suggestion for Python

[=?iso-8859-1?Q?Fran=E7ois=20Miville=2DDech=EAne?=]

I find your language very nice, it is actually more than three quarters
of what I had been dreaming for years a programming language should be.
But I mourn the passing of APL, another interpreted language. I like
interpreted languages, for they can be used in calculator mode, testing
function after function without bothering to code long protocols in the
language and outside it.

One suggestion I would propose would be to make common operators
distributive by enclosing them in square brackets. For instance:
[1, 3, 5, 7] [*] [1, 2, 1, 3] would be equivalent to
[[1, 3, 5, 7] * [1, 2, 1, 3] for i in range(3)]
In APL the common operators were implicitly distributive over vectors,
but it is not possible for Python due to the more evident concatenation
meaning of + and *. The map function does the job for a user
defined-function, as would do in this instance something like
map(multiply, [1, 3, 5, 7], [1, 2, 1, 3]), but it is cumbersome to
redefine simple operations.

The same brackets enclosing the operator with a point could be used to
define outer product: [1, 3, 5] [*.] [1, 3, 2] would be equivalent to
the result:
[[1, 3, 5], [3, 9, 15], [2, 6, 10]] (a regular matrix shorter to list
than to express in the Python way). The same brackets enclosing two
operators separated by a point would do the job of the vector or matrix
product:
[1, 3, 5, 7] [+.*] [1, 2, 1, 2] would thus be
sum (map(multiply, [1, 3, 5, 7], [1, 2, 1, 2]))

I give very straightforward examples, but it is easy to imagine that
those operators thus enclosed would not only apply to simple lists, but
to matrices and tree-like structures defined as lists of lists, provided
they have a common structural basis.

One problem with Python is the necessity to define very short functions
to enable operators to be the object of function of functions such as
map. Another problem is that common operators cannot be easily
redifined onto larger mathematical objects one might conceive and could
be defined by the means of classes. This could be solved by setting
that such functions named as add, mult, div, defined as two-parameter
functions, would automatically correspond to the dyadic +, *, /, etc.

And conversely, any name of a two-parameter function, let it be
hypothenuse, would signify when enclosed in parentheses the
corresponding dyadic operators.

so that 3(hypothenuse)4 would be 5

This would give Python all the advantages of late APL and much more, for
APL could process only regular data structures such as matrices whereas
Python is so good with data more to be encountered in real life such as
ill-formatted texts to be divided in chapters, sections and subsections
in order to fit in a certain software, just to name one intance, by the
means of the use of reg exps and tree-processing.

Thanks for your attention. The sole objection to the development I
propose for Python would be a certain loss of readability for persons
not acquainted with the language, but reg exps are not that readable
neither and yet what would we do without them?

 

[John Roth]

It's an interesting thought, certainly. I like the conciseness
involved. APL certainly looked like an interesting language,
although I never took the opportunity to really play with it.

Have you looked at the NumPy extension? It does a
great deal in the area of mathematical functionality.

John Roth

I find your language very nice, it is actually more than three quarters
of what I had been dreaming for years a programming language should be.
But I mourn the passing of APL, another interpreted language. I like
interpreted languages, for they can be used in calculator mode, testing
function after function without bothering to code long protocols in the
language and outside it.

One suggestion I would propose would be to make common operators
distributive by enclosing them in square brackets. For instance:
[1, 3, 5, 7] [*] [1, 2, 1, 3] would be equivalent to
[[1, 3, 5, 7] * [1, 2, 1, 3] for i in range(3)]
In APL the common operators were implicitly distributive over vectors,
but it is not possible for Python due to the more evident concatenation
meaning of + and *. The map function does the job for a user
defined-function, as would do in this instance something like
map(multiply, [1, 3, 5, 7], [1, 2, 1, 3]), but it is cumbersome to
redefine simple operations.

The same brackets enclosing the operator with a point could be used to
define outer product: [1, 3, 5] [*.] [1, 3, 2] would be equivalent to
the result:
[[1, 3, 5], [3, 9, 15], [2, 6, 10]] (a regular matrix shorter to list
than to express in the Python way). The same brackets enclosing two
operators separated by a point would do the job of the vector or matrix
product:
[1, 3, 5, 7] [+.*] [1, 2, 1, 2] would thus be
sum (map(multiply, [1, 3, 5, 7], [1, 2, 1, 2]))

I give very straightforward examples, but it is easy to imagine that
those operators thus enclosed would not only apply to simple lists, but
to matrices and tree-like structures defined as lists of lists, provided
they have a common structural basis.

One problem with Python is the necessity to define very short functions
to enable operators to be the object of function of functions such as
map. Another problem is that common operators cannot be easily
redifined onto larger mathematical objects one might conceive and could
be defined by the means of classes. This could be solved by setting
that such functions named as add, mult, div, defined as two-parameter
functions, would automatically correspond to the dyadic +, *, /, etc.

And conversely, any name of a two-parameter function, let it be
hypothenuse, would signify when enclosed in parentheses the
corresponding dyadic operators.

so that 3(hypothenuse)4 would be 5

This would give Python all the advantages of late APL and much more, for
APL could process only regular data structures such as matrices whereas
Python is so good with data more to be encountered in real life such as
ill-formatted texts to be divided in chapters, sections and subsections
in order to fit in a certain software, just to name one intance, by the
means of the use of reg exps and tree-processing.

Thanks for your attention. The sole objection to the development I
propose for Python would be a certain loss of readability for persons
not acquainted with the language, but reg exps are not that readable
neither and yet what would we do without them?

 

[Erik Max Francis]

I find your language very nice, it is actually more than three
quarters
of what I had been dreaming for years a programming language should
be.
But I mourn the passing of APL, another interpreted language. I like
interpreted languages, for they can be used in calculator mode,
testing
function after function without bothering to code long protocols in
the
language and outside it.

The meat of this suggestion is what PEP 225 is about:

http://www.python.org/peps/pep-0225.html

 

[Alex Martelli]

I find your language very nice, it is actually more than three quarters
of what I had been dreaming for years a programming language should be.
But I mourn the passing of APL, another interpreted language. I like

Having worked a LOT with both APL and APL2 around 1977-1986, I don't
really mourn them at all (except when I'm in my cups, sometimes, but
then I'm mostly mourning the fact that 20 years ago I was younger.

One suggestion I would propose would be to make common operators
distributive by enclosing them in square brackets. For instance:
[1, 3, 5, 7] [*] [1, 2, 1, 3] would be equivalent to
[[1, 3, 5, 7] * [1, 2, 1, 3] for i in range(3)]
In APL the common operators were implicitly distributive over vectors,
but it is not possible for Python due to the more evident concatenation
meaning of + and *. The map function does the job for a user
defined-function, as would do in this instance something like
map(multiply, [1, 3, 5, 7], [1, 2, 1, 3]), but it is cumbersome to
redefine simple operations.

import operator
map(operator.mul, [1,3,5,7], [1,2,1,3])

and you're done. Or, to taste,

[ a*b for a, b in zip([1,3,5,7], [1,2,1,3]) ]

One problem with Python is the necessity to define very short functions
to enable operators to be the object of function of functions such as

Yes, we do lack a _good_ syntax for "codeblock literals" -- about a
month ago we were discussing possibilities quite actively here.

map. Another problem is that common operators cannot be easily
redifined onto larger mathematical objects one might conceive and could
be defined by the means of classes. This could be solved by setting
that such functions named as add, mult, div, defined as two-parameter
functions, would automatically correspond to the dyadic +, *, /, etc.

Standard library module operator has those. But I'm not sure what you
mean by "redefined onto larger mathematical objects" -- you can get
operator overloading on any type by defining __add__ __mul__ etc.

And conversely, any name of a two-parameter function, let it be
hypothenuse, would signify when enclosed in parentheses the
corresponding dyadic operators.

so that 3(hypothenuse)4 would be 5

Yep, Haskell has something like that, except that instead of
yet another overloading for parentheses (which have too many
already!-) it uses `...` which in Python are an unfortunate and
useless duplciate for repr(...). However to make sense you need
to be able to define priority and associativity and that gets hairy.

This would give Python all the advantages of late APL and much more, for
APL could process only regular data structures such as matrices whereas

APL2 could have arrays with each row a different length, which helped.

Python is so good with data more to be encountered in real life such as
ill-formatted texts to be divided in chapters, sections and subsections
in order to fit in a certain software, just to name one intance, by the
means of the use of reg exps and tree-processing.

Thanks for your attention. The sole objection to the development I
propose for Python would be a certain loss of readability for persons
not acquainted with the language, but reg exps are not that readable
neither and yet what would we do without them?

They're in a separate standard library module -- the criteria are laxer.
Getting into core Python demands passing MUCH stricter tests on all scores.


Alex