. Kossebau wrote:
> > Each app controls up to x "windows" (read documents/files/...) in a
> > process. For some further "windows" a new process is started. This way
> > the amount of lost "windows" is bordered.
> 
> If I start Konqui processes, I want to "isolate" things. For example a process
> for unknown sites, or a separate process for KDocs.

Alright. Looks familiar ;) 
But there is room for improvements as you cannot say by a visual flag or
like which windows are run by the same process (and by this share some
settings!). 
No idea how this might help joe user (and not confuse ;). Okay, I'll
keep this in my mind until I have worked out something and then come
back perhaps.
 
> > To ask more precisely:
> > What is faster, starting a new executable or forking a already running
> > instance? After a fork I expect all the data of the forked process have
> > to be resetted to get the state of a fresh started app, right?
> 
> Forking is much faster. The kernel just copies the page entries for the new
> process (memory will only be copied when one of the processes writes). All
> file descriptors will be shared. But this is what makes it so difficult: you
> dont want the processes to share file descriptors. Or the X11 connection. Or
> the DCOP handle.

So forking is faster only for the start of the process, not the
resetting involved? But as you wrote below the burden might be the
programming...

> It is probably possible to solve these problems, but it is so much work that
> you better wait for glibc 2.3 which will hopefully solve the linking problem
> in a clean way. The time for loading and relinking an application with its
> shared libs should be negligible then...

I see. Thank you for this information, some more pieces to my puzzle of
understanding the way everything works and is going :)

But still I wonder why there cannot be a reuse of the readonly parts of
an executable already loaded in memory which I understood machine
instructions to be, too (section ".text", isn't it?). Sorry, I once read
something about it, but forgot again so I have no clear picture. Are
there machine instructions that might be readwrite, so a new relinking
is needed for the same executable? Wouldn't it be enough to reload those
sections that are readwrite? But they may have to relinked to the
machine instructions part again somehow... well...

Any place on the net where I can learn about this? Any good book
(preferrable in german) about this? What topic is this? (where to search
in the library?)
 
> bye..

Bye

Friedrich
 
>> Visit http://mail.kde.org/mailman/listinfo/kde-devel#unsub to unsubscribe <<
[prev in list] [next in list] [prev in thread] [next in thread]