FSet v2.3.0: Transients!

FSet v2.3.0 added transients!  These make it faster to populate new collections with data, especially as the collections get large.  I shamelessly stole the idea from Clojure.

They are currently implemented only for the CHAMP types ch-set, ch-map, ch-2-relation, ch-replay-set, and ch-replay-map.

The term "transient" contrasts with "persistent".  I'm using the term "persistent" in its functional-data-structure sense, as Clojure does: a data structure is persistent if multiple states of it can coexist in memory efficiently.  (The probably more familiar use of the term is in the database sense, where it refers to nonvolatile storage of data.)  FSet collections have, up to now, all been persistent in this sense; a point modification to one, such as by with or less, takes only O(log n) space and time to return a new state of the collection, without disturbing the previous state.

A transient encapsulates the internal tree of a collection so as to guarantee that it holds the only pointer to the tree; this allows modifications to tree nodes to be made in-place, so long as the node has sufficient allocated space.  Once the collection is built, the tree is in the same format that existing FSet code expects, and can be accessed and functionally updated as usual.

Some quick micro-benchmarking suggests that speedups, for constructing a set from scratch, range from 1.6x at size 64 to as much as 2.4x at size 4096. 

You don't necessarily even have to use transients explicitly in order to benefit from them.  Some FSet builtins such as filter and image use them now.  The GMap result types ch-set etc. also use them.

For details, see the GitLab MR.

FSet v2.2.0: JSON parsing/printing using Jzon

FSet v2.2.0, which is the version included in the recent Quicklisp release, has a new Quicklisp-loadable system, FSet/Jzon.  It extends the Jzon JSON parser/printer to construct FSet collections when reading, and to be able to print them.

On parsing, JSON arrays produce FSet seqs; JSON objects produce FSet replay maps by default, but the parser can also be configured to produce ordinary maps or FSet tuples.  For printing, any of these can be handled, as well as the standard Jzon types.  The tuple representation provides a way to control the printing of `nil`, depending on the type of the corresponding key.

For details, see the GitLab MR.

NOTE: unfortunately, the v2.1.0 release had some bugs in the new seq code, and I didn't notice them until after v2.2.0 was in Quicklisp.  If you're using seqs, I strongly recommend you pick up v2.2.2 or newer from GitLab or GitHub.