Rainer Glaschick, Paderborn
2025-01-14
Corresponds to version 0.10.3
This is a short tutorial for the programming language TAV-PL; for details, see the language specification and the documentation for the standard library functions.
TAV-PL is a precompiler for the C programming language, thus it is fairly easy to port the compiler — written in TAV-PL — to most computers.
The differences and highlights are:
The example programs may be tested online at http://tavpl.de/TAV-Online.
The term digraph is used for two special characters without white space in between.
All programs need to include the definitions of the standard library, as shown in the first example. For easier reading, it will be omitted from the other examples.
Here is the inevitable program that just prints a message:
\ the famous sample program prints just a greeting main (parms):+ print "Hello, world!" \+ stdlib
producing the output
Hello, world!
First of all, the backslash exclusively starts comments and pragmas (that control the compiler); in our example, they start in column 1 (remember that the examples are indented for readability).
As in C, the main program is called main with the parameter map parms,
that will be explained later.
A function definition like main (parms) always starts in column 1 and
ends in a colon; here with the digraph :+ to make it externally
visible.
The main program has only one statement, that must be indented, i.e. not
start in column 1 of a line.
Here, it is a call to the standard library function print with one parameter,
the string "Hello, world!".
The missing parenthesis around the parameter are correct; you will soon observe — and hopefully appreciate — the few situations in which (balanced) parenthesis are needed.
The last line includes from a file the templates (i.e. definitions)
of the library functions used, print () in this case.
You may select a different library, e.g. one using French:
\ célèbre programme pour émettre un salut main (paramètres):+ tirer "Bonjour, le monde!" \+ stdlib.fr
Please supply one in Russian, Italian, Spanish or whatever you like.
TAV-PL is fully UTF-aware, and it's easy, as there are no keywords; just translate the function templates.
As the very first computers were made to compute numbers, they could not print character strings, just numbers. Thus, printing a table of squares is one of the earliest programming tasks:
\( squares: Print squares of numbers
\)
main (parms):+
i =: 1
?* i < 11
print i __ i*i
i =+ 1
\+ stdlib
The output should be:
1 1 2 4 3 9 4 16 5 25 6 36 7 49 8 64 9 81 10 100
First, the block comment is a bit unusual, using \( to open and \)
to close; but as said before, a backslash is only used for comments,
and everything that starts with a backslash is a comment or a pragma.
The first line is an assignment that puts the integer number 1 into the variable i;
but — as PASCAL and ALGOL users might appreciate, and C and JAVA programmers
might find at least funny — the assignment symbol
is not just the equals sign, but the digraph =: (two special characters
glued)1.
In contrast to ALGOL and PASCAL, the symbol starts with an equals sign,
because there are more assignment symbols like =+ and =-.
The next line is a simple loop that repeats while the expression i<11
is true.
The body of the loop is again indented, two lines to print and to increment
the index i.
There are no parenthesis needed to indicate the parameters for the print function; which is no special case. Parenthesis are only required to ensure correct grouping.
To compose the print line, the underscore is used to combine strings.
While in general numbers and strings may not be used interchangeably,
the print function as well as the catenation operator denoted by the
underscore _ use a built-in conversion function to convert any number
to a string in a standard format.
Thus, the print function still sees only one parameter, and this
is the string expression i __ i*i.
(The double underscore inserts a blank, while the single underscore glues
two strings tightly together).
Finally, we have to increment the number in i.
For this, the increment variant of an assignment is used, which is
actually a shortcut for
i =: i + 1
Of course, any expression that provides an integer value is possible. Note again the inversion over the increment in C, in order that all assignments start with an equals sign. Moreover, the amount to be added is after the plus character.
There is no such thing as the ++ operator in C and JAVA.
While I loved it for the compact and tricky programming it made possible
when I learned C 30 years ago, it makes programs harder to understand.
At that time the performance gains might have been valuable,
but the risk of unreliable and buggy programs is much larger.
As some sensible programming instructions wrote:
Avoid tricky programming; the compiler knows the rules exactly,
but you as a human might not.
Of course, the normal way to write the body is:
?# i =: from 1 upto 10 print i, i*i
using the scan function from () upto () to provide integers as
a number generator.
Scan functions (also called enumerations) are simple yet powerful
and can generate all kinds of items.
Python users should observe that its range() function excludes
the upper bound.
Also, the print statement accepts a list of expressions, the numbers are delimited by a blank space.
You might already know that old-fashioned programmers like me use a command line interface that allows to provide parameters.
The following program prints these parameters:
\ printparms: Just echo the parameters
main (parms):+
?# i =: from 1 upto parms.Last
print i, parms[i]
print "Total: ", i - 1
\+ stdlib
If run with the parameters A bc 9, the result should be:
1 A 2 bc 3 9 Total: 3
As there are many parameters possible, the command line shell creates a
list of them, which the TAV-PL runtime converts to a table (a row),
which is the parameter of main().
Again a scan function is used to deliver integer numbers, this time
up to the value of the expression parms.Last,
which is the .Last attribute (property) of parms, the largest index.
After the printed list, the total number of lines before is printed; as this scan provides the next (out of bounds) integer on termination, we have to subtract 1.
In the next example, the squares are calculated without multiplications, and their number is given as a parameter:
main (parms):+
num =: string parms[1] as integer else 10
\ squares by x² = (x-1)² + 2x - 1
x =: 0
sum =: 0
?# i =: from 1 upto num
x =+ i + i - 1
print i, x
sum =+ x
print "sum=" _ sum
\+ stdlib
giving
1 1 2 4 3 9 4 16 5 25 6 36 7 49 8 64 9 81 10 100 sum=385
The maximum number to be squared is given as a parameter on the command
line; this is a string which must be converted to a (binary) integer number
using the library function string (input) as integer else (default).
So its time to explain now the probably most intriguing feature of TAV-PL: function calls do not need parenthesis. 2
Each function is a template of words and parameters, separated by blanks; the template for the above function written as:
string (str) as integer else (default):
Looking for a function call, the words in the expression
are compared to the function templates; and if one fits, the scan advances,
which is the case for string, as, integer and else.
Note that these are not reserved words, just words matching the
function template.
If there is no such word, as is the case for the 2nd (parms),
it is checked if there is a parameter symbol, which is a word in parenthesis
in the template.
If it is, an expression is parsed, which is in this case parms[1],
the first string on the command line.
Then, matching resumes matching the words as, integer and else.
Finally, the expression 10 matches the second parameter in the template.
So, in C, the statement would be written as
max = as_integer_else(parms[1], 20);
It is good programming style in TAV-PL to use function templates with more
than one word (required for functions without parameters).
Because it is needed so often, print is one of those few functions
with one word only.
Using print and the other one-word function words as variable names
is discouraged, because the function invocation has higher priority;
to avoid surprises, the compiler will issue a warning.
Variables need not and cannot be declared; the just come into existence by using the variable name; and are initialised to void. Misspelled variable names that are thus void can lead to subtle programming errors; but — with a bit of experience — the source is quickly identified, even if the compiler does not complain about variables that maybe used before initialised, i.e. are not target of an assignment before used in an expression.
There are no type declarations in TAV-PL, any variable can hold any kind of item, and may even change it:
x =: 'x' x =_ 1 \ same as x =: x _ 1 and thus converts integer to string ! x = 'x1'
The last line is an assertion statement, that will stop if the condition is not met. Note that comparison for equality uses a single equals sign.
A program to calculate a square root:
main(parms):+
x =: string parms[1] as float else ()
? x = ()
error print "Parameter " _ parms[1] _ " invalid, use decimal point: 2.0"
:> 1
print "sqrt(" _ x _ ")=", square root x
\( Iterative square root. If the initial value is > 1,
then the values are monotonically falling, no epsilon needed
\)
square root (a):
! a >= 0.0 \ argument must be a positive rounded number
? a %= 0.0 \ nearly equal zero
:> 0.0
? a < 1.0 \ iteration needs a >= 1
a =: 1.0 / a
:> 1.0 / square root a
x =: a
?* \ loop until break
y =: (x + (a/x)) / 2.0 \ new value from iteration
? y >= x \ done if no longer monotonically falling
?> \ stop the loop
x =: y
:> x
and, called with 2.0 as parameter, gives
sqrt(2)=1.41421
Besides numbers, strings and others, there is a special contents of a
variable, called void, that is not a zero, not an empty string, neither false
nor true, just nothing, indicated by a pair of parenthesis (the digraph ()).
(Well, if found where number for addition is expected, it is treated as 0,
but not for multiplications and divisions.
This allows in the first line (of main) to indicate that the parameter string
was not a valid number.
In the next line a plain condition — not a loop — is indicated by just
a question mark, that questions the boolean expression that compares x to void.
If the condition holds, processing continues with the indented block,
giving a message to the error output, thus the function error print ()
is used.
It follows a return with integer 1 as a return code from the whole program.
This time we use a function to do the calculation just to avoid to deeply indented code.
A function definition always starts in column 1, followed by the template, followed by a colon as last on the line. The function body is then just indented as usual.
The first statement is an assertion: an exclamation mark, followed by
a boolean expression. The programmer wants to ensure that the parameter
is not negative, and is a floating point number.
If this is not the case, the program goes to an error stop.
This is also the case if x is not a floating point number, as there are
no automatic conversions between integer and floating point number
and clearly not from strings.
The algorithm used is a bit unusual; you might miss an epsilon to terminate
the iteration. The latter is annoying at least,
as it requires to know the accuracy of the machine.
But if started with a number greater 1.0, the values
are monotonically falling in value,
and finally staying the same or oscillating.
If there is no more progress, the condition y >= x holds,
and the loop is terminated with the break symbol ?>.
The symbols for function return and break are intentionally similar.
When the loop terminates, the value in x is the root, and returned to the caller.
If a negative floating point number is supplied, there will be a failure message like:
Fatal: Assertion failed: ! a >= 0.0 at 12 in square root (a) Called at 6 in main (parms) Called at 5636 in SYSTEM
telling that the co-operation between the function programmer and the main programmer was not close enough. I know of no type system that allows to declare a non-negative floating point number to catch such an error at compile time.
If you enter 0.0 at the command line, you would expect to crash with a division by zero in the line
:> 1.0 / square root 1.0 / a
But modern floating point number arithmetic results in inf for infinity,
and its reciprocal is 0.0.
Instead of using the absolute value in the function, the input check is made better:
x =: string parms[1] as float else ()
? x = () | x < 0.0
error print ...
You might try to use the lines
? x = 0.0
:> 0.0
unaware of the language description telling that equality comparisons for
floating point numbers are dubious and thus not allowed; only order comparisons
or the nearly equal comparison are supported.
(Yes, you may write ~ (x < 0.0 | x > 0.0) to circumvent, signalling any
sensible programmer that reads the code a dubious comparison,
because an argument x very close to zero may fail for 1.0/x)
Here is a version that uses more advanced features:
main (args):+
a, x =: string args[1] as float else 2.0
?# x =- give while positive (x - a/x) / 2.0
()
print format "%'f.16; %f;" x, a - x^2
Integer numbers may be — as in PYTHON — arbitrary long and are not restricted to 64 bits; so there is no integer overflow. If accurate numbers are required that are not whole numbers, rational numbers are supported (currently only as 32 bit numbers, just to prove the concept; will be extended to arbitrary precision):
\ sum up the fractions: 1 + 1/2 + 1/3 + 1/4 + ...
main (parms):+
x =: 0
?# i =: from 1 upto 10
x =+ 1 / i
print i, x, x.float \ or ##x or float x
\+ stdlib
giving
1 1 1 2 3/2 1.5 3 11/6 1.83333 4 25/12 2.08333 5 137/60 2.28333 6 49/20 2.45 7 363/140 2.59286 8 761/280 2.71786 9 7129/2520 2.82897 10 7381/2520 2.92897
The division operator may be used either for a pair of floating point numbers or a pair of accurate numbers, which are integers and rationals (integers are rationals with denominator 1). As can be seen from the 4th line on of the output, common divisors in nominator and denominator are automatically cancelled (and if the denominator is 1, the result is given back as integer number).
The truncated division for integers is //, while the commonly used
computer remainder is %%. See the documentation for reasons and variants.
The already used scan function from () upto () is written in TAV-PL:
from (start) upto (end):
:> from start upto end step 1
from (start) upto (end) step (step):
? $ = () \ first call?
$ =: start \ yes, use start value
|
$ =+ step \ no, increment
rv =: $ \ save
? $ > end | $ < start \ done ?
$ =: () \ yes
:> rv
main(parms):+
?# i =: from 1 upto 12
print i \ prints 1 to 12
print i \ prints 13
\+ stdlib
giving
1 2 3 4 5 6 7 8 9 10 11 12
The first two lines just map the absent upper limit to 1.
In the body for the next function, the special variable $ is
used, called the local context of a scan.
It is set to void at the begin of each scan.
Upon entry to the function, this context variable is checked for void;
if it is, its the first call, and the context variable set to it.
Otherwise, it is incremented.
Next, it is checked if the upper limit last has been exceed;
if yes, the context variable is cleared to void.
This signals the calling loop control that the loop is terminated,
and the body of the loop, i.e. the print, is now skipped.
Because the check takes place after the assignment, the return value,
which is the next value, is assigned to i.
This is necessary in case the scans ends prematurely and returns
a fault item to indicate this case.
Dropping the line rv =: $ and returning $ supplies void
to i when the loop is ended.
Note that the scan can also be used to supply floating point or rational numbers:
main(parms):+
?# i =: from 1 upto 2 step 1/10
print i
\+ stdlib
gives
1 11/10 6/5 13/10 7/5 3/2 8/5 17/10 9/5 19/10 2
Scan functions look a bit clumsy because their initialisation takes space,
and the context variable $ is unusual,
but are much easier to implement in a precompiler and more efficient
than true coroutines.
Surely scan functions can nested, e.g. to supply a multiplication table:
main (parms):+
?# i =: from 1 upto 9
?# j =: from 1 upto 9
print (pad left i*j to 3) nonl
print ''
\+ stdlib
gives
1 2 3 4 5 6 7 8 9 2 4 6 8 10 12 14 16 18 3 6 9 12 15 18 21 24 27 4 8 12 16 20 24 28 32 36 5 10 15 20 25 30 35 40 45 6 12 18 24 30 36 42 48 54 7 14 21 28 35 42 49 56 63 8 16 24 32 40 48 56 64 72 9 18 27 36 45 54 63 72 81
The print function print () nonl does not terminate with a line feed
(the parenthesis were written for clarity; they are not necessary).
Note that the pad function accepts either numbers or strings,
and always delivers a string.
A scan function can often be used to deliver less simple integer sequences by using e.g. tuples (pairs, immutable rows) as scan context:
give fibonacci numbers upto (n):
$ =~ 0, 1 \ initial tuple if first call
r =: $.1 + $.2 \ next number
$ =: $.2, r \ save previous and current value
? r >= n
$ =: () \ end of scan
:> r \ return next number in any case
main (p):+
?# f =: give fibonacci numbers upto 100
print f
\+ stdlib
giving
1 2 3 5 8 13 21 34 55 89
To publish a code fragment, it is better to use keywords like
if, while etc instead of symbols, and the single equals
sign for an assignment.
This might also make TUF-PL more accepable.
For this purpose, the \C pragma switches to C mode:
\C
square root (a):
assert a > 0.0 \ argument must be a positive number
if a < 1.0 \ iteration needs a > 1
a = 1.0 / a
return 1.0 / square root a
x = a \ start value
while \ no condition means until break
y = (x + (a/x)) / 2.0 \ next iteration value
if y >= x \ still monotonically falling?
break \ no, root found
x = y
return x
Or:
\C
main (parms):+
for i =: from 1 upto 9
for j =: from 1 upto 9
print (pad left i*j to 3) nonl
print ''
Because string catenation is quite comfortable, composing strings, in particular for printing, can be easily done this way. However, if many numbers must be converted in specific formats, the result is quite bulky.
Interpolated formatting, in particular the well-known printf function, has some advantages has thus be adopted to TAV-PL, not a least the better chance for localisation.
Here, it is a function of its own, which composes a string under control of a format string from the elements of a tuple:
print format "max: %d; %d; steps ( %_d; iterations)" imax, cmax, total
The format string is compatible with C (s)printf, meaning it differs in crucial differences and some extra features, because the conditions due to the language are different. In particular, each format ends with a semicolon.
There are two primary types of aggregates:
As maps may also be addressed by integers, a row is a special kind of item optimised if the keys are integers within a decent range that allows to allocate all cells in the range used actually.
Both are expanded as required, but rows keep every cell ever used (unless manually condensed), while maps keep only the the set of used keys.
To create a table square numbers would read:
collect squares upto (n):
r =: new row
?# i =: from 1 upto n
r[i] =: i*i
:> r
The word new is not an operator, just a convention for a factory function,
and may be different in other localisations.
To count the words of a file, using the function file (f) give words:
count words of (fn):
m =: new map
?# w =: file fn give words
m{w} =+ 1
\ print unsorted:
?# w =: map m give keys
print w, m{w}
:> m.Count
The cells (elements) of rows and maps can be any mix of references to any item, be it an number, string, map or row.
As the primary function of a map is to store items under a key and return them if the same key is used, only items that can be sensibly compared for equality are accepted as keys. This excludes floating point numbers, which are equal only within an application dependent difference.
For both, the usual notion like row[integer] or map{string} is used.
To remind the programmer to the difference of maps and rows,
maps are indexed by curly braces.
Rows and maps may additionally have any number of fields,
denoted by the common field notation using dots,
i.e. map.field and row.field.
They are attached dynamically to a map or row with a name
given at compile time; similar to static map elements.
As an example, the maximum could could be kept in a field and returned for further processing:
count words of (fn):
m =: new map
m.max =: 0
?# w =: file fn give words
m{w} =+ 1
? m{w} > m.max
m.max =: m{w}
:> m.max
Field names should start with a small letter by convention.
Some metadata of items can be obtained using field notation.
The number of actually used cells in a map or row is obtained
by using the word .Count in field notation:
m =: new map
m{1} =: 3.5
m{'y’} =: 17
! m.Count = 2 \ field notation
If a attribute can be the target of an assignment to change an item, the are called controls.
Attributes and controls start with a capital letter; in old code, they may be used with a small one.
As the names of attributes and controls are always available in English, they are not localisation agnostic as such, but as the namespace is separate, localisation aliases may be provided.
While TAV-PL is a rather conventional language — except the function templates — it has some features that make it easier to follow an object oriented design.
As a syntactical aid, the first parameter may be the single scroll symbol(@),
same as this in other languages:
integer (@) square:
:> @^2
No restriction is imposed on this;
but normally the use of the scroll (@) indicates that a map (or row)
is used as first parameter, that can be changed inside:
new my bunch:
r =: new map
r.Tag =: 'my bunch'
:> r
my bunch (@) set attr (val):
! @.tag = 'my bunch'
@.attr =: val
my bunch (@) get attr:
! @.tag = 'my bunch'
:> @.attr
Still, the example just obeys the convention that objects are created using factory functions, and the objects are tagged by a sequence of words that are used a the first words upto the first parameter, denoted by the scoll symbol.
Nothing changes so far if any other word, e.g. this, is used instead of
the scoll symbol.
With these conventions, an object oriented design could be programmed fairly clear without any special language features.
As there is no automatic conversions of accurate to rounded numbers (except that integers and rationals are converted to floats in expressions), two functions would be needed to halve a number (note that truncating division is used for integers):
integer (@) halve:
:> @ // 2
float (@) halve:
:> @ / 2.0
halve (@):
? is @ float
:> float @ halve
:> integer (@) halve
This silly example is used to introduce a language feature, the class function call, using the double colon:
x =: 3.0 print x::halve y =: 3 print y::halve
The double colon is a function call operator, that dynamically determines the class of the variable (in this case its kind), and selects the function for which the tail after the first parameter (which must be a scroll) matches like any other function template match.
Although the details are not very complex, they should not be elaborated here.
For convenience, instead of the scroll in parenthesis (@),
a double colon may be used in the function template:
integer :: halve:
:> @ // 2
Strings can be delimited either by quotes (") or primes (').
The former ones can be automatically converted by a user-given map
to a different, in particular localised, variant.
Typed programming languages declare each variable to conform to certain rules, be it integers, floating point number, strings or whatever. They could be regarded as predefined constraints, in that integer types restrict the value to certain ranges of numbers.
TUP-PL is designed to support such constraints flexibly, declared within the language. A number can be constrained to a range, and a variable to contain only numbers or a mix of numbers and strings.
This constraints system is discretionary by design and may be used to automatically generate assertions and optimise code.
Whether this concept is practical, is still open; the current compiler just ignores constraints.