Andrew Cooke | Contents | Latest | RSS | Previous | Next


Welcome to my blog, which was once a mailing list of the same name and is still generated by mail. Please reply via the "comment" links.

Always interested in offers/projects/new ideas. Eclectic experience in fields like: numerical computing; Python web; Java enterprise; functional languages; GPGPU; SQL databases; etc. Based in Santiago, Chile; telecommute worldwide. CV; email.

Personal Projects

Choochoo Training Diary

Last 100 entries

Felicitations - Empowerment Grant; [Bike] Fixing Spyre Brakes (That Need Constant Adjustment); [Computing, Music] Raspberry Pi Media (Audio) Streamer; [Computing] Amazing Hack To Embed DSL In Python; [Bike] Ruta Del Condor (El Alfalfal); [Bike] Estimating Power On Climbs; [Computing] Applying Azure B2C Authentication To Function Apps; [Bike] Gearing On The Back Of An Envelope; [Computing] Okular and Postscript in OpenSuse; There's a fix!; [Computing] Fail2Ban on OpenSuse Leap 15.3 (NFTables); [Cycling, Computing] Power Calculation and Brakes; [Hardware, Computing] Amazing Pockit Computer; Bullying; How I Am - 3 Years Post Accident, 8+ Years With MS; [USA Politics] In America's Uncivil War Republicans Are The Aggressors; [Programming] Selenium and Python; Better Walking Data; [Bike] How Fast Before Walking More Efficient Than Cycling?; [COVID] Coronavirus And Cycling; [Programming] Docker on OpenSuse; Cadence v Speed; [Bike] Gearing For Real Cyclists; [Programming] React plotting - visx; [Programming] React Leaflet; AliExpress Independent Sellers; Applebaum - Twilight of Democracy; [Politics] Back + US Elections; [Programming,Exercise] Simple Timer Script; [News] 2019: The year revolt went global; [Politics] The world's most-surveilled cities; [Bike] Hope Freehub; [Restaurant] Mama Chau's (Chinese, Providencia); [Politics] Brexit Podcast; [Diary] Pneumonia; [Politics] Britain's Reichstag Fire moment; install cairo; [Programming] GCC Sanitizer Flags; [GPU, Programming] Per-Thread Program Counters; My Bike Accident - Looking Back One Year; [Python] Geographic heights are incredibly easy!; [Cooking] Cookie Recipe; Efficient, Simple, Directed Maximisation of Noisy Function; And for argparse; Bash Completion in Python; [Computing] Configuring Github Jekyll Locally; [Maths, Link] The Napkin Project; You can Masquerade in Firewalld; [Bike] Servicing Budget (Spring) Forks; [Crypto] CIA Internet Comms Failure; [Python] Cute Rate Limiting API; [Causality] Judea Pearl Lecture; [Security, Computing] Chinese Hardware Hack Of Supermicro Boards; SQLAlchemy Joined Table Inheritance and Delete Cascade; [Translation] The Club; [Computing] Super Potato Bruh; [Computing] Extending Jupyter; Further HRM Details; [Computing, Bike] Activities in ch2; [Books, Link] Modern Japanese Lit; What ended up there; [Link, Book] Logic Book; Update - Garmin Express / Connect; Garmin Forerunner 35 v 230; [Link, Politics, Internet] Government Trolls; [Link, Politics] Why identity politics benefits the right more than the left; SSH Forwarding; A Specification For Repeating Events; A Fight for the Soul of Science; [Science, Book, Link] Lost In Math; OpenSuse Leap 15 Network Fixes; Update; [Book] Galileo's Middle Finger; [Bike] Chinese Carbon Rims; [Bike] Servicing Shimano XT Front Hub HB-M8010; [Bike] Aliexpress Cycling Tops; [Computing] Change to ssh handling of multiple identities?; [Bike] Endura Hummvee Lite II; [Computing] Marble Based Logic; [Link, Politics] Sanity Check For Nuclear Launch; [Link, Science] Entropy and Life; [Link, Bike] Cheap Cycling Jerseys; [Link, Music] Music To Steal 2017; [Link, Future] Simulated Brain Drives Robot; [Link, Computing] Learned Index Structures; Solo Air Equalization; Update: Higher Pressures; Psychology; [Bike] Exercise And Fuel; Continental Race King 2.2; Removing Lowers; Mnesiacs; [Maths, Link] Dividing By Zero; [Book, Review] Ray Monk - Ludwig Wittgenstein: The Duty Of Genius; [Link, Bike, Computing] Evolving Lacing Patterns; [Jam] Strawberry and Orange Jam; [Chile, Privacy] Biometric Check During Mail Delivery; [Link, Chile, Spanish] Article on the Chilean Drought; [Bike] Extended Gear Ratios, Shimano XT M8000 (24/36 Chainring); [Link, Politics, USA] The Future Of American Democracy; Mass Hysteria

© 2006-2017 Andrew Cooke (site) / post authors (content).

Unifying OO, FP, Asynch Messages

From: "andrew cooke" <andrew@...>

Date: Wed, 30 May 2007 06:42:36 -0400 (CLT)

Some ideas evolving and/or expressed in different ways...

Version 1

Writing Erlang code to receive messages is very like writing functions. 
Why isn't it identical?

First, there are the largely syntactical issues of function names and
currying.  These might be addressed by using "bundles" of anonymous
functions, or associating the tag in a tagged tuple with a function name.

But secondly there is also the problem of thread-local state.  When a
thread is waiting on a message it alread has some state.  This seems to be
similar to partial evaluation with explicit state.

And there is also the idea of synchronisation - messages are only
"received" when the thread is ready.

At the same time, there's the idea that one sends messages to objects. 
And objects with explicit state are similar to the partialy evaluated
functions described above.

Binding a partially evaluated function to a public name would "expose" it
for evaluation.  This could be within an existing thread, in which case it
becomes normal evaluation, or it could be associated with a new thread, in
which case it becomes message passing.

The new ideas are:
- Publishing functions with names
- Evaluating functions synchonously or asynchronously
- Bundling a group of functions into a single function
- Partially evaluating functions

An object is a partially evaluated bundle - the partial evaluation is the
state and the bundle defines the methods.

Version 2

Ignore the partial evaluation/currying and relating that to objects -
that's just the usual objects as closures thing.

Instead, consider how to unify message passing and function calling.  What
does that require?  If functions receive messages then the mailbox must be
a separate entity.  But then you are sending messages to mailboxes which
isn't the same as sending them to functions.

Stuck here.

Version 3

Classes are functions that return a pair - the "next" function and a
result.  Functions are anonymous.  Here is a class:

> MkFoo = [A inc B -> Foo A, B + A
>         |A inc B -> Foo A, B - A]
> Foo = MkFoo 2
> Foo inc 3
> Foo dec 2

Here is another way to define that:

> MkFoo = [A ->
>            Foo A, [inc B -> B + A
>                   |dec B -> B - A]]

Here's another way:

> Inc = [A B -> B + A]
> Dec = [A B -> B - A]
> MkFoo = [A ->
>            Foo A, [inc -> Inc A
>                   |dec -> Dec B]]

What does "," mean?  Seems to be list construction.

"[...]" is not list construction but syntax for function definition.  Or
maybe they are the same?    But then how do we distinguish between plain
lists and parameters without looking ahead to "->"?  And does that mean
"->" is first class?  No.  This is not Lisp-like.

A different example to show persistent state

> MkBar = [A B -> MkBar B, A + B]
> Bar = MkBar 2
> Bar 3
> Bar 1
> Bar 0

Version 4

Bad thing about Version 3 is that there is no clear distinction between
variables bound to pure functions and variables bound to mutable objects. 
Maybe the syntax should be such that they can't be bound to objects (think
list comprehensions, for example).

Version 5

Objects cannot be references from more than one thread or we get nasty
things happening.  Easiest way to prevent this is to restrict initial code
to export only functions and constants (not variables9 (as Erlang now).

Version 6

These objects are a kind of fixed point (or iterated point?).  Maybe a
syntax something like:

> {State, Value} <-> function(State, Arg)

where "State" is intentionally repeated and indicates the iterative
nature, Value is returned and Arg is the parameter.  Hmmmm.  Not perfect

Or maybe

> myFunc State Arg -> ....
>   return State2, Result
> myObj Arg -> {State, Bar} = myFunc State Arg -> Bar

But what is the initial value of State?

> myFunc State Arg -> ....
>   return State2, Result
> myObj Arg -> {Start = State, Bar} = myFunc State Arg -> Bar

Or, returning to something like list comprehensions

> myObj Arg = {Bar | (State, Bar) <- myFunc State Arg, State = Start}

Ooooh!  That works!  Means we need list comprehensions, though :o(
Something like a generator, which is close to a loop...

Version 7

C-- is sleeping.  LLVM is incomprehensible.  GCC is scary.  Parrot seems

More on OO/FP/Asynch

From: "andrew cooke" <andrew@...>

Date: Wed, 30 May 2007 07:50:36 -0400 (CLT)

The final list comprehension syntax was wrong because we need to return
not the state but the partially evaluated function (otherwise, when used
to receive messages in a state machine, you are stuck in the same state).


> myFun State Arg ->
>   ....
>   (myFun State, Result)
> myObj Arg = {Result | (Fun, Result) <- Fun Arg, Fun = myFun Start}

Result is not in a tuple - that must be a list (variable number of return
values).  In an untyped languge do we need tuples?  Not really - lists are

Also, to get something like inheritance, you need function bundles

> myFun = {myFun1, myFun2 SomeState, myFun3 _ OtherState}

Using some kind of partial evaluation syntax.

What does foldl look like in this syntax?  The above gives a kind of
generator.  It's not a function - can't be partially evaluated.  Bad
question.  What about parser/combinator?  That doesn't seem right either. 
Have I missed what I was aiming at?


More Jabberings on Syntax

From: "andrew cooke" <andrew@...>

Date: Sat, 2 Jun 2007 11:33:02 -0400 (CLT)

I'm sure this makes sense to no-one, and only partially to me but it is
coming together,I hope.

In parallel with this I am exploring Parrot - see wiki at


assuming we have an eval(function, args) for automated code, I think it's
a net win to have a special syntax for 0-arity functions.  This seems to
be necessary in an eager language if you want Haskell's currying and
"open" syntax.

Also, for dispatch on atoms to simulate message passing to objects, we
want to be able to define anon functions easily.

So the following seems to work:

somefun = {arg1 arg2 arg3 -> some definition
          |other args     -> other definition
          |yet more args  -> another}

For 0-arity functions you simply have {....}.  Evaluation is with "!".

> {print "hello world}
> {print "hello world}!
hello world

Note that parsing requires special lookahead, but that's just a technical

This gives a very nice way of introducing laziness, which gives you
control functions for free:

if = {true  block1 block2 -> block1!
     |false block1 block2 -> block2!
     |true  block -> block!
     |false block -> {}}

(Is it possible to have variable arity functions?)

> if true {print yes} {print no}
> if false {print yes} {print no}

Which is probably similar to code blocks in smalltalk or whatever?

One thing you seem to lose is the ability to do

somefun arg1 arg2 arg3 -> some definition
somefun other args     -> other definition
somefun yet more args  -> another

Maybe that can be added as sugar?  A lot seems to depend on how you use
indentation to delimit expressions.  Need to read up on Haskell's offside


One More Step

From: "andrew cooke" <andrew@...>

Date: Sun, 10 Jun 2007 09:05:45 -0400 (CLT)

I've been trying to write this up in a paper and was fed up by the end of
yesterday, because I couldn't pull it together into a coherent whole. 
However, in bed last night I think I found the key.

One of the common threads is the iteration of a function over a mutable
kernel.  This is what happens in state machines (processes running against
mailboxes), allows "objects" to be emulated, and accumulates data in list

Last night I realised the obvious - that it is also present in the "right
hand side" of list comprehensions, which motivates a syntax:

 > X = [1, 2, 3];
 > Y <- X;
 > X;
 [2, 3]
 > Y;
 > mkCounter Count inc -> [Count, mkCounter $ Count + 1];
 > C = mkCounter 0;
 > A <- C inc;
 > B <- C inc;
 > A;
 > B;

One remaining problem is what to do if you *only* want the side effect.
 > <- C inc;
seems a bit silly.



From: "andrew cooke" <andrew@...>

Date: Thu, 14 Jun 2007 08:04:57 -0400 (CLT)

An update on this, because last night I felt I reached what could be
described as the end of an iteration.  That's the positive spin - the
negative version is that it didn't work.  But continuing with the
positive, here are lessons learned so far:

Good ideas related to function evaluation:
- combining functions in series and parallel
- functions with clauses with different numbers of arguments
- ordering of clauses
- partial evaluation of functions
- ability to "freeze" evaluation

Good ideas related to syntax:
- syntax without parentheses (a la Haskell)
  (allows eager/lazy mix described above)
- atoms and variables semantically distinct
  (allows pattern matching and binding)
- {...} for function clauses
- [...] for lists (with | for append)

Bad ideas that led to dead-end:
- forcing list comprehension syntax to mean something
- chaining state with <- operator ([A|B] = B ...)
  (in the end, just ugly and uncomfortable)

Emerging good ideas for next iteration:
- use mailbox to hold state
- partial evaluation with mailbox gives "objects"
- need for "nil" as false?!


Iteration 2

From: "andrew cooke" <andrew@...>

Date: Sun, 17 Jun 2007 18:30:57 -0400 (CLT)

I discarded list comprehension and focussed on mailboxes.

The most interesting new idea comes from a deeper understanding of how
eager evaluation, partial evaluation, variable arity, and "case"
statements in function definitions interact.  While the syntax may look
like currying, functions need to be treated "whole" if the correct case is
to be selected.

For example in:
  myfun .atom1 .atom2 42 -> ...;
  myfun .atom1 -> {.atom2 "foo" -> ...};
  myfun .atom1 var -> ...;
all 3 cases will match myfun .atom1 .atom2, but what happens when a third
argument is applied?  In the first case it must be integer 42.  If it's
not, we expect a different case to be chose - the second case (truly
curried) would be selected but would then fail if the thirs argument was
not the string "foo".

For this iteration the second case would be selected, and then (with a
third argument of .atom3, for example) the program would fail.  To
improves this I seem to need to do at least one of:
- change the syntax to delimit function arguments
- remove variable arity
- disallow partial evaluation
- add backtracking

Obviously the last of these is much the most interesting.  So I need to
read up on Snobol...  From wikipedia:  "providing operators for pattern
concatenation and alternation" (which is exactly what is happening here
with function arguments).


Lessons from Icon

From: "andrew cooke" <andrew@...>

Date: Mon, 18 Jun 2007 06:33:42 -0400 (CLT)

It was Icon, more than Snobol, that I needed to investigate (Icon being a
more modern development and a more general language - Snobol is perhaps
more like Awk, with string matching and a reliance on "goto").

At first I couldn't see how Icon's treatment of failure, looping, and
coroutines were related, but I think the following implementation
mechanism clarifies things (no idea if it is correct, but it was useful).

Consider a language with support for both exceptions and continuations. 
My impression is that Icon-like functionality can be implemented in such a
language by doing the following:

- Adding an implicit parameter to function calls, which is a stack of

- On failure, throwing an exception which contains the continuation stack.

- On requiring a repeat, pushing the current continuation to the stack,
and throwing an exception containing the stack.

- Some operations (fork points), on entry, push continuations for
alternatives onto the stack.

- Various operations (fork points and, fir example, explicit "every")
catch the exceptions described above, pop the topmost continuation from
the stack, and call it (passing the rest of the stack as normal).

- In addition to this "automatic" handling of continuations, also allow
explicit creation and calls to coroutines.

The above, taken together, provides for both backtracking and
coroutine-like looping.

For SFL things can be simplified considerably, since only backtracking is
needed (instead of coroutines, threads and mailboxes are used).  Also, to
reduce memory usage we need either (or both):

- An explicit "cut" that discards the coroutine stack

- Automatic detection and elimination of continuations within repeated
(tail) calls to the same routine (not so sure about this - seems to make
sense until you consider very abstract, high level, general routines like

So, to be more concrete:

- Joining a pair of functions into a bundle creates a fork point as
described above.

- Failure to pattern match throws an exception with continuation stack.

- Explicit notation for failure and cut also needed.

Ideas that might be worth exploring:

- Labelling fork points and allowing a cut to that label (or restarting
from that label).

- Supporting coroutines anyway.



From: "andrew cooke" <andrew@...>

Date: Tue, 19 Jun 2007 09:22:54 -0400 (CLT)

Backtracking isn't trivial to implement :o)

A bundle of single parameter functions forms a choice point.  The
functions are ordered (from whether they appear to the left or right of |
and the first function is evaluated.  If that evaluation fails (typically
because of pattern matching failure between parameter and argument) then
the next function is evaluated instead.  If evaluation succeeds then
evaluation of the bundle as a whole is a success, a value is returned, and
the choice point is discarded.

Failure may occur ``deeper'' within the evaluation of a function ---
through the failure of nested and composed functions, or variable binding
with patterns, or even via explicit failure (the <<< operator).  In all
these cases control returns to the choice point and the next function in
the bundle is evaluated.

The same process extends directly to multiple parameter functions (through
currying) and to the SFL program as a whole.

The backtracking implementation should impose minimal overhead.  In
particular, care needs to be taken with (not) breaking tail call 

There are two issues here.  First, how to avoid consuming memory on tight
tail call loops.  Second, how to clear choice points that are no longer
relevant (because the program has ``returned through them'').  The second
is only an issue if the solution to the first moves choice point handling
from the stack.

Possible (partial) solutions:

- Separate stack of continuations.
- Specify a maximum depth.  Ugly and broken by tight inner loops.
- Name stack entries (via user annotation) and delete any stack between
repeated names.  Works well for high level loops, but not for utility
- Allow forcing --- none, or perhaps one level only of backtracking. Might
be useful for inner loops.  A use case with higher level functions would
be instructive; want to be as flexible as possible.
- Do use the stack, but ameliorate with some of the above.
- Allow backtracking to be disabled in certain scopes.


- One use case for backtracking is accumulate data on success and return
it on failure.  In other words, as loop control.  Here it is the driver
function (the fold or whatever) that needs to be efficient --- the inner,
evaluated function, may use backtracking, but is called separately on each
- Main loop in a server, handling incoming messages.  Needs backtracking
but also needs to tail call for next message.


Comment on this post