I liked Travis's style. He began the talk by almost opening the conference, introducing himself, thanking people, taking a survey of Smalltalk's in use, industries, etc. I thought it was very slick. He has a calm presence in front of the audience.

I got a lot out of Travis's presentation. There are some new and interesting things I've learnt about Smalltalk and about optimising Smalltalk. I'd say I got more out of his talk than I usually get out of talks (and I usually get a fair bit).

One big thing I realised from his presentation was that the assumptions behind your optimisation may change. So, your optimisation may work really well, but then the OS changes, the hardware changes, a new version of VW comes out, etc.. and these changes can invalidate your optimisations assumptions.

To combat this, what we need to do is write unit tests that -test- our expectations of a new algorithm or technique. That way, when the assumptions change, our test will fail and we'll know we should do two things: (a) remove the optimisation from use (not necessarily delete it) and (b) invert our test so that if the new assumption changes again in the future, we'll know about it.

So thank you Travis, what a wonderful way to start the conference.