This section contains many ideas to insert in
the text as it is rewritten.
The ideas are in no particular order
(not even order of chronological appearance),
having been put at random places in the file
as they came, or were moved from the written text,
since late january 1995 when redacting this article began.
A.1 About the whole article
Some of this draft should definitely be moved
to other Tunes documentation files,
or expanded into independent articles.
A.2 Part I
Part I would:
- Show that OS utility lies in its influence on dynamic CS behavior
- The OS is not as much the software as the protocols
- Show that this influence is in the way the common background
allows to increase signal/noise ratio,
that is to give meaning to observable data,
to provide expressive languages using the obsersable world as
- The role of the "kernel" is to provide some central authority
as a ultimate resource to arbitrate conflicts
and guarantee consistency.
- Constraints of an Operating System:
it may contains only a tiny fraction of the total information
in the CS, as its information is bounded by what one computer
can know, whereas the system is bounded by what N computers can know.
- evolves slowly, in a conservative way
so that dataflow can rely on it.
- (old stuff)
computing is a recent art
whose evolution is well-known
- multimedia is the latest OS slogan;
when we see through this veil of illusion,
the trend is toward adding new functions
to the OS.
and the trend in which they evolve
show that they fail
[u vs p]
see what is their approach,
- see why it fails
The essence of an OS is no more
in a kernel that would supervise
all forms of communication between objects,
than the essence of civilization
lies in a central administration
that would supervise
all forms of communication between humans.
The essence of an OS
is in the abstract property of
allowing objects to communicate,
through any possible decentralized means;
it is in its utility as a general context for communication,
much as civilization is an intangible
set of said or unsaid traditions and rules,
that allow humans to rely on each other.
focus on what they should do,
not on what they do
(what defines a place setting
is not its having the shape of a fork or that of a spoon,
but its ability to ease lunch activity,
that is, its function, not its implementation).
(xref to PartII: centralize)
utility -- correlation to static ou dynamic features
informational basis that gives meaning
to the flux of raw information;
- (I.11 ?)
- (I.12 ?) The ultimate source of meta(n)-information: Man
- Security is being able to devise arbitrary contracts,
and have the guarantee that if agreed upon,
the contract will be fulfilled.
Systems that don't allow you to express the contract you want
are stupid unsecure systems.
Systems that do allow you to express the contract you want,
but have no way to enforce it (e.g. literate programming)
are ineffective unsecure systems.
Systems that enforce contracts that you don't want
are fascist unfree systems.
and only such information can eventually and enrich the whole system.
basis of any reliable information upon which new information can be built
that will enrich the whole system;
when this information eventually settles, it enriches in turn the OS,
and can serve as a universal basis for even further enhancements.
That is the utility of Operating Systems.
That's why the power and long-term utility of an OS
mustn't be measured according to what the OS does currently allow to do,
but according to how easily it can be extended
so that more and more people share more and more complex software.
That is, the power of an OS
is not expressed in terms of services it statically provide,
but in terms of services it can dynamically manage;
intelligence is expressed not in terms of knowledge,
but in terms of evolutivity toward more knowledge.
A culture with a deep knowledge
but that would prevent or considerably slowdown further innovations,
like the ancient chinese civilization, would indeed be quite harmful.
An OS providing lots of services,
but not allowing its user to evolve
would likewise be harmful.
Utility lies in new, original information;
a large body of acquired information is a sign of past utility,
but quite independent from current utility.
Again, we find the obvious analogy with human culture
for which the same stands;
the analogy is not fallacious at all,
as the primary goal of an operating system
is allowing humans
to communicate with computers more easily to achieve better software.
So an operating system is a part of human culture,
though a part that involves computers.
Multiplying the actual services provided by an operating system
may be an expedient way to solve computer problems,
in the same way that multiplying welfare institutions
may be an expedient way to solve the everyday problems of a human system;
the progress of the system ultimately means that those services
will actually be multiplied in the long run.
However, from the point of view of utility,
what counts is not any the objective state of the system at any given moment,
and its ephemeral advantages,
but the dynamic project of the system across time,
and its smaller, but growing, long-standing advantages.
the information in an OS is virtually (not forcibly physically)
duplicated at each node.
Hence growing the OS for ever more feature is harmful,
as it would involve an ever increased waste of resources
duplicated at each node, instead of letting each node develop
original information in a way adapted to its immediate environment.
A.3 Users are Programmers
The only source of information in the UCS that we can directly act upon,
hence what counts with respect to utility, is the Humans.
Therefore, Operating Systems should structure the Computing System
so that the fullest possible human creativity is promoted.
The deepest flaw in computer design
is this idea that there is a fundamental difference
between system programming and usual programming,
between usual programming and "mere" using.
The previous point shows how false is this conception.
The truth is any computer user, whether a programming guru or a novice
user, is somehow trying to communicate with the machine. The easier
the communication, the quicker better larger the work is getting done.
Of course, there are different kinds of use;
actually, there are infinitely many.
You can often say that such kind of computer use is
much more advanced and technical than such other;
but you can never find a clear limit,
and that's the important point
(in mathematics, we'd say the space of kinds of computing is connected).
Of course also,
any given computer object has been created by some user(s),
who programmed it above a given system,
and is being used by other (or the same) user(s),
who program using it, above the thus enriched system.
That is, there are computer object providers and consumers.
But anyone can provide some objects and consume other objects;
providing objects without using some is unimaginable,
while using objects without providing any is pure useless waste.
The global opposition between users and programmers that roots
the computer industry is thus inadequate;
instead, there is a local complementarity between providers and consumers
of every kind of objects.
Some say that common users are too stupid to program;
that's only despising them;
most of them don't have time and mind
to learn all the subtleties of advanced programming;
Most of the time, such subtleties shouldn't be really needed,
and learning them is thus a waste of time
but they often do manually emulate macros,
and if shown once how to do it,
are very eager to use or even write their own macros or aliases.
Others fear that encouraging people to use a powerful programming
language is the door open to piracy and system crash,
and argue that programming languages are too complicated anyway.
Well, if the language library has such security holes and cryptic syntax,
then it is clearly misdesigned;
and if the language doesn't allow the design of a secure, understandable
library, then the language itself is misdesigned (e.g. "C").
Whatever was misdesigned, it should be redesigned, amended or replaced
(as should be "C").
If you don't want people to cross an invisible line, just do not draw roads
that cross the line, write understandable warning signs, then hire an army of
guards to shoot at people trying to trespass or walk out of the road.
If you're really paranoid, then just don't let people near the line:
don't have them use your computer. But if they have to use your computer,
then make the line appear, and abandon these ill-traced roads and fascist
So as for those who despise higher-order and user-customizability,
I shall repeat that there is NO frontier
between using and programming.
Programming is using the computer
while using a computer is programming it.
Which does not mean there is no difference between various users-programmers;
but creating an arbitrary division in software
between "languages" for "programmers" and "interfaces" for mere "users"
is asking reality to comply to one's sentences
instead of having one's sentences reflect reality:
one ends with plenty of unadapted, inefficient, unpowerful tools,
stupefies all computer users
with a lot of unuseful ill-conceived, similar but different languages,
and wastes a considerable lot of human and computer resources,
writing the same elementary software again and again.
A.4 Operating System Kernel
In traditional OS design,
the kernel is some central piece of software through which any
communication between first-class system objects is done...
But this accounts only for centralized design;
it appears that what system acknowledge as first-class objects
are actually very coarse-grained information concepts,
and that a meaningful study of information flow should take into
account much finer-grained information,
that such system just do no consider at all,
hence being unadapted to the actual use that is done of them.
How does this design generalize to arbitrary OSes?
What do OS kernels provide that is essential to all OSes,
and what do they do that is costly noise?
To answer such questions, we must depart from the traditional
OS point of view that we know is flawed,
and see how are OSes doing, that we recognized as such,
that traditional design refuses to consider this way,
and what the analogy to human systems lead to.
Thus, we see that of course, centralization of the information flow
through the kernel is not needed:
hence, information most often is much more efficiently passed directly
from object to object without any intermediate.
To conclude, we'll say that the kernel is the central authority
used to coordinate software components,
and solve conflicts, in a computer system.
A.5 Current state of System software
It is remarkable that
while since their origins,
computer hardware have grown in power and speed
at a constant exponential rate,
system software only slowly evolved in comparison.
It does not offer any new tools
to master the increasing power of hardware,
but only enhancements of obsolete tools,
and new "device drivers" to access new kinds of hardware as they appear.
System software becomes fatware (a.k.a. hugeware),
as it tries to cope differently
with all the different users' different but similar problems.
It is also remarkable that
while new standard libraries arise,
they do not lead to reduced code size
for programs of same functionality,
but to enhanced code size for them,
so that they take into account all the newly added capabilities.
As a blatant example
of the lack of evolution of system software quality
is the fact that
the most popular system software in the world (MS-DOS)
is a fifteen-year old thing that does not allow the user
to do either simple tasks, or complicated ones,
thus being a no-operating system,
and forces programmers to rewrite low-level tasks
everytime they develop any non-trivial program,
while not even providing trivial programs.
This industry-standard has always been designed
as a least sub-system possible for the Unix system,
which itself is a least subsystem of Multics
made of features assembled in undue ways
on top of only two basic abstractions,
the raw sequence of bytes ("files"),
and the ASCII character string.
As these abstractions proved not enough to express adequately
the semantics of new hardware and software that appeared,
Unix has had a huge number of ad-hoc "system calls" added,
to extend the operating system in special ways.
Hence, what was an OS meant to fit the tiny memory of
then available computers,
has grown into a tentaculous monster with ever growing pseudopods,
that wastes without counting the resources of the most powerful workstations.
And this, renamed as POSIX,
is the new industry standard OS to come,
whose promoters crown as the traditional, if not natural, way
to organize computations.
Following the same tendency, widespread OSes are
found upon a large number of human interface services,
video and sound.
This is known as the "multi-media" revolution,
which basically just means that your computer produces
high-quality graphics and sound.
All that is fine:
it means that your system software
grants you access to your actual hardware,
which is the least it can do!
But software design, a.k.a. programming,
is not made simpler for that;
it is even made quite harder:
while a lot of new primitives are made available,
no new combinatorials are provided
that could ease their manipulation;
worse, even the old reliable software is made obsolete
by the new interface conventions.
Thus you have computers with beautiful interfaces
that waste lots of resources,
but that cannot do anything new;
to actually do interesting things,
you must constantly rewrite everything from almost scratch,
which leads to very expensive low-quality slowly-evolving software.
A.6 An Ancien Régime
[= most of the energy is wasted in a fight for supremacy
The current computing world is anything but a failure.
So many things are now done by computers that relieve people
from stupid repetitive work, and so many things are done that
just could not be done without computers,
that nobody can deny the utility of today's computers
relatively to the implicit reference being the absence of computers.
But somehow, programming techniques are finding their limits
as programs reach the size beyond which
no human can fully understand the whole of one.
And the current OS trend, by generating code bloat,
makes those limits reached much faster than they should,
while wasting lots of human resources.
It is thus necessary to see
why current programming techniques lead to code bloat,
and how this trend can be slowed down, set back, or reversed.
Of course, we easily can diagnose about the "multimedia revolution"
that it stems from the cult of external look, of the container,
to the detriment of the internal being, the contents;
such cult is inevitable whenever non-technical people have
to choose without any objective guide among technical products,
so that the most seductive wins.
So this general phenomenon,
which goes beyond the scope of this paper,
though it does harm to the computing world,
and must be fought there as well as elsewhere,
is a sign that computing spreads and benefits to a large public;
by its very nature, it may waste a lot of resources,
but it won't compromise the general utility of operating systems.
Hence, if there is some flaw to find in current OS design,
it must be looked for deeper.
Computing is a recent art, and somehow,
it left its Antiquity for its Ancien Régime.
Its world is dominated by a few powerful companies,
that wage a perpetual war to each other,
At the same time, there are heavens where computists
can grow in art while freely benefitting
isn't the deeply rooted
the informational status of the computer world
is quite remindful of the political status of
A.8 Contents of an Operating System
What are the characteristic components of an operating system ?
Well, firstly, we may like to find some underlying structure of mind
in terms of which everything else would be expressed,
and that we would call "kernel".
Most existing OSes, at least, all those software that claim to be an OS,
are conceived this way.
Then, over this "kernel" that statically provides most basic services,
"standard libraries" and "standard programs" are provided
that should be able to do all that is needed in the system,
that would contain all the system knowledge,
while standard "device drivers" would provide
complementary access to the external world.
We already see why such a conception may fail:
it could perhaps be perfect for a finite unextensible static system,
but we feel it may not be able to express a dynamically evolving system.
However, a solid argument
why it shouldn't be able to do so is not so obvious at first sight.
The key is that like any complex enough systems,
like human beings, computer have some self-knowledge.
The fact becomes obvious when you see a computer being used
as a development system for programs that will run on the same computer.
And indeed the exceptions to that "kernel" concept are
those kind of dynamic languages and systems
that we call "reflective", that is,
that allow dynamical manipulation of the language constructs themselves:
FORTH and LISP (or Scheme) development systems,
which can be at the same time editors, interpreters, debuggers and compilers,
even if those functionalities are available separately,
are such reflective systems.
so there is no "kernel" design,
but rather an integrated OS.
And then, we see that if the system is powerful enough
(that is, reflective),
any knowledge in the system can be applied to the system itself;
any knowledge is also self-knowledge; so it can express system structure.
As you discover more knowledge,
you also discover more system structure,
perhaps better structure than before,
and certainly structure that is more efficiently represented directly
than through stubborn translation to those static kernel constructs.
So you can never statically settle once and for all the structure
of the system without ampering the system's ability to evolve toward a better
any structure that cannot adapt, even those you trust the most,
may eventually (though slowly) become a burden as new meta-knowledge
is available. Even if it actually won't, you can never be sure of it,
and can expect only refutation, never confirmation of any such assumption.
The conclusion to this is that you cannot truly separate a "kernel"
from a "standard library" or from "device drivers";
in a system that works properly, all have to be integrated
into the single concept, the system itself as a whole.
Any clear cut distinction inside the system is purely arbitrary,
and harmful if not done due to strong reasons of necessity.
A.9 Toward a Unified System
From what was previously said, what can we deduce about how
an OS should be behaved for real utility ?
Well, we have seen that an OS' utility is not defined in terms
of static behavior, or standard library functionality; that it
should be optimally designed for dynamic extensibility, that it
shall provide a unified
interface to all users, without enforcing arbitrary layers
(or anything arbitrary at all). That is, an OS should be primarily
open and rational.
But then, what kind of characteristics are these ? They are features
of a computing language. We defined an OS by its observational semantics,
and thus logically ended into a good OS being defined by a good way to
communicate with it and have it react.
People often boast about their OS being "language independent", but what
does it actually mean ?
Any powerful-enough (mathematicians say universal/Turing-equivalent)
computing system is able to emulate any language, so this is no valid argument.
Most of the time, this brag only means that they followed no structured plan
as for their OS semantics, which will lead to some horrible inconsistent
interface, or voluntarily limited their software to interface with the
least powerful language.
So before we can say how an OS should be, we must study computer languages,
what they are meant to, how to compare them, how they should be or not.Comparing computers and cars:
people say that computers, like cars,
should have everything done by the machine,
with the user never having to modify anything.
- but cars are rarely creative objects
Most people use cars to move from some place to another,
which they don't consider as a piece of art,
as a work they produce.
They rather feel it's some inevitable noise,
that should be reduced as much as possible.
- cars are merely tools to relieve people from the burden of displacement,
and even then, we don't forbid people from repairing their car themselves,
or adding something to it, or making it.
Of course, there are laws about how cars should or should not be done,
that these persons should follow like all manufacturers,
for security reasons.
- Thus, in so far as computers are tools that people are not developing,
everything should be made to relieve people from the hassle of using
the computer, to hide all the nasty details,
to provide everything possible to make their daily computer usage
easy and secure, fool-proof, etc,
to the detriment of raw performance,
and even of some "liberties" that bring only chaos
(like the liberty to drive on either side of the road would be).
- This is a sign that Computing as a project evolves,
and the obtained computerware are objects that this project leaves
behind it; the more advanced the project, the more elaborate these
- Now, information technology, under its particular form of computers
as well as all of its forms, is precisely not a complete project,
but a project in continuous development.
- Surely, people should not have to worry about completed parts,
(though they should not be prevented to worry about them either).
- but more importantly, they should be able to freely contribute to
- The problem is that the society as it is does not regard
meta-information as more powerful than terminal information;
it tends to judge things according to their cost instead
of judging them according to their value.
as if only a class of people should learn to read,
and do all the work of reading in place of other
butis precisely an art that is ever-developing.
It could be said that what the
are not in development anymore,
that become more and more objects,
- Defining an OS as a set of low-level abstractions:
If freight technology had been left to similar companies and academies,
they would have put the horse as the only basic abstraction on which to
They could have made a new standard when it would have been obvious
that steam engines should have been adopted twenty years ago,
and similarly for all newer technologies...
In any case, it doesn't directly tackle the real problem,
which is reliable transport of goods;
it just forces to people to use standard technology,
however it be obsolete,
and prevents them from developping structures
that would survive this technology.
Also see the disaster of State managing services.
- Under a free programming system,
independent software modules are made independently for independent purposes.
Under current bound programming systems,
software are not modular, not independent,
and if you can't convince an established company
that your purpose does match that
of enough ready-to-pay people (money-wise),
you just can't write it at all.
Hence Free software means that more software will be written,
that it will have more feedback, hennce will be better in turn, etc.
- See the failure of the french industry,
because of its illusory policy of developing their "own standard"
(i.e. not a standard, as they're not strong enough to impose it by force)
in a way bot internally centralized and externally isolated (!!!);
such an industry can survive only by constantly stealing the taxpayers,
and that's exactly what it has been doing for decades.
- Let's use the limited metaphor of the computing system as a human society:
- at its basis is a Constitution,
which has the double role of
acknowledging the few informal rules
that are found as universal requirements for a just society,
and of settling arbitrary general settings as an agreed-upon frame
in which those requirements can be provided.
- then are laws, bills, and conventions,
that are arbitrary binding but renegociable contracts,
made whenever a common solution is needed to some shared problem.
- then are the executives,
who must do the minimal work
of verifying that the legal constraints are always respected,
that information does flow freely,
and that noise and disinformation are discouraged.
the operating system "kernel" is like State
-- it regulates interaction between objects;
- the very basics of the system are like the constitution
-- a set of informal rules that explains the
general principles of the system and establish a common arbitration.
- Standard protocols are like Laws and Rules
-- they provide common features at the expense of common constraints,
that defines the way object interact.
- Laws et al should include arbitrary decisions
insofar as and as long as the fact that an arbitrary decision is made
itself is not arbitrary:
see the classical problem of everyone driving on the same side of the roads;
whether everyone should drive on the right side or on the left side
of the road is essentially arbitrary;
that everyone should either drive on the right side or on the left side
of the road is not.
- The most common error about State and Operating systems
is to believe that either should actually MANAGE
the system and ultimately do or have do everything about it.
That's completely WRONG.
They should regulate things
- States are the skeleton of Societies.
if the skeleton was all that counted in a man,
men would be ... skeletal!
Surely the skeleton is important,
but it is not it that will make the man move;
it will only serve as a background that supports the move.
- As an example, the Académie Française,
meant to represent France's most proeminent litterary authors,
is NOT meant to write all possible french litterature,
or to sum it up, or to establish what litterature should be or not.
Instead, it will be an authority as to what are the rules of the french
language and litterature, and keep a standard of it,
not inventing it but rather making a synthesis of what exists,
so that people speaking french do have a common reference.
- Similarly, the OS authorities should not provide the ONE TRUE PROGRAMS
that will perform each single task in the system,
but instead will maintain a public, open reference
of how people are meant to communicate,
and should be required to communicate whenever
a disagreement appears that cannot be otherwise fairly settled.
- There need not be a single administration that would
manage all laws and regulations.
Instead, it is much better that various specialized administrations
made of proficient people each manage the fields where their members
- Choosing people in each of these specialized administration is not
harder than choosing people in a one centralized administration:
specialization restricts the choice of potential nominees,
and should also modify the weight of voters according to
their expected knowledge about how to judge proficiency on the
- Of course, there need be a constitutional means
to settle disagreements,
and this means eventually there is a ultimate authority
(because all ultrafilters on finite sets are principal);
but central arbitration doesn't mean central management at all.
A central arbitration would not take any initiative by itself,
and would not rule anything,
only judge between alternative when asked to.
Refering to it should be an exceptional event;
when a submitted case clearly matches a field
for which an established authority already exists,
the central authority would always follow
the opinion of the competent authority,
so that people wouldn't argue over and over when
the competent authority decided something.
- If privacy was one of the constitutional principles,
then laws can't uselessly constraining the private behavior of objects.
- the protection-handling or proof-checking "microkernel"
is like the executive
-- it enforces the respect of the rules of the system.
Seeing how existing human States and computer kernels fail
to do their job is left as an exercise to the reader.
The point is that all this infrastructure is meant to help
objects (people) communicate with each other in fair terms,
so that the global communication is faster, safer, and more accurate,
with less noise, while consuming less resources.
It should make the objects nearer to each other.
The role of State id to allow people to communicate.
To stay politically as neutral as possible
(after all, this is a technical paperr),
the paper should try to not explicitly use a reference to State, if possible.
Instead, it would conclude with a note according to which
the very same argument would hold when applied to human societies
as similar dynamical systems.
- Contrarily to the socialists,
who say that a state-ruled society is the End of History,
the Authentic Liberals do not say that a free, fair, market
is the end of history;
on the contrary, they say that a free, fair, market,
is the beginning of history;
it is a prerequisite for information to pass well,
for behaviors to adapt,
for changes to operate,
for history to exist.
The freer, fairer the market, the more history.
- It may be said that computing has been doing quantitative leaps,
but has not done any comparable qualitative leap;
computing grows in extension, but does not evolve toward intelligence;
it sometimes rather becomes more largely stupid.
This is the problem of operating systems not having
a good conceptual kernel:
however large and complete their standard library,
their utility will be essentially restricted to the direct use of the library.
A.10 Newest Operating Systems: the so-called "Multimedia revolution"
This phenomenon can also be explained by the fact that programmers,
long used to software habits from the heroic times
when computer memories were too tight
to contain more than just the specific software you needed
(when they even could),
do not seem to know how to fill today computers' memory,
but with pictures of gorgeous women and digitized music
(which is the so-called multimedia revolution).
Computer hardware capabilities evolved much quicker
than human software capabilities;
thus humans find it simpler to fill computers
with raw data (or almost raw data) than with intelligence.
Those habits, it must be said,
were especially encouraged by the way information
could not spread and augment the common public background,
since because of lack of theory and practice of what a freely
communicating world could or should be,
only big companies enforcing "proprietary" label could
up to now broadcast their software;
people who would develop original software
thus had (and sadly still have) to rewrite everything from almost scratch,
unless they could afford a very high price for every piece of software
they may want to build upon,
without having much control on the contents of such software.
- The role of the OS infrastructure in the computer world
is much like that of the State in human societies:
it should provide justice by guaranteeing,
by force if needs be, that contracts will be fulfilled, and nothing more.
In the case of computer software,
this means that it will guarantee that contracts passed between
objects will be fulfilled,
that objects should fulfill each other's requirements
before they can connect.
When there is no Justice, there is no society/OS, but only chaos.
- Because it ain't in the Kernel doesn't mean it ain't done.
[Because the government doesn't do it doesn't mean nobody does it].
- The Kernel is there as a warrant that voluntarily agreed
contracts between objects be respected:
if function F is ready to trade a golden coin from some quantity of
gold powder, the kernel will see that people trading with F
will actually provide the right amount of gold powder,
whereas F will actually return a gold coin.
- Part II would discuss programming language utility,
stating the key concepts about it.
Any reuse includes some rewrite, which is to minimize.
Similarly, when we "rewrite",
we often reuse a lot of the formal and informal ideas from
and even when we reinvent,
we reuse the inspiration,
or sometimes feedback from people already inspired.
- Notably after discussing how to be able to construct as
many new concepts as possible,
it should explain that the key to concept expressivity
(that reflectivity cannot indefinitely postpone)
is their separation power,
and thus the capability to affirm one of multiple alternatives,
to express different things, to negate and deny things.
- Literate Programming, and D.E. Knuth's attempts with WEB and C/WEB
(see this interview of D.E. Knuth
are actually ways to pass more information about programs.
To pass information that programming languages themselves
through well-organized human-readable documentation.
This is A GOOD THING,
because there will ALWAYS be things that
humans can (already) express that machines cannot express (yet).
But this is NOT THE PANACEA,
because there ARE things that the machines
ACTUALLY COULD express with high-level languages,
that pure literate programming over low-level languages
require the human to not only to write, but to check,
when a computer is much better suited to check them.
Knuth completely ignores the meta- capabilities of computers.
- Each independent part is subject to the limit of a one-man's understanding
so it be reliable.
Existing systems are coarse-grained,
which means that independent parts or large portions of programs,
so that a complete program is made of few independent parts,
and that total program complexity is limited to the sum of a few
direct human understandings.
- Tunes will be fine-grained and reflective,
so that a complete program is made of arbitrarily many limited parts,
and can arbitrarily grow in complexity.
- Paradoxically, while their user-interface abstractions
are coarse-grained and "high-level" (in a complex),
current OSes only provide a very low-level set of programming abstractions
to combine at a fine-grained level for any reliability/efficiency.
There is double-speach here,
and both users and programmers are hindered.
- If safety criteria are not expressible by the computer system,
then to be safe, programs must be understandable by men.
And because it is not essentially harder to express the
criteria to the machine than to another man,
this most likely means that a one man.
Because that man won't ever be there to maintain code,
because armies of "maintainers" won't replace him,
because there is no tool to safely adapt old programs to new points of views,
then every so often, code must go to the dust bin.
No wonder why software evolves so slowly:
only some small human experience remains,
and even then,
because there is no way to express what that experience is,
it cannot spread in technology fast ways,
but only man to man.
- Having generic programs instead of just specific ones is
exactly the main point that we saw about having a good grammar to introduce
new generic objects, instead of just an increasing number of terminal,
first order objects, that actually do specific things (i.e. extending the
- What is really useful is a higher-order grammar, that allows to manipulate
any kind of abstraction that does any kind of things at any level. We call
level 0 the lowest kind of computer abstraction (e.g. bits, bytes, system
words, or to idealize, natural integers). Level one is abstractions of these
objects (i.e. functions manipulating them). More generally, level n+1 is made
of abstractions of level n objects. We see that every level is a useful
abstraction as it allows to manipulate objects that would not be possible
to manipulate otherwise.
But why stop there ? Everytime we have a set of level, we can define a new
level by having objects that arbitrarily manipulate any lower object (that's
ordinals); so we have objects that manipulate arbitrary objects of finite
level, etc. There is an unbounded infinity of abstraction levels. To have the
full power of abstraction, we must allow the use of any such level; but why
not allow manipulating such full-powered systems ? Any logical limit you put
on the system may be reached one day, and this day, the system would become
that's why any system to last must potentially contain
(not in a subsystem) any single feature that may be needed one day.
The solution is not to offer any bounded level of abstraction, but
unlimited abstracting mechanisms; instead of offering only terminal operators
(BASIC), or first level operators (C), or even finite-order
offer combinators of arbitrary order.
offer a grammar with an embedding of itself as an object. Of course, a simple
logical theorem says that there is no consistent internal way of saying
that the manipulated object is indeed the system itself, and the system state
will always be much more complicated than it allows the system to understand
about itself; but the system implementation may be such that the manipulated
object indeed is the system.
This is having a deep model of the system inside itself; and this is quite
useful and powerful. This is what I call a higher-order grammar -- a grammar
defining a language able to talk about something it believes be itself.
And this way only can full genericity be achieved: allowing absolutely
anything that can be done about the system, from inside, or from outside
(after abstracting the system itself).
First, we see that the same algorithm can apply to arbitrarily complex data
structures; but a piece of code can only handle a finitely complex data
structure; thus to write code with full genericity, we need use code as
parameters, that is, second order. In a low-level language (like "C"),
this is done using function pointers.
We soon see problems that arise from this method, and solutions for them.
The first one is that whenever we use some structure, we have to explicitly
give functions together with it to explain the various generic algorithm
how to handle it. Worse even, a function that doesn't need some access method
about an the structure may be asked to call other algorithms which will
turn to need know this access method; and which exact method it needs may not
be known in advance (because what algorithm will eventually be called is not
known, for instance, in an interactive program). That's why explicitly passing
the methods as parameters is slow, ugly, inefficient; moreover, that's code
propagation (you propagate the list of methods associated to the structure --
if the list changes, all the using code changes). Thus, you mustn't pass
explicitly those methods as parameters. You must pass them implicitly;
when using a structure, the actual data and the methods to use it are embedded
together. Such a structure including the data and methods to use it is
commonly called an object; the constant data part and the methods,
constitute the prototype of the object; objects are commonly grouped into
classes made of objects with common prototype and sharing common data.
This is the fundamental technique of Object-Oriented
some call it that Abstract Data Types (ADTs) and say it's only part of the
"OO" paradigm, while others don't see anything more in "OO". But that's only
a question of dictionary convention. In this paper, I'll call it only ADT,
while "OO" will also include more things. But know that words are not settled
and that other authors may give the same names to different ideas and vice
BTW, the same code-propagation argument explains why side-effects are an
especially useful thing as opposed to strictly functional programs (see pure
ML :); of course side effects complicate very much the semantics of
programming, to a point that ill use of side-effects can make a program
impossible to understand or debug -- that's what not to do, and such
possibility is the price to pay to prevent code propagation. Sharing
data (data subject to side effects) between different embeddings (different
users) for instance is something whose semantics still have to be
clearly settled (see below about object sharing).
The second problem with second order is that if we are to provide functions
other functions as parameter, we should have tools to produce such functions.
Methods can be created dynamically as well as "mere" data, which is all the
more frequent as a program needs user interaction. Thus, we need a way to
have functions not only as parameters, but also as result of other functions.
This is Higher order, and a language which can achieve this has a
reflective semantics. Lisp and ML are such languages; FORTH also, whereas
standard FORTH memory management isn't conceived for a largely dynamic use of
such feature in a persistent environment. From "C" and such low-level
languages that don't allow a direct portable implementation of the
higher-order paradygm through the common function pointers (because low-level
code generation is not available as in FORTH), the only way to achieve
higher-order is to build an interpreter of a higher-order language such as
LISP or ML (usually much more restricted languages are actually interpreted,
because programmers don't have time to elaborate their own user customization
language, whereas users don't want to learn a new complicated language for
each different application and there is currently no standard user-friendly
small-scale higher-order language that everyone can adopt -- there are just
plenty of them, either very imperfect or too heavy to include in every
With respect to typing, Higher-Order means the target universe of the
language is reflective -- it can talk about itself.
With respect to Objective terminology, Higher-Order consists in having
classes as objects, in turn being groupable in meta-classes.
And we then see that it does prevent code duplication,
even in cases where the code concerns
just one user as the user may want to consider concurrently two -- or more --
different instanciations of a same class (i.e. two sub-users may need toe
have distinct but mostly similar object classes). Higher-Order is somehow
allowing to be more than one computing environment: each function has its own
independant environment, which can in turn contain functions.
To end with genericity, here is some material to feed your thoughts about
the need of system-builtin genericity: let's consider multiplexing.
For instance, Unix (or worse, DOS) User/shell-level programs are ADTs,
but with only one exported operation, the "C" main() function per executable
file. As such "OS" are huge-grained, with ultra-heavy inter-executable-file
(even inter-same-executable-file-processes) communication semantics no one can
afford one executable per actual operation exported. Thus you'll group
operations into single executables whose main() function will multiplex those
Also, communication channels are heavy to open, use, and maintain, so you
must explicitly pass all kind of different data & code into single
manually multiplexing them (the same for having heavy multiple files or a
manually multiplexed huge file).
But the system cannot provide builtin multiplexing code for each single
program that will need it. It does provide code for multiplexing the hardware,
memory, disks, serial, parallel and network lines, screen, sound. POSIX
requirements grow with things a compliant system oughta multiplex; new
multiplexing programs ever appear. So the system grows, while it will never
be enough for user demands as long as all possible multiplexing won't have
been programmed, and meanwhile applications will spend most of their time
manually multiplexing and demultiplexing objects not yet supported by the
Thus, any software development on common OSes is hugeware. Huge in hardware
resource needed (=memory - RAM or HD, CPU power, time, etc), huge in resource
spent, and what is the most important, huge in programming time.
The problem is current OSes provide no genericity of services. Thus they can
never do the job for you. That why we really NEED generic system
multiplexing, and more generally genericity as part of the system. If one
generic multiplexer object was built, with two generic specializations
for serial channels or flat arrays and some options for real-time behaviour
and recovery strategy on failure, that would be enough for all the current
multiplexing work done everywhere.
So this is for Full Genericity: Abstract Data Types and Higher Order.
Now, if this allows code reuse without code replication -- what we wanted --
it also raises new communication problems: if you reuse objects especially
objects designed far away in space or time (i.e. designed by other
people or an other, former, self), you must ensure that the reuse is
consistent, that an object can rely upon a used object's behaviour. This is
most dramatic if the used object (e.g. part of a library) comes to change
and a bug (that you could have been aware of -- a quirk -- and already have
modified your program accordingly) is removed or added. How to ensure object
combinations' consistency ?
Current common "OO" languages are not doing much consistency checks. At
most, they include some more or less powerful kind of type checking (the most
powerful ones being those of well-typed functional languages like CAML or
SML), but you should know that even powerful, such type checking is not
yet secure. For example you may well expect a more precise behavior from
a comparison function on an ordered class 'a than just being
i.e. telling that when you compare two elements the result can be
"lesser than", "equal", or "greater than": you may want the comparison
function to be compatible with the fact of the class to be actually ordered,
that is x<y & y<z => x<z and such. Of course, a typechecking
which is more than useful in any case, is a deterministic decision system,
and as such cannot completely check arbitrary logical properties as expressed
above (see your nearest lectures in Logic or Computation Theory). That's why
to add such enhanced security, you must add non-deterministic behaviour to your
consistency checker or ask for human help. That's the price for 100%
secure object combining (but not 100% secure programming, as human error is
still possible in misexpressing the requirements for using an object, and
the non-deterministic behovior can require human-forced admission of unproved
consistency checks by the computer).
This kind of consistency security by logical formal property of code is
called a formal specification method. The future of secure programming lies
in there (try enquire in the industry about the cost of testing or
debugging software that can endanger the company or even human lives if ill
written, and insurance funds spent to cover eventual failures - you'll
understand). Life concerned industries already use such modular formal
In any cases, we see that even when such methods are not used automatically
by the computer system, the programmer has to use them manually, by including
the specification in comments or understanding the code, so he does
Now that you've settled the skeleton of your language's requirements, you
can think about peripheral deduced problems.
- When the best fit technique is known,
only this technique, and none else, should be used.
any other use may be expedient, but not quite useful.
Moreover, it is
very hard to anticipate one's future needs; whatever you do, there will
always be new cases you won't
lastly, it doesn't replace combinators
as of the combinatorials allowed
allowing local server objects to be saved by the client is hard to
implement eficiently without the server becoming useless, or creating a
At best, your centralized code will provide not only the primitives
you need, but also the combinators necessary; but then, your centralized code
is a computing environment by itself, so why need the original computing
environment ? there is obviously a problem somewhere; if one of the two
computing environment was good, the other wouldn't be needed !!!;
All these are problems with servers as much as with libraries.
- With a long training, people can avoid most bugs
that typing would have detected;
but this long training has a human cost.
And even then, all bugs are not guaranteed to be avoided,
so insurance is still needed against huge occasional catastrophes,
which also involves a high, non-linear cost.
Actually, the same holds for any kind of static information
that might have been gathered about programs:
you can live without the computer checking it,
by checking it yourself.
But then you must do computer work,
are not guaranteed to do it properly,
and cannot offer the guarantee to your customers,
as youuur proof is all inside your mind
and not repeatable!!!
- Paul R Wilson said:
BTW, this whole wrangle is exactly why I recommend avoiding the term
"weakly typed." It means at least three different things to different
people, and various combinations to other people:
- implicit conversions, and
- unchecked types
implicit vs explicit is what differentiates a HLL from a LLL.
A LLL will require the
- not building an artificial border between programmers and users
=> not only the system programming language must be OO,
but the whole system.
- easy user extensibility -> language-level reflection.
- sharing mutable data: how ? -> specifications & explicitly
(or more or less mutation-prone ?) & time & locking -- transactions.
- objects that must be shared: all the hardware resources
-- disks & al.
- sharing accross time -> persistence
- reaching precision/mem/speed/resource limit:
what to do ? -> exceptions
- recovering from exceptional situations:
how ? -> continuations (easy if higher-order on)
- tools to search into a library ->
must understand all kind of morphism in
a logically specified structure.
- sharing accross network -> distribution
- almost the same: tools for merging code ->
that's tricky. Very important
for networks or even data distributed on removable memory (aka floppies) --
each object should have its own merging/recovery method.
- more generally tools for having side effects on the code.
we consider Logical Structures: each structure contains some types, and
symbols for typed constants, relations, and functions between those types.
Then we know some algebraic properties verified by those objects,
i.e. a structure of typed objects, with a set of constants & functions
& relations symbols, et al.
A structure A is interpreted in another structure B if you can map the
symbols of A with combinations of symbols of B (with all the properties
conserved). The simplest way to be interpreted is to be included.
A structure A is a specialization of a structure B if it has the same
symbols, but you know more properties about the represented objects.
A.13 Mutable objects
We consider the structure of all the possible states for the object. The
actual state is a specialization of the structure. The changing states
accross time constitute a stream of states.
A.14 Sharing Data
The problem is: what to do if someone modifies an object that others see ?
Well, it depends on the object. An object to be shared must have been
programmed with special care.
The simplest case is when the object is atomic, and can be read or modified
atomically. At one time, the state is well defined, and what this state is
what other sharers see.
When the object is a rigid structure of atomic objects, well, we assume that
you can lock parts of the object that must be changed together -- in the
meantime, the object is unaccessible or only readable -- and when the
modification is done, everyone can access the object as before. That's
Now, what to do when the object is a very long file (say text), that each
user sees a small part of it (say a full screen of text), and that someone
somewhere adds or deletes some records (say a sentence) ? Will each user's
screen scroll according to the number of records deleted ? Or will they
stay at the same spot ? The later behaviour seem more natural. Thus, a file
has this behaviour that whenever a modification is done, all pointers to the
file must change. But consider a file shared by all the users across a
network. Now, a little modification by someone somewhere will affect
everyone ! That's why both the semantics and implementation of shared objects
should be thought about longly before they are settled.
A.15 Problem: recovery
What to do when assumptions are broken by higher priority objects ?
e.g. when the user interrupts a real-time process, when he forces a
modification in an otherwise locked file, when the process is out of memory,
Imagine that a real-time process is interrupted for imperative reasons (e.g.
a cable was unplugged; a higher-priority process took over the cpu, etc):
will it continue where it stopped ?
or will it skip what was done during the interruption ?
Imagine the system runs out of memory ? Whose memory are you to reclaim back ?
To the biggest process ?
The smallest ?
The oldest ?
The lowest real-time priority ?
The first to ask for more ?
Or will you "panic" like most existing OSes ?
If objects spawn, thus filling memory (or CPU), how to detect "the one"
responsible and destroy it ?
If an object locks a common resource, and then is itself blocked by a failure
or other unwilling latency, should this transaction be cancelled, so others can
access the resource, or should all the system wait for that single transaction
to end ?
As for implementation methods, you should always be aware that
defining those abstraction as the abstractions they are,
rather than hand-coded emulation for these,
allows better optimizations by the compiler,
quicker write phase for the programmer,
neater semantics for the reader/reuser,
no implementation code propagation for the reimplementer,
Partial evaluation should also allow specialization of code that don't use
all the language's powerful semantics, so that standalone code be produced
without including the full range of heavy reflective tools.
- all the requirements to be used as for Tunes,
or design a new one if none is found.
- That is, without ADTs, and combinating ADTs, you spend most of your time
manually multiplexing. Without semantic reflection (higher order), you spend
most of your time manually interpreting runtime generated code or manually
compiling higher order code. Without logical specification, you spend most of
your time manually verifying. Without language reflection, you spend most of
your time building user interfaces. Without small grain, you spend most of
your time manually inlining simple objects into complex ones, or worse,
simulating them with complex ones. Without persistence,
you spend most of your time writing disk I/O (or worse, net I/O) routines.
Without transactions, you spend most of your time locking files. Without
code generation from constraints, you spend most of your time writing
redundant functions that could have been deduced from the constraints.
To conclude, there are essentially two things we fight: lack of feature
and power from software, and artificial barriers that misdesign of former
software build between computer objects and others, computer objects and
human beings, and human beings and other human beings.
A.16 Centralized code
There's been a craze lately about "client/server" architecture for computer
hardware and software. What is "client/server" architecture that many
corporations boast about providing ?
conceptually, a server is a centralized implementation for a library;
centralized => coarse-grained;
now, coarse grained => evil;
hence centralized => evil.
we also have centralized => network bandwidth waste.
only "advantage": the concept is simple to implement even by the dumbest
Do corporations boast about their programmers being dumb ?
A very common way to share code is to write a code "server" that will
include tests for all the different cases you may need in the future and
branch to the right one.
Actually, this is only some particular kind of library making,
but much more clumsy,
as a single entry point will comprise all different behaviours needed.
This method proves hard to design well,
as you have to take into account all possible cases to arise,
with predecided encoding,
whereas a good encoding would have to take into account actual use
(and thus be decided after run-time measurements).
The obtained code is slow as it must test many uncommon cases;
it is huge, as it must take into account many cases,
most of them seldom or never actually used;
it is also uneasy to use, as you must encode and decode the arguments
to fit its one entry point's calling convention.
It is very difficult to modify,
but by adding new entries and replacing obsolete subfunctions by stubs,
because it would else break existing code;
it is very clumsy to grant partial access to the subfunctions,
as you must filter all the calls;
security semantics become very hard to define.
Centralized code is also called "client-server architecture";
the central code is called the server,
while those who use it are called clients.
And we saw that a function server is definitely something that no sensible man
would use directly;
human users tend to write a library that will encapsulate calls to the server.
But it's how most operating systems and net-aware programs are implemented,
as it's the simplest implementation way.
Many companies boast about providing client-server based programs,
but we see there's nothing to boast about it;
client-server architecture is the simplest and dumbest mechanism
even a newbie is able to do that easy.
What they could boast about would be not using client-server
but truely distributed yet dependable software.
A server is nothing more than a bogus implementation for a library, and
shares all the disadvantages and limits of a library, with enhanced
extensibility problem, and additional overhead. It's only advantage is to have
a uniform calling convention, which can be useful in a system with centralized
security, or to pass the stream of arguments through a network to allow
distant client and servers to communicate. This last use is particularly
important, as it's the simplest trick ever found for accessing an object's
multiple services through a single communication line. Translating software
interface from library to server is called multiplexing the stream of
library/server access, while the reverse translation is called
Then what are "intelligent" ways to produce reusable, easy to modify code?
Such a method should allow reusing code without duplicating it, and without
growing it in a both unefficient and uncomplete way: an algorithm should be
written once and for once for all the possible applications it may have,
not for a specific one. We have just found the answer to this problem:
the opposite of specificity, genericity.
So we see that system designers are ill-advised when they provide such
specific multiplexing, that may or may not be useful, whereas other kind
of multiplexing is always needed (a proof of which being people always
boasting about writing -- with real pain -- "client/server" "applications").
What they really should provide is generic ways to automatically
multiplex lines, whenever such thing is needed.
More generally a useful operating system should provide a generic way to
share resources; for that's what an operating system is all about: sharing
disks, screens, keyboards, and various devices between multiple users and
programs that may want to use those accross time. But genericity is not
only for operating systems/sharing. Genericity is useful in any domain; for
genericity is instant reuse: your code is generic -- works in all
cases -- so you can use it in any circumstances where it may be needed,
whereas specific code must be rewritten or readapted each new time it must
be used. Specificity may be expedient; but only genericity is useful on the
Let us recall that genericity is the property of writing things in their
most generic forms, and having the system specialize them when needed,
instead of hard-coding specific values (which is some kind of manual
Now, How can genericity be achieved ?
- Machines can already communicate; but with existing "operating systems"
the only working method they know is "client/server architecture", that is,
everybody communicating his job to a one von Neuman machine to do all the
computations, which is limited by the same technological barrier as before.
The problem is current programming technology is based on coarse-grained
"processes" that are much too heavy to communicate; thus each job must be
done on a one computer.
- There is lots of laughable hype about network computers (NCs).
NCs are the hardware embodiment of the client/server architecture:
you plug NCs on the net, and the only configuration needed,
that can be done automatically, is assigning them a network name/address.
All the data is on the server side.
This just has all the advantages and disadvantages of the client/server:
surely this ensures consistency of data,
but efficiency is the worst possible among systems that ensure it,
because of centralization.
An efficient system would achieve consistency of installed software
without sacrificing performance, and without requiring a modification
of current hardware,
by doing all the consistency enforcement in software.
Memory that is local to network hosts is then used
for data cacheing and replication;
CPU resources can be used to do distributed computations, etc.
A software solution could do things great.
A hardware solution like NCs is just waste of resources.
All the more, NCs do not even have compatibility,
and do not allow any particular leverage of existing software,
so that they are really a big gratuitous waste of resources.
For a fraction of the price of all the wasted hardware resources,
a proven distributed OS could be developed and marketed!
reflective language frame,
code & data persistence,
explicit batch processing,
sessions & files,
- because many semantical changes are to be manually propagated
accross the whole program.
- "Compilers can not guarantee a program won't crash."
have I been told.
Surely, given an expressive enough language,
there is no compiler that can tell for an arbitrary program
whether it will crash or not.
Happily, computers are not used to run random arbitrary programs!!!
Well, there's genetic programming and corewar, fair enough.
When you program,
Well, instead of all these knowledge and proofs to stay forever untold
and unchecked, it is possible to use a language that can express them all!
A computer language could very well express all the requirements
for the program, and logically check a proof that they are fulfilled.
you know what you want, so you know what a correct program is,
and more easily even what a non-crashing program is.
- you write your program specifically so you can prove to yourself
that the program is correct.
- if programming under contract, you are expected to have done the former,
even though the customer has no way to check, but perhaps having you
fill red tape.
A.18 Part III
- Part III would apply the concepts to existing technology.
It would have to discuss what is tradition,
what is its role, how it should or not be considered,
what it currently does wrong,
how the Tunes approach inserts in it.
- It would debunk myths
- Efficiency,Security,Small-grain: take two, you have the third.
That is, when you need two of them, you also need the third,
but when you have two of them, you automatically have the third.
- with OO, people discovered that implicit binding is needed.
Unhappily, most "OO" only know as-late-as-possible binding
and no such thing as reflectivity (=implicitness control)
or migration (=modification of implicitness control).
"Tradition and Revolution" ?
- "Hierarchy vs Liberty" ?
- "Myths and reality" ?
- "The burden of the past" ?
- "No computer is an island, entire in itself" ?
This part would explain how we apply the principles
from part I and II to actual computing
- It would recall what tradition is,
what the two meanings for revolution are,
and why a one applies and not the other.
- It would try debunk some myths:
- Tapes vs Files vs Persistency
- Linear ASCII Text vs hypertext vs Meta-text,
- Single-Computer OS vs Networked OS vs Distributed OS
- Single-User vs Multi-user vs dynamic user
- console vs GUI vs decoupling of programming and IO
- They are all instances of the
"Flat Resource vs Hierarchical Layering vs Higher-order modularity"
- Text as source files: derives from
the silly notion that because people type programs
by hitting a sequence of keys, each being marked with a symbol,
the program should be stored and manipulated as
the sequence of symbol typed.
That because books are read as such a sequence,
because time is linear and thus any exploration of the text,
then the text should be linear, too.
This also derives from the fact that early computers where so
slow and primitive, with such tight memory,
that programmers had to care a lot about the representation of computer
data, with the sequential nature of their being fed to the computer
through punched paper or magnetic tape.
Derives from belief that the object is its representation,
and that the only valid representation is the "usual" one.
- The Web allows casual users to publish new information
from where they are.
This is quite a progress.
But only passive documents can be published;
any inter-site reference is unreliable.
The most advanced programming techniques (cgi-bin)
only allow unsafe localized low-level computations.
- A common myth about programming is that low-level programming
allows more efficiency than high-level programming.
This is completely untrue, while the opposite is quite true.
Actually, people spend several million dollars at developping
optimizing C and FORTRAN compilers, but a much cheaper Common LISP
compiler (CMU Common LISP, developped by a few students and teachers),
achieve similar performance,
while allowing the whole expressivity of a real high-level language.
Also, people may see that a large part of modern optimizers consist
in making the whole code higher-level, so it can be better
understood and optimized by the compiler.
Any amount of time spent at manually optimizing some routine,
could be equally spent at developping some specialized optimizing
heuristics of same effect on the particular low-level routine,
but that could generalize to further modified versions of the routine,
or of similar routines,
thus improving reliability and maintainability as well as performance,
and saving a lot of time.
Of course, this means that compiler technology with the ability to accept
user-defined optimizing heuristics be widely available.
But this is just possible and will be case.
Instead of losing ever more time at low-level coding,
most low-level people should consider making such a compiler appear sooner.
Actually, a trivial theoretical argument could have told us that already:
high-level programs contain more information and less noise than
low-level programs, hence, can be manipulated and compiled more efficiently,
with proper tools; and anything that can be done in low-level can be
done at least as well, and surely more cleanly and genericly, in high-level.
"No man should do what the computer can do quicker for him (including
time spent to have the computer understand what to do)" -- that's why we need
to be able to give order to the computer, i.e. to program.
- "Do not redo what others already did when you've got more important work"
-- that's why we need code reuse.
- "no uncontrolled code propagation" -- that's why we need genericity.
- "security is a must when large systems are being designed"
-- that's why we need strong typechecking and more.
- "no artificial border between programming and using"
-- that's why the entire system should be OO with a unified language system,
not just a hidden system layer.
- "no computer user is an island, entire by itself"
-- you'll always have to
connect (through cables, floppies or CD-ROMs or whatever) to external
networks, so the system must be open to external modifications, updates and
- Current computers are all based on the von Neumann model in which
a centralized unit executes step by step a large program composed of
While this model is simple and led to the wonderful computer technology
we have, laws of physics limit in power future computer technology
to no more than a grand maximum factor 10000 of what is possible today
This may seem a lot, and it is, which leaves room for many improvement
in computer technology;
however, the problems computer are confronted to are not limited anyway
by the laws of physics.
To break this barrier, we must use another computer model,
we must have many different machines that cooperate,
like cells in a body, ants in a colony,
neurones in a brain, people in a society.
- More than 95about Interfaces: interfaces with the system, interfaces with the human.
Actual algorithms are very few,
heuristics are at the same time few and too many,
because the environment makes them unreliable.
Interfaces can and should be semi-automatically deduced.
- More generally, the problem with existing systems is lack of reflectivity,
and lack of consistency:
you can't simply, quickly, reliably, automate any kind of programming.
in a way such that system consistency be enforced.
- Persistence is necessary for AI:
Intelligence is the fruit of a long tradition.
Even a most intelligent and precocious human being
must be carefully bred for years
before yielding the faintest result.
- How could you expect a machine to become intelligent
as soon as it is built and powered-up,
or even after being powered-up for some hours, or some days ?
- computers currently do not allow any information to persist reliably
more than a few months, and won't translate information from old software
to newer ones.
- Hence, artificial intelligence is not possible with existing architecture.
- However, systems with persistent memory could be a first step toward AI.
- unindustrialized countries:
the low reliability of power feeds make resiliant persistency a must.
- Why are existing OS so bad ?
For the same reason that ancient lore is completely irrelevant in
At a time when life was hard, memories very small and expensive,
development cost very high, people had to invent hacker's techniques to
survive; they made arbitrary decisions so survive with their few resources;
They behaved dirtily, and thought for the short term.
- They just had to.
- Now, technology has always evolved at an increasing pace.
What was experimental truth is always becoming obsolete,
and good old recipes are becoming out of date.
Behaving cleanly and thinking for the long term is made possible.
- It is made compulsory.
- The problem is, most people don't think, but blindly follow traditions.
They do not try to distinguish what is truth and what is falsehood
in traditions, what is still true, and what no longer stands.
They take it as a whole, and adore it religiously,
sometimes by devotion, most commonly by lack of thinking,
often by refusal to think,
rarely but already too often by a hypocrit calculus.
Thus, they abdicate all their critical faculties,
or use it against any ethics.
As a result, for the large majority of honest people,
their morals are an unspeakable burden, mixing common sense,
valid or obsolete experimental data, and valid, outdated, or false rules,
connected and tangled in such a way that by trying to extract something
valid, you come up with a mass of entangled false things that are associated,
and that when extirping false things, you often destroy the few that were
The roots of their opinions are not in actual facts, but in lore,
hence their being only remotely relevant to anything.
- Tunes intends to rip off all these computer superstitions.
A.19 Down to actual OSes
A.20 Humanly characteristics of computers
persistence, resilience, mobility, etc....
response to human
- The Internet is a progress,
in that people can publish documents.
But these documents are mostly passive.
Those that are not suppose highly-qualified specialists to care about;
A.21 Multiplexing: the main runtime activity of an OS
Putting aside our main goal, that is, to see how reuse is possible in
general, let us focus on this particular multiplexing technique, and see
what lessons we can learn that we may generalize later.
Multiplexing means to split a single communication line or some other
resource into multiple sub-lines or sub-resources, so that this resource
can be shared between multiple uses. Demultiplexing is recreating a single
line (or resources) from those multiple ones; but as dataflow is often
bi-directional, this reverse step is most often unseparable from the first,
and we'll only talk about multiplexing for these two things. Thus,
multiplexing can be used to share a multiple functions with a single stream
of calls, or convertly to have a function server be accessed by multiple
Traditional computing systems often allow multiplexing of some
resources, thus spliting them into a first (but potentially very large) level
of equivalent logical resources. For example, a disk may be shared with a
file-system; CPU time can be shared by task-switching; a network interface
is shared with a packet-transmission protocol. Actually, what any operating
system does can be considered multiplexing. But those same traditional
computing systems do not provide the same multiplexing capability
for arbitrary resource, and the user will eventually end-up with having to
multiplex something himself (see the term user-level program to multiplex a
serial line; or the screen program to share a terminal; or window systems,
etc), and as the system does not support anything about it, he won't do it
the best way, and not in synergy with other efforts.
What is wrong with those traditional systems is precisely that they
only allow limited, predefined, multiplexing of physical resources
a small, predefined, number of logical resources; there they create
difference between physical resources (that may be multiplexed), and
logical ones (which cannot be multiplexed again by the system). This gap
is completely arbitrary (programmed computer abstractions are never purely
physical, neither are they ever purely logical); and user-implemented
multiplexers must cope with the system's lacks and deficiencies.
- More generally, in any system, for a specialized task,
you may prefer dumb workers that know well their job
to intelligent workers that that cost a lot more,
and are not so specialized.
But as the tasks you need to complete evolve, and your dumb workers don't,
you'll have to throw them away or pay them to do nothing
as the task they knows so well is obsolete;
they may look cheap, but they can't adapt,
and their overall cost is high for the little time when they are active;
In a highly dynamic world, you lose at betting on dumbness,
and should invest on intelligence.
whereas with the intelligent worker,
you may have to invest in his formation,
but will always have a proficient collaborator after a
short adaptation period.
After all, even the dumb worker had to learn one day,
and an operating system was needed as a design platform for any program.
- People tend to think statically in many ways.
- At the time when the only metaprogramming tool was the human minds
of specialized engineers, because memories were too small,
which is very expensive and cannot deal with too much stuff at once,
a run-time hardware protection was wishable to prevent bugs in
existing programs from destroying data,
even though th
But now that computers have enough horsepower to be useful metaprogrammers,
the constraints change completely.
- Dispell the myth of "language independence",
particularly about OSes.
which really means "interfaces to many language implementations";
any expressive-enough language can express anything you expressed
in another language in many ways.
- And as the RISC/CISC then MISC/RISC concepts showed,
the best way to achieve this is to keep the low-level things
as small as possible, so as to focus on efficiency,
and provide simple (yet powerful enough) semantics.
The burden of combining those low-level things into useful
high-level objects is then moved to compilers,
that can do things much better than humans,
and take advantage of the simpler low-level design.
- Now, the description could be restated as:
"project to replace existing Operating Systems, Languages,
and User Interfaces by a completely rethough Computing
System, based on a correctness-proof-secure
higher-order reflective self-extensible fine-grained
distributed persistent fault-tolerant version-aware
decentralized (no-kernel) object system."
- GC&Type checking need be in developing version,
not forcibly in developed version.
- Nobody should be forced by the system itself
into proving one's program correctness with respect to any specification.
Instead, everyone is enabled to write proofs,
and can require proofs from others.
Thus, you can know precisely what you have and what you don't
when you run code.
When the code is safe, you know you can trust it.
When it ain't, you know you shouldn't trust it.
Surely, you will object that because of this system,
such man will now require you to give a proof that you
can't or won't give to him, so NOW you can't deal with him anymore.
But don't blame it on the system.
If the man wants the proof,
it means he'd expected your provided software to behave
accordingly in the past, but just couldn't require a proof,
which was impossible.
By dealing with the man, you'd have been morally and/or legally
bound to provide the things that he now asks a proof for.
Hence the proofable system didn't deprive you from making
any lawful thing. It just helped formalize what is lawful
and what isn't.
If the man requires so difficult proofs that he can't find any
provider to that, he will have to adapt, die, or pay more.
If the man's requirements are outrageously excessive,
and no-one should morally provide him the proofs,
then he obviously is a nasty fascist pig, or whatever,
and it's an improvement that no-one will now deal with him.
To sum up things,
being able to write/read/provide/require proofs
means being able to transmit/receive more information.
This means that people can better adapt to each other,
and any deal that the system will cancel was an unlawful deal,
replaced by better deals.
Hence this technology increases the expressivity of languages,
and does not decrease it.
The system won't have any statical specification,
but will be a free market for people having specifications
and people having matching software to safely exchange
code against money,
instead of being a blind racket.
- People like that the cryptic Perl syntax be ambiguous
and guess what you mean from context,
because it allows rapid development of small programs,
and Perl usually guesses right
what you want it to do.
other people will object that because your programs
will then depend on guesses, you can't reliably develop
large programs and be confident that you don't depend
on a guess that may prove wrong in certain conditions,
or after you modify your program a bit.
But why should you depend on dynamic guesses?
A Good programming language would allow you
Of course, all this require a much more reflective platform than we have,
with interactive tools being much more integrated to the compiler than
to control how the guesses are done,
enable some tactics, disable some others, and write your own.
- to make the guesses explicitly appear or disappear in the program
by automatic semantic-preserving source-to-source transformations.
- resolve all the ambiguities in a static way through some
interactive tool, with a reasonable guess as the default, but with the
program's source being statically disambiguated by the machine.
- an open system,
where computational information can efficiently flow
with as little noise as possible.
Open system means that people can contribute any kind of information
they want to the available cultural background,
without having to throw everything away and begin from scratch,
because the kind of information they want to contribute does not
fit the system.
Example: I can't have lexical scopes in some
wordprocessor spell-checker, only one "personalized dictionary"
personalized at once (and even then, I had to hack a lot
to have more than one dictionary,
by swapping a unique global dictionary). So bad.
I'll have to wait for next version of the software.
Because so few ask for my feature, it'll be twenty years until
it makes it to an official release. Just be patient.
Or if I've got lots of time/money, I can rewrite the whole
wordprocessor package to suit my needs. Wow!
On an open system, all software components must come in small grain,
with possibility of incremental change anywhere,
so that you can change the dictionary-lookup code to handle
multiple dictionaries merged by scope, instead of a unique global one,
without having to rewrite everything.
Current attempts to build an open system have not been fully successful.
The only successful approach to offer fine-grained control on objects
has been to let sources freely available,
allowing independent hackers/developers to modify and recompile;
but apart from the object grain problem,
this doesn't solve the problems of open software.
Other problems include the fact
This offers no semantical control of
seamless data conservation accross code modification;
contributions are not really incremental in that
the whole software must be integrally recompiled, stopped, relaunched;
Changes that involve propagation of code among the whole program
cannot be done incrementally with non
- "as little noise as possible": this means
that algorithmic information can be passed
without any syntactical or architectural constraint in it
that would not be specifically intended;
that people are never forced to say
either more than they mean or less than they mean.
Example: with low-level languages like C,
you can't define a generic function to work on any integer,
then instanciate to the integer implementation that fits
the further problem.
If you define a function to work on some limited number type,
then it won't work on longer numbers than the limit allows,
while being wasteful when cheaper more limited types might have been used.
Then if some 100000 lines after, you see that after all,
you needed longer numbers, you must rewrite everything,
while still using the previous version for existing code.
Then you'll have two versions to co-debug and maintain,
unless you let them diverge inconsistently, which you'll have to document.
So bad. This is being required to say too much.
And of course, once the library is written,
in a way generic enough so it can handle the biggest numbers you'll need
(perhaps dynamically sized numbers),
then it can't take advantage of any particular situation
where the known constraints on numbers
could save order of magnitudes in computations;
of course, you could still rewrite yet another version of the library,
adapted to that particular knowledge,
but then you again have the same maintenance problems as above.
This is being required to say too little.
Any "information" that you are required to give the system
before you know it, without your possibly knowing it,
without your caring about it,
with your not being able to adjust it when you further know more,
all that is *noise*.
Any information that you can't give the system,
because it won't heed it, refuse it as illegal,
implement in so inefficient a way that it's not usable,
is *lack of expressiveness*.
Current languages are all very noisy and inexpressive.
Well, some are even more than others.
The "best" available way to circumvent lack of expressiveness from
available language is known as "literate programming",
as developed, for example, by D.E.Knuth with his WEB and C/WEB packages.
With those, you must still fully cope with the noise of a language like C,
but can circumvent its lack of expressiveness,
by documenting in informall human language
what C can't express about the intended use for your objects.
Only there is no way accurately verify that
objects are actually used consistently
with the unformal documented requirements,
which greatly limits the (nonetheless big) interest of such techniques;
surely you can ask humans to check the program for validity
with respect to informal documentation,
but his not finding a bug could be evidence
for his unability to find a real bug,
as well as the possible absence of bug,
or the inconsistency of the informal documentation.
This can't be trusted remotely as reliably as a formal proof.
The Ariane V spacecraft software
had been human-checked thousands of times against informal documentation,
but still, a software error would have $ 109 disappear in fumes;
from the spacecraft failure report,
it can be concluded that the bug
(due to the predictable overflow
of an inappropriately undersized number variable)
could have been *trivially* pin-pointed by formal methods!
Please don't tell me that formal methods are more expensive/difficult
to put in place than that the rubbish military-style red-tape-checking
that was used in place.
As a french taxpayer,
I'm asking immediate relegation of the responsible egg-heads
to a life-long toilet-washing job
(their status of french "civil servants" prevents their being fired).
Of course my voice is unheard.
Of course, there are lots of other software catastrophes
that more expressive languages would have avoided,
but even this single 10 G$ crash would pay more than
it would ever cost to develop formal methods
and (re)write all critical software with!
- It is amazing that researchers in Computer Science are not
developing branches of a same software,
but everytime rewriting it all from scratch.
For instance, people experimenting with persistence, migration,
partial evaluation, replication, distribution, parallelization, etc,
just cannot write their part independently from the others.
Their pieces of code cannot combine,
and if each isolated technical point is proven possible,
they are never combined,
and techniques can never really be fairly compared,
because they are never applicable to the same data.
It is as if mathematicians would have to learn a completely new language,
a completely new formalism, a completely new notation,
for every mathematical theory!
As if every book couldn't actually assume results from other books,
unless they were proven again from scratch.
It is as if manufacturers could not assemble parts
unless they all came from the same factory.
Well, such phenomena happen in other place than computer software, too.
Basically, it might be conceived as a question of lack of standards.
But it's much worse with computer software,
because computer software is pure information.
When software is buggy, or unable to communicate,
it's not worth a damn;
it ain't even good as firewood, as metal to melt or anything to recycle.
Somehow, the physical world is a universal standard for the use
of physical objects.
There's no such thing in the computer world
where all standards are conventional.
Worse, progress in hardware and software implementation techniques
is incompatible with the advent of definitive computerware standards,
so that either standards are made ephemeral,
or implementational progress is throttled.
And the solution is Reflection.
- The other day, I tried to explain what Reflection is
to a mathematician friend of mine.
But Reflection is so natural a thing for mathematicians
(and my math background is perhaps what makes it hard for me
to live without it in the computer world),
that I could only try to describe what lack of Reflection would be in math:
It would mean that you could only combine theorems that were not
developped with the very same formalism.
For instance, you would not even be able to apply to classic results,
unless you could provide some actual derivation of both well-known theorems
from scratch using the same formalism,
which for a mathematician would mean
a conventional minimalistic theory
like peano's axioms for arithmetics, or some flavor of set theory.
This very notion of "scratch" will seem silly to a mathematician,
as he knows from Goedel that there is no absolute "scratch"
from which to build mathematics,
so that theorems should instead be produced in
whatever form seems the most adequate for its current use
(its being proved or reused, etc).
The computer engineer would then say that "scratch" is the
actual operating software/hardware he's got to work on NOW,
which is very concrete.
But this notion of scratch should be silly to him, too,
if only he were conscious how fast hardware technology evolves,
and building from current "scratch"
only ties his programs to current technology,
preventing computerware upgrade, or limiting the benefits thereof.
Reflection does allow to refine implementations,
to move between standards,
to prove new meta-theorems and use them,
to juggle between representations so as to pick up
the most adequate for a given task
without sacrificing consistency,
to reuse (meta)theorems from other people,
A.22 Miscellaneous notes
- I saw your answer about an article in the news, so i wanna know,
what is Tunes ?
Well, that's a tough one.
Here is what I told Yahoo:
"TUNES is a project to replace existing Operating Systems, Languages,
and User Interfaces by a completely rethought Computing System,
based on a correctness-proof-secure
higher-order reflective self-extensible fine-grained
distributed persistent fault-tolerant version-aware
decentralized (no-kernel) object system."
Now, there are lots of technical terms in that.
Basically, TUNES is a project that strives to develop a system where
computists would be much freer than they currently are:
in existing systems, you must suffer the inefficiencies of
If any of these seems unclear to you, I'll try to make it clearer in
centralized execution [=overhead in context switching],
- centralized management [=overhead and single-mindedness in decisions],
- manual consistency control [=slow operation, limitation in complexity],
- manual error-recovery [=low security],
- manual saving and restoration of data [=overhead, loss of data],
- explicit network access [slow, bulky, limited, unfriendly, unefficient,
wasteful distribution of resource],
- coarse-grained modularity [=lack of features, difficulty to upgrade]
- unextensibility [=impossibility to do things oneself,
people being taken hostage by software providers]
- unreflectivity [=impossibility to write programs clean for both human
and computer; no way to specify security]
- low-level programming [=necessity to redo things again everytime one
- Note that Tunes does not have any particular technical aim
any particular technique intended for inclusion in the system
has most certainly already
been implemented or proposed by someone else already,
even if we can't say where or when.
Tunes does not claim any kind of technical originality.
Tunes people are far from being the most proficient in
any of the technical matters that they'll have to use,
and hope that their code will be eventually replaced by better
code written by the best specialists wherever applicable.
But Tunes is not an empty project for that.
Tunes does claim to bring some kind of original information,
just not of a purely technical nature,
as a global frame to usefully combine
those various techniques as well as arbitrary future ones
into a coherent system,
rather than have them stay idle gadgets that can't reliably
communicate with each other.
We Tunes people hope that our real contribution
will be the very frame in which the code
from those specialists can positively combine with each other,
instead of being isolated and helpless technical achievements.
Even if our frame doesn't make it into a worldwide standard,
we do hope that our effort will make such a standard appear
sooner than it would have without us (if it ever would),
and avoid the traps that we'll have uncovered.
- In this article,
we have started from a general point of view of moral Utility,
and by applying it to the particular field of computing,
we have deduced several key requirements
for computing systems to be as useful as they could be.
We came to affirm concepts like
dynamism, genericity, reflectivity, separation and persistency,
which unhappily no available computing system fully implements.
So to conclude,
there is essentially one thing that we have to fight:
the artificial informational barriers
lack of expressivity
of former software,
due misknowledge, misunderstanding, and reject
of the goals of computing,
computer objects and others computer objects,
computer objects and human beings,
human beings and other human beings.
- People are already enough efficiency-oriented so that
TUNES needn't invest a lot in it,
just providing a general frame for others to insert optimization into.
In the case of a fine-grained dynamic reflective system,
this means that hooks for dynamic partial evaluation must be provided.
This is also an original idea, that hasn't been fully developed.
contribution that TUNES
- When confronted with some proposition in TUNES,
people tend to consider it separated from the rest of the TUNES ideas,
and they then conclude that the idea is silly,
because it contradicts something else in the traditional system design.
These systems indeed have some coherency, which is why they survived and
were passed by tradition. But TUNES tries to be much more coherent even,
- When I begun this article, long ago, I believed that multiplexing
was the main thing an OS would do. Now, I understand that the main thing
is trust. Multiplexing can be readily done with a powerful language
(ok, OSes are not currently powered by such languages, so that multiplexing
is a system-level problem, with them!)
- Not seeing the importance of TRUST, I didn't at the time realize
the effect of proprietary vs free software issues in the design of the system.
Indeed, closed vs open source has a great impact on the dynamics of
trust-building, on the need to have features and multiplexing in the "kernel";
on the separation between users and programmers, etc, etc.
- About reuse and copy/paste:
copy paste, of course, is evil.
that's precisely why I cite it as the simplest and dumbest way.
It's the way we use when better ways are available. We all use it a lot.
Making other ways to reuse code difficult just makes us use this one,
with all maintainability nightmare and development cost that this induces.
Of course there's even worse. In a language like unlambda
(voluntary designed for obfuscation, yet based on nice theory),
you mostly cannot even copy/paste code then insert modifications
to do "simple" things like add a variable (well, then reason you cannot,
and how it generalizes to other languages is interesting,
and deserves treatment).
Not to talk about binary executable objects,
that are typically a language where you have a hard time copy/pasting
routines (a reason why we use assemblers and symbolic languages).
- Algorithms provide trust in the well-defined behaviour of a program:
the systematic coverage of every case in some space,
the strict abiding by rules known in advance,
the controlled usage of space and time resources,
are valuable meta-level knowledge about a program.
``AI'' techniques on the other hand, attempt to be somewhat creative,
and hence unpredictable, and by definition,
try to destroy any such meta-level knowledge,
although in most cases, we like to have some meta-meta-level knowledge
about the goals that the AI seek, and some double-check
on the fulfillment of these goals.
There are many cases such meta-level knowledge is necessary
and where ``AI'' kind of techniques just cannot be used within a program,
or can only be used with a algorithmic backup plan,
when it is possible to provide such a plan
(for instance, any somewhat real-time control problem
might like to use AI to find interesting ``optimized'' solutions,
but will require an algorithm guaranteed to give a sensible response
in case the AI doesn't provide a satisfying answer in time).
However, just because a program's run-time must be somewhat algorithmic
doesn't mean that AI cannot be used during development-time,
Hopefully, AI can provide unpredictable creative help
in generating predictable stubborn algorithms.