Version 1 of stasher

This page is about (the features that should be produced within) an ongoing project of the Google Summer Of Code 2011.

The idea (abstractly): Suppose you've got some value X, and you know you'll need to know f(X) quite often, where f is some function. Then a stasher can let you compute f(X) once and then "stash" the result within the Tcl_Obj that is X, so that when subsequently you need to know f(X), computing it is "free".

The Tcl core uses this idea in a number of cases, usually when X is a piece of code written by a programmer and f is the operation of compiling that code: Tcl scripts can be bytecompiled, expressions can be bytecompiled too, and regexps compile to a finite automaton. But traditionally, you could only do that if you had implemented f in C. A stasher makes it possible to use any Tcl command as an f here, and additionally it lets you cache g(X), h(X), etc. for as many functions g, h, … as you bother to define.

Interface

(The details below are preliminary.)

Small glossary:

stash: Where cached values are stored. Technically this is going to reside within the internal representation of a Tcl_Obj, so that when the original value is forgotten, everything that was cached for it goes away too.
stasher: A command used to access a stash.
stasher: The command used to create stashers.
property: A computed value that can be cached within a stash.

Stashers are created using the stasher command, which has the syntax

: stasher name setup-script

A stasher has a number of subcommands, the most important of which are

: stasher get value property

that returns the property of the give value, e.g.

  strange get {some stupid() #computer<program>using*strange@programming<language>} bytecode

that would return the bytecode for some stupid() #computer<program>using*strange@programming<language> in the strange programming language.

When called, the get subcommand does one of two things:

It passes the value on to the property property subsubcommand, and uses whatever it returns.
It may return some result obtained previously when doing 1.

Doing 1 may be expensive, but doing 2 is essentially free (time-wise).

(To be continued)

Example

First, an example to illustrate the idea of "properties" of a value.

 stasher complexnumber {
    # Lots of code defining properties of complex numbers in various formats
 }
 complexnumber get 3+4i real                    ; # => 3
 complexnumber get 3+4i imag                    ; # => 4
 complexnumber get 3+4i abs                     ; # => 5
 complexnumber get 3+4i arg                     ; # => 0.9272952180016122
 complexnumber get 3+4i parsed                  ; # => rect 3 4
 complexnumber get 5cis0.92 parsed              ; # => polar 5 0.92
 complexnumber get 5cis0.9272952180016122 real  ; # => 3.0000000000000004
 complexnumber get 5cis0.9272952180016122 imag  ; # => 3.9999999999999996
 complexnumber get 3 real                       ; # => 3
 complexnumber get 3 imag                       ; # => 0
 complexnumber get 3 parsed                     ; # => rect 3 0

Next, an example to illustrate how stashing can make a difference for speed

 stasher book {
    # Lots of code defining properties of "book"s and how to compute or look them up
 }
 set pptt isbn:0-13-038560-3
 book get $pptt title

Some delay when book retrieves a library catalog entry for the book with ISBN 0-13-038560-3, and then returns: Practical programming in Tcl/Tk.

 book get $pptt authors

No delay before returning: {Welch, Brent B.} {Hobbs, Jeffrey} {Jones, Ken}, as the full catalog entry is stashed within the $pptt value.

 set pptt

Returns: isbn:0-13-038560-3. We're not automagically mutating values, we're doing things to the internal representation that supplements that value.

 book get isbn:0-13-038560-3 authors

Returns {Welch, Brent B.} {Hobbs, Jeffrey} {Jones, Ken} as above, but after a delay. Though the two strings are equal, they are not stored in the same Tcl_Obj, so when book is handed one it cannot access the data stashed within the other.

Category Internals