Re: gnome-metadata

[Havoc Pennington <hp redhat com>]

On Sun, 29 Aug 1999, Elliot Lee wrote:
> 
> My main point is that guessing & wishing that it won't slow things down is
> not enough. We need to make a conscious effort to verify that GNOME is
> going to be fast, so that it works nicely on those pokey P5-90's that
> 

Randomly avoiding features is no better than randomly chopping out
features or randomly obfuscating code in the name of optimization.

We have to get good profiling tools and use them. Period. That is the only
way we will get meaningful speedups.

Even a good gprof effort might be worthwhile, though it means a really
painful rebuild of all packages (starting with libc, really...) - we could
do this on a test box.

What happened to that Gnumeric profile that revealed the Imlib problem?
What were time-sucks #2 and #3 after the Imlib color snafu?

If we're going to guess at random slowdowns I suspect the sheer size of
our libraries is a big part of the problem; demand-loading and
demand-initialization might help a lot to reduce the number of pages that
get hit on startup, and we could do that as part of a new GnomeApplication
object.

We can actually get a big "speedup" by just getting application windows on
the screen ASAP, and postponing everything else until just after the user
sees the window. A nice desktop-wide busy cursor couldn't hurt either.

Havoc