immediate Characters
Last updated at 3:28 pm UTC on 29 February 2020
The new Spur VM memory model supports immediate Characters so that Characters with Unicode code points greater than 255 can be tested via #==, and wide string access is much faster.
This was introduced with Squeak 5.0.
For the binary representation see Spur image format.
A practical example of the effect of immediacy is demonstrated by the deprecation of 'Character characterTable at: 64' in favor of 'Character value: 63'.
Character allInstances size=0. Why?
If you are accustomed to Character as represented in earlier Squeak (or other Smalltalk) versions, then this result will be surprising.
The reason is that the internal representation of Character has changed in recent versions of Squeak. We now use an object memory design (called "Spur") developed by Eliot Miranda that enables a number of improvements and optimizations.
One important optimization is that instances of Character can now be represented as "immediate" objects, in which the data for an instance of a class is actually hidden directly in the "object pointer" that refers to the instance. That instance is in every way a real object but is optimized in such a way that the information (in this case the character value) is encoded directly ("immediately") inside the object pointer that specifies the instance.
This optimization allows certain simple objects to be encoded very efficiently. And it also has the somewhat surprising side effect of making it appear that there are no "real" instances of the class when you use #allInstances to scan the object memory looking for instances of the class.
In earlier versions of Squeak, only class SmallInteger was able to benefit from this optimization. So in older versions, you will see this for the optimized SmallInteger:
SmallInteger allInstances size ==> 0
But for Character, which did not benefit from an internal representation as an "immediate" class, you might see something like this:
Character allInstances size ==> 257
Since that time, Character has now been updated to be an immediate class, and it now behaves much like SmallInteger. Thus on modern Squeak you will now have:
Character allInstances size ==> 0
Spur enabled another similar optimization for floating-point objects. On older Squeak versions, you will see that all instances of Float are ordinary non-immediate objects. So you might see something like this in an older image:
Float allInstances size ==> 8362
But on newer Squeak images with the Spur object format, most floating-point objects can now be optimized with an immediate representation. It depends on the actual numeric values involved, but most Float objects can be represented by class SmallFloat64, which is an optimized "immediate" representation that (like SmallInteger) appears to have no instances when you scan the objects memory:
SmallFloat64 allInstances size ==> 0
However, some floating point values cannot be packed inside an object pointer, and these need to be represented by "normal" objects. This is done with class BoxedFloat64, so your image might show a number of instances of these non-immediate objects:
BoxedFloat64 allInstances size ==> 34
Modern Squeak images on Spur still have a class called Float. But all of the actual instances of Float are represented by two concrete subclasses. Floating point values that can be packed into an object pointer are now instances of SmallFloat64, and values that cannot fit into that optimized format are represented by the "normal" class BoxedFloat64.
And of course, the original class Float is now abstract, so it has no instances at all:
Float allInstances size ==> 0
(David T. Lewis, http://forum.world.st/Character-allInstances-size-0-for-5-3-latest-tp5112604p5112616.html)
