Xhelp code on launchpad ready

[Eero Tamminen]

Hi,

On Wednesday 08 August 2007, Cody A.W. Somerville wrote:
> > > The prominence of the Xubuntu help on the desktop is an issue.
> > > However, the issue we are trying to address is that people do not
> > > want to view documentation in Firefox because it takes too long to
> > > load. If they already have firefox open, it isn't as bad but I can
> > > imagine there will be times when a user wants help on a topic that
> > > does not involve the internet/web-browser and as a consequence simply
> > > do not have firefox open.
> >
> > If Firefox is already open, starting something like Yelp is definately
> > slower than opening a new window in Firefox for the help page and as
> > it's a new UI process, it most likely takes more memory from the system
> > too. Most users have browser open all the time.
>
> Could you please provide some hard evidence to support this.

On my 1Ghz 1/2GB desktop machine running Kubuntu, starting first instance
of Firefox takes ~10 secs, closing that and restarting Firefox takes ~3
seconds, whereas calling Firefox when there's already Firefox window open
takes ~1 second to open the given (plain) HTML page into another tab
(whether it's window or new tab is an FF preference setting).

Starting Yelp for the first time takes ~4 seconds, second time takes ~3 
seconds and starting yelp one more time while the previous instance is open
takes again ~3 secs.  I.e. Yelp is slower than Firefox when there's already
an instance of the same program running.

But please test it yourself, it takes only a couple of minutes.


I think Xhelp could be quite a bit faster than Yelp, but because firefox
just tells the existing instance (if firefox is already running) to load
the page, it's pretty fast and as after that the new instance exits, you
don't have the UI process memory overhead.  So using FF or any other
browser that does something similar is not *that* bad.

If users don't normally have browser running, then the FF first startup 
speed is a real problem and I would recommend using something faster
than a browser or having some good user feedback on its startup.


Btw. If you're interested about details of what FF does, try something like:
LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/lib/mozilla-firefox/ 
strace /usr/lib/firefox/firefox-bin index.html 2> ff.log

(or use ltrace instead of strace.  I called above firefox-bin directly
instead of the firefox wrapper script because ltrace cannot trace
over clone() syscall.)


> > In these cases, the wait time is simply unbearable on lower-end
> >
> > > systems. Often users are unaware that they are launching firefox
> > > and are confused by the immediate performance hit they feel.
> >
> > This is a UI feedback issue.  I'm (sometimes) using Epiphany on P166
> > (Epiphany is slightly faster to start and uses slightly less memory
> > than Firefox) with IceWM and because it has a CPU meter, there's
> > feedback that something actually happening (besides modem noise :-))
> > when it's launched.


	- Eero

PS. I don't really understand why Xubuntu tries so hard to avoid Gnome
dependencies in applications used already on Ubuntu that are not always
running. Could this be explained a bit better somewhere? Or is it there and
I've just missed it?

(Reasonable) Gnome dependencies don't take that much space on harddisk
(or even on install-CD) because they are shared between all the
applications. IMHO memory footprint and startup speed are more important,
but that should be measured, not guessed based on package dependencies.
This could be gotten e.g. by calculating from /proc/meminfo
buffers+cached+free after booting to the default Xubuntu desktop,
then after starting the application, after doing some normal use-cases with
that application and after closing it.

Additional dependencies on the Desktop processes itself are bad because it
increases the memory use etc, but for applications that user starts
separately and quits after using them I don't see why it would be an issue.  
Unless there are good (functionality, support...) lighter alternatives and 
the difference is as large as it's between e.g. OO writer and Abiword...