As I told you in an earlier mail I'm back cleaning up and reworking guile-log and
refreshing the memory of the inner details of that code base enabled me to rewrite
the spec for redo safe variables considerable. I think that it is much cleaner now and
should be worthy of a good discussion.
-----------------------------------------------------------------------------------------------------------------------
Dear editors of srfi,
SUMMARY:
========
Reinstating continuation in order to capture a state and later restart
from the same state is a concept that at least in proof assistants and
similar software with user interaction is important. But also direct
algorithms in logic programming can take advantage of this concept.
Unfortunately scheme is partly broken when trying to implement these
constructs effectively with little effort and hence it could be an idea
to try improve on the issue. This srfi describes semantics that can help
improve the current state without sacrificing code clarity and the
elegance of scheme. We will refer to this concept as redo semantics and
hence name the new variable types as redo safe variables and a new
parameter type redo safe parameters. Also we should distinguish between
different kinds of undo Capability, full capability or region guarded
capability.
The basic idea to conceptually implement this is to keep a list of
guarded variables and parameters and then use this to guide storage
and retrieval of states. The complexity
comes from managing this list and keep it's size small. The main trick
to accomplish this is to intelligently add and remove guarded variables
depending on where the location of mutation on the stack is. Also one
can target the guarding towards a special undo usage pattern and e.g.
save only on backtracking over specific regions of the stack.
AUTHORS BACKGROUND:
===================
I'm an active developer of scheme centered around the guile scheme
ecosystem and the developer of guile-log, a logic programming system on
top of guile. Guile log implements all of kanren but for efficiency
reasons and also semantically reasons are able to use stack based logic
programming as well as association list or functional tree versions of
assoc for managing variable bindings and implement the proper
backtracking. One concept that this enables is guarded variables.
A SHORT BACKGROUND ON THE ISSUE:
================================
The background has a plurality of origins that converged into one
concept.
1. Current scheme is a beautiful language but has a drawback. If you
set! a variable that variable will be boxed and any uses of redo
semantic becomes broken. This is not acute in scheme because we try
to avoid set! and that is well. But there are cases where you would
like to use set!. One example being utilizing a simple generator e.g.
(next! n) -> (begin (set! n (+ n 1)) n)
This can many times be optimized efficiently and with a good semantic
in a looping macro for example but it will not be scheme. To
efficiently achieve that without using set! seams to be difficult.
Also the resulting loop! macro will not mix well with redo features
on top of the system. What one would like is to have unboxed variables
if the variables are local only. The problem is that the theoretical
motivation behind keeping local variables unboxed is unsound at first
sight because you will have different semantics for local variables
compared to if you put a non local one in a lambda. Now by viewing
these variables as redo safe variables we can treat mutating non local
variables as well as non boxed mutating local ones into one semantic
meaning which means there is a sound theoretical semantic for this
optimized variables e.g. being redo-safe variables.
2. Emacs lisp is coded very much in an imperative style and with guile
we would like to sell guile scheme as a back end for emacs lisp and
emacs as a whole. One of the selling point to do this could be that we
could use continuations to seamlessly add undo redo like semantics to
already written programs in emacs lisp. The problem is twofold here.
i) Typically emacs variable are dynamical variables. Dynamic variables
in scheme is modeled on parameters and they posses the property that
for each dynamic state S and parameter p the evaluation of p under S is
always the last value setted e.g. it is working like a local variable
in the S environment. Because each instance of a continuation
get's it's own dynamic state, the possibility of freezing a state that
can be restarted multiple times seams to be there assuming there is no
other non local exits. The problem is that a dynamic state can be
referenced as a scheme object and used as the current state multiple
times as well for the same continuation and hence we must overwrite
the initial value at a normal return from the dynamic state and hence
loose the ability to store and restore naturally.
ii) Scheme local variables does also show up in the compilation of emacs
lisp to guile. The problem with this is that typically lisp programs
in emacs is littered with set!-ing of local variables. And the semantic
for this in scheme is to box them and again you loose the ability to
redo and undo without a hefty reprogramming or write an advanced
compiler to scheme - which is typically not a beautiful solution. Again
a good concept for enabling seamless and efficient and well designed
code that can be undone and redone multiple times is needed.
3. Guile log is a combination of both traditional stack based logic
programming concept and the kanren concept and together with
a separate continuation implementation targeted for logical variables.
To be able to code all of kanrens concept including a stack as well
for speed and extra semantics like easy tracing or more generally
to have a dynamic-wind like idioms. The ability to redo and undo is
really helpful and enables interesting logic programs to be written. To
accomplish this effectively the stack is needed So here it was found out
that having variables that dynamically can change between behaving like
normal variables or during other dynamical context behave like a state
preserving variable yield very satisfying semantics. So guile log uses a
separate stack(s) that have quite a lot of optimization's and one of
them is efficient redo safe variables. Knowing the usefulness of the
concept for advanced logic programming, it seams interesting to add the
same concept as a srfi to increase scheme's ability to express similar
algorithms.
RATIONAL FOR LIFTING THIS TO A SRFI REQUEST:
============================================
The concept have been proven useful and also implementing this in the
right way do touch on the fundamentals of scheme in such a way that it,
from the perspective of the author, needs help from the greater scheme
community in order to improve upon the idea leading to a good
specification. Also having it as a srfi will mean that redo semantics
will have the potential to be better served in a portable manner. The
main drawback refers to making the ability to reason about code
incorporating different kinds of variables one way to mitigate this
could be to mark variables binding to items with this complexity like
we do with mutation. But a general discussion and idea creation how to
be able to make code transparent would be welcomed.
MAIN CAVEATS:
=============
1. A good implementation of this concepts that are lightweight and will
not disturb the speed of scheme programs much, this means that the
targeted VM instructions and compilers might need to change.
2. The functionality without a good language support risk of messing
up the ability to reason about scheme code. Therefore we want support
in scheme to make it possible to introduce this concept in a sane way.
SUGGESTED SPECIFICATION:
========================
In order to specify the logic we need:
1. dynamic-wind from r5rs. call/cc from r5rs, and parameters from
srfi-39 to mainly to have thread local variables.
2. It is suggested a new variable kind will be called a redo safe
variable and a new parameter object called redo safe parameter and the
concept will be marked with a ~, like with ! a macro or function that
spread this concept should be marked with ~. Else if the semantic follow
what is expected by the standard use the usual symbol nomenclature.
3. We will need a concept of a redo safe wind level. It is a bit more
involved then just a number. Anyhow with this concept goes a
parameter that we denote k for the actual wind level of the guard and
K for the wind-level of the reinstating of the continuation. So when we
guard a variable we will associate a k with that guard and when
reinstating a redo-safe continuation we will need to supply a parameter
K that represent the Wind level of the instance.
4. The logic for restoring a state or not in the wind levels k and K are
in order,
i) if k is a boolean value then that decides the restore or not
ii) if K is #t then we will restore
iii) if K and k is a number then we will restore a state if K < k
iv) if K is a procedure then K should be a predicate and (K k)
decides if we should restore or not.
v) else do not restore.
The rational for i) is that this enables important optimizations that
can be done if the compiler knows that k is #t or #f.
The rational for ii) is that we can force a re-storage of the state
that is convenient in the important case of redo and undo semantics
regarding user interaction logic for e.g. proof assistants.
The rational for iii) is that in e.g. logic program we might want to
keep a variable that behaves just as a variable regarding the logic
of the idiom in question e.g. postpone semantics, and that the body
of the code in this idem with redo safe data be stored and restored
multiple perhaps times. Then an enclosing idem could fire a restore
at a higher level hence the name redo safe wind level and at that time
the variables that previously behaved as ordinary variables would
restore it's state.
The rational for iv) is just to make the notion extensible to
logic not foreseen by the author.
5. The main concept we will use here is a thread local list,
redolist
that represents the current list of guards for different variables.
The list is a logical set not necessary a scheme list.
6. The variable arguments to the idioms below can be either parameters
or scheme variables, we will use the term variables for both these
concepts for now in the spec, the reason is that other overhead
should be more important then the introduction of a check for
parameter or variable in the code, it is indeed more important to be
able to batch them together and hence treat variables and parameters
with the same constructs.
7. The lowest level code will just take a variable and make a guard
object of that variable. In essence we will at this moment fix the
k wind level and produce an identity that we can place in different
guard operations. So we will have a macro,
(make-redo-safe-guard var k)
This creates a guard object identity with the frozen wind level k and
var identity. To Note here is for parameters the guarded variable the
guard should refer to the current active parameter slot e.g. if we
later issue a with-parameters ... idiom we will introduce a new slot
associated with the parameter and hence to redo guard it we must
issue a new redo-safe-guard "dance".
The main operations are,
1 (add-variable-guard f)
Push guard identity f onto the front of the redo list
2 (remove-variable-guard f)
Removes the guard identity f from the redo list preserving the
order of the other guards on the list
8. We will need to be able to store a continuation and later restore
it and by doing that maybe restore some of the guarded variables.
Now we introduce a new call/cc e.g.
(redo-safe-call/cc cc)
As usually cc should be a lambda e.g.
(lambda (cont) ...)
The difference from standard call/cc is that
1. at the call of redo-safe-call/cc, the redo-list is stored and
is traversed for all guarded variables and the by the guarded
variables and parameters referenced value will be stored.
2. cont is now a "lambda" of the form
(lambda (K) ...)
E.g. when we instantiate it, it has the reinstation wind-level
as a parameter.
3. When the cont is instantiated, the standard call/cc reinstation
will take place and after that the saved redo-list will
be traversed and the stored values will be restored into the
different variables.
9. With this we have a minimal working system to implement
redo-safeness. But it has a drawback. The redo-list can become
really huge, and using the idioms correctly and effectively can be
difficult. The opportunity to improve on this is that quite often
one have good control on where on the stack layout we do mutation.
Let Set! represent an location of the stack where we do mutation and
mark with X non mutation regions. Also let Define represent the
definition of the variable and we will assume that the variable is
never referenced "below" the Define, then a common idiom is
a) Define Set! X ...
b) Define Set! X ...' Set! X ... Set! '' X ...
We will concentrate on redo-safeness for state storage after the Define
on the stack. In case the variable get pushed into a lambda and later
used although the control has passed below Define in the stack and
there gets mutated and stored we refer to the lower level framework. And
generally is difficult to automate in a sensible way.
For a) we note that we really just need to guard the variable directly
after the Set! and put the necessary logic in a dynamic wind there
an optimization and hence keep the redo-list small. Not only this, by
using this idiom for case b) we can still support a weaker form of
redo safeness e.g. for cases when we basically always when we store
a state also backtrack back over the guard. With this logic guarded
parameters would mean essentially just the ordinary parameters but
in some implementations e.g. guile there is a small extra overhead to
make the semantic mix with other features e.g. the ability to swap
dynamic state.
For b) We need between ' and '' to have a guard on the redo-list, but
not after '' where we could as in a) in stead again add a guard on the
dynamic stack.
So in this light the following idioms seams natural:
i) guards
(redo-safe-variable-guard guard ((v k) ...)
code ...)
Here v and k will be freezed for guarding and corresponding guard
identities will be made the symbol will be used as
(guard thunk)
In thunk if the guarded variable(s) are not mutated then it is
guaranteed that any stored state inside thunk will be correctly
restored. If there is no mutation between the storage operation and
the control leaving the thunk the semantic would be redo-safe in
the strong sense. Else the redo safeness is not specified and
implementations as the reference implementation below can make
it redo safe, or a more optimized implementation could chose to
have the weak form implemented.
ii) LR guards,
(redo-safe-variable-lr-guard left-guard right-guard ((v k) ...)
code ...)
Here v and k will be freezed for guarding and corresponding guard
identities will be made the two symbols will be used as a guard
and used as
(left-guard thunk)
(right-guard thunk)
The semantics is that when control enters the left-guard thunk the
guard identities will be added to the redo-list and when it leaves,
the guard identity will then be removed from the list.
For the right guard the behavior of the redo-list is reversed with
the addition of the guard behavior in i)
This finishes the specification.
HOW DOES THIS MEET THE EXAMPLES THAT WE STARTED FROM?
=====================================================
a) For the case with code where set! is needed and the data is mainly
local data one would typically use something like
(let ((v val) ...)
(redo-safe-variable-lr-guard lguard rguard ((v #t) ...)
(lguard thunk)))
A well designed compiler would then see that the local variables can
be put onto the call frame unboxed and the code would be very efficient.
If one places the variable in a lambda and it is mutated it will be
boxed and an appropriate guard will be added, one also one can
optionally expose rguard in such a way that one can prevent an
explosion of guards. This concept diverges if we consider state storage
after the control leaves the let and v is put in a live mutating lambda.
b) For integrating undo/redo semantics in emacs with guile-scheme one
could extend elisp with some compiler directives like
#parameter-guard-semantic = #t / #f
#parameter-lguard?
#variable-guard-semantic = #t / #f
#variable-lguard?
With implementing the same guard as a) for k = #t and skipping it
altogether with k = #f. The kind of guard could be controlled by the
lguard? versions which meas use lguard (#t) or guard (#f) as a guard.
The with e.g.
#parameter-guard-semantic #t
#parameter-lguard? #f
#variable-guard-semantic #t
#variable-lguard? #t
in the source file we would have efficient lightweight guarding with
really low overhead or sometimes even less overhead
(local variables oriented code) but still have the weak form of
redo safeness.
c) To implement the same logic as in guile-log but using delimited
continuation and a function stack is with this concept doable (although
you will lose quite a lot of optimizations) Anyhow this imply that scheme
with this addition is more expressible. As an example in a logic program
like in kanren having a idiom that fold each element into a accumulator
is difficult to do in a redo safe way. So say these accumulators are used
in a setup where you take a breath first search together with subset
selection of active continuations which use again variables and then
another postpone like idiom is working on-top of this and another one
etc. getting this complex creature will be a breeze with redo safe
variables and redo safe parameters.
REFERENCE IMPLEMENTATION:
=========================
Assume that we have syntax-rules macrology system together with a
parameters (srfi-39) with the addition of the reference implementation
therein of an idiom:
(get-parameter-cell parameter)
which returns a cons cell with the cdr being the parameters current
value and by setting the cdr we will change the parameter value at that
frozen state.
(define redolist (make-parameter '()))
(define (add-variable-guard f)
(redolist (cons f (redolist))))
(define (remove-variable-guard f)
(redolist
(let loop ((l (redolist)))
(if (pair? l)
(if (eq? (car l) f)
(cdr l)
(cons (car l) (loop (cdr l))))
'()))))
(define (do-restore? k K)
(cond
((eq? k #t)
#t)
((eq? k #f)
#f)
((eq? K #t)
#t)
((number? K)
(and (number? k) (< K k)))
((procedure? K)
(K k))
(else #f)))
(define-syntax make-redo-safe-guard
(syntax-rules ()
((_ v k)
(let ((kk k))
(if (parameter? v)
(let ((cell (get-parameter-cell v)))
(lambda ()
(let ((V (cdr cell)))
(lambda (K)
(if (do-restore? k K)
(set-cdr! cell V))))))
(lambda ()
(let ((V v))
(lambda (K)
(if (do-restore? k K)
(set! v V))))))))))
(define-syntax add-fnames
(syntax-rules ()
((_ fname ((v k) . l) (m ...) (u ...) th . l)
(let ((f (make-redo-safe-guard v k)))
(add-fname l (m ... f)) (f u ...) th . l))
((_ fname () l u th . l)
(fname l u th . l))))
(define-syntax redo-safe-variable-lr-guard
(syntax-rules ()
((_ left right l code ...)
(add-fname redo-safe-variable-lr-guard-0 l () () th
left right code ...))))
;; We make a simplistic implementation and only use a strong
;; guard here for the stack.
(define-syntax redo-safe-variable-lr-guard-0
(syntax-rules ()
((_ (f ...) (f2 ...) thunk lguard rguard code ...)
(let ((lguard
(lambda (thunk)
(dynamic-wind
(lambda ()
(add-variable-guard f2) ...)
thunk
(lambda ()
(remove-variable-guard f) ...))))
(rguard
(lambda (thunk) (thunk))))
code ...))))
(define-syntax redo-safe-variable-guard-0
(syntax-rules ()
((_ (f ...) (f2 ...) thunk guard code ...)
(let ((guard
(lambda (thunk)
(dynamic-wind
(lambda ()
(add-variable-guard f2) ...)
thunk
(lambda ()
(remove-variable-guard f) ...)))))
code ...))))
(define-syntax redo-safe-variable-rguard
(syntax-rules ()
((_ g l code ...)
(add-fname redo-safe-variable-guard-0 l () () g code ...))))
(define (redo-safe-call/cc cc)
(let ((L (map (lambda (f) (f)) (redolist)))
(K #f))
(call/cc
(lambda (cont)
(let ((Cont (lambda (KK)
(set! K KK)
(cont))))
(cc Cont)))
(for-each (lambda (f) (f K)) L))))