> 
> > The reason for OrderedCollections is so that text can be
> > *built* quickly -- this is important in processing HTML.
> 
> Yes, I am well aware of this - and I checked the implications beforehand.
> You only pay a (minor) price when a new run begins, in all other cases the
> tweaks make it much faster than before. Thus, if your runs are not extremely
> short you'll win in the big picture.

This case happens a LOT for formatting HTML.  Every link, bold, italics,
indentation, and font-size change all start a new set of text
attributes.  It's really a downer -- Squeak can handle batch execution
of anything by adding a primitive for it, but Html pages are made up of
lots of little segements that, with my lack of cleverness, ended up
being processed one by one.





> 
> > Or at least, it used to be. Right now it seems to make a 10% slowdown
> > for the main Swiki page, so maybe it's not a terribly big
> > deal.
> 
> 10% slowdown?! Sounds strange. No matter how hard I try I'm not seeing a
> slowdown - in fact, I'm seeing a slight speedup (in the range of 3%). And
> that's certainly not amazing considering the above.
> 

What can I say?  Your download didn't fail, did it, leaving you
formatting an empty page?  Here are my timings, this time with 10
executions instead of just 1:


OrderedCollection:  464 464 463 464 460   avg: 463
Array:  496 497 495 496 479    avg: 492.6

difference: (492.6 - 463) / 463.0 * 100  ==>  6.39


Okay, it's less than 10%, but it's there.  And I still don't think it's
a terribly big deal -- there are probably bigger speedups to be had in
the HTML formatter than this, and anyway, it's fast enough for the pages
I actually use with Scamper.

At the time, though, nothing seemed more important than speeding up HTML
formatting!


-Lex

PS -- sure wish I could have just "unloaded" Andreas's patch to do the
comparison, instead of removing it method by method!