Squeak 2.3 news?

Dan Ingalls DanI at wdi.disney.com
Sun Jan 3 22:03:13 UTC 1999


Andrew -

>I am always curious where a 30% 
>improvement was found, by the way -- where'd you find it this time?

There were two major improvements:

1.  By declaring the stack pointer to be the REAL end of a context, we saved the time to nil out all those fields above the stack pointer on every send.  There were some secondary saving here also because without this burden one can use a single size of contexts, and strip out all the logic for different frame sizes, plus get better sharing of recylced contexts.

2.  By keeping a tiny cache of indexable objects, at: and at:put: are able to skip a lot of the logic for bounds checking and element type.  In the process, by testing a bit more on the first access, it became possible to do fast access on more than just a couple of system classes.

We also picked up a bit by reordering the instruction fetch sequence and persuading C to keep a few things in registers.

There are also numerous improvements such as folding in various contributions to support 32-bit image sizes, and strings and files with positions and sizes that are > 30 bits.  Plus a couple of others to make things work better with Jitter.  I'm assuming Ian will write the full story of Jitter II as soon the dust settles -- it's really a jewel.

>This is not threat -- it is the point of OSS -- uncontrolled spinoffs are a 
>virtue, not a vice.  If the present custodian isn't doing something you 
>want or need -- go ahead and do it yourself.  If the present software 
>doesn't suit your interest and you are not inclined to make the change or 
>wait for it -- use something else that does.  If it doesn't yet exist, make 
>it (and please consider making it OSS).
>
>Changes can be added and incorporated in the "central" release, or not, as 
>the "market" for information demands.  If it is a minority interest, the 
>spinoff will eventually wither and "die" (only in the sense of falling into 
>obscurity), or become a feature for a small sub-community.

One thing we can do to assist in this coevolution is to communicate a bit more about our directions.  It's obviously time for another installment of "Where Squeak is Headed" from the point of view of Squeak Central.  We are right in the midst of some important presentations, and the release of 2.3.  Then during the week of the 20th we will take some time out to catch our breath and do a bit of strategic planning for '99.  I will make it a point to summarize the relevant points of that meeting for everyone on this list.  That way, at least you will have some sense of what to expect, and what not to expect (from us, at least), in the year ahead.

Thanks for your support

	- Dan





More information about the Squeak-dev mailing list