Search                        Top                                  Index
HELP FOR_FORM            Ian Rogers, Adrian Howard, David Young Jan 1994
                                           Revised: John Gibson Jan 1996

For-forms are used to extend the syntax of "for ... endfor" to give more
varied forms of looping constructs. The user can add new forms.

         CONTENTS - (Use <ENTER> g to access required sections)

  1   Overview

  2   Predefined Extended for Forms
      2.1   in_vectorclass and in_vector
      2.2   in_string
      2.3   in_dstring
      2.4   in_subscripted
      2.5   with_index
      2.6   in_list
      2.7   on_list
      2.8   in_property
      2.9   on_property
      2.10  from_repeater
      2.11  in_region
      2.12  in_array
      2.13  equalto and allequalto

  3   How to Write Extensions

  4   Predefined General Utilities
      4.1   Using class_subscr_loop to Define New for-forms
      4.2   Using subscr_loop
      4.3   Adding with_index Hooks

  5   fast_for Details

1  Overview

This provides extensions to the normal looping constructs of the forms
    for ... in  ... do ... endfor
    for ... on  ... endfor
    fast_for ... in ... do ... endfast_for
    fast_for ... on ... do ... endfast_for


The extended for forms are

    for varlist type expression-list do ... endfor;

    fast_for varlist type expression-list do ... endfast_for;

where type is a syntax word to indicate an extended for iteration

2  Predefined Extended for Forms

Several frequently required "for ...type" forms are provided.
Additional forms can be defined by the user, as described below.

Descriptions of the built in forms follow.

2.1  in_vectorclass and in_vector

    for var1, var2 ... in_vectorclass vector1, vector2, ... do

Each of the variables varN iterates over elements of the corresponding
vector vectorN. In the case where there is only one variable and one
vector, this is a more efficient replacement for the tortuous

    for item in vector using_subscriptor subscrv do

using_subscriptor does does not allow iteration through multiple vectors


    vars a,b;
    for a, b in_vectorclass  'ABC', {ant bat cow}  do [^a ^b] => endfor;
    ** [65 ant]
    ** [66 bat]
    ** [67 cow]

Notice that in this example there are two vectors of different types: a
character vector and a full vector. Any number of vector expressions
with any mixture of vector types is permitted (unlike the in_vector
construct, defined below). The only constraint on this form is that
there must be a class_fast_subscr present in the system for each type of
vector used. If something other than a vector type is found, e.g. a
list, then an error will result.

The number of iterations to be done is determined by the size of the
smallest vector.

    for a, b in_vectorclass {1 2 3 4 5}, {dog bat ant} do
        [^a ^b]=>
    ** [1 dog]
    ** [2 bat]
    ** [3 ant]

fast_for can also be used. In that case it is assumed that the length of
the first vector will determine the number of iterations. This can lead
to bizarre results if it is longer than one of the other vectors, e.g.

    fast_for a, b in_vectorclass {1 2 3 4 5}, {dog bat ant} do
        [^a ^b]=>
    ** [1 dog]
    ** [2 bat]
    ** [3 ant]
    ** [4 <SYSTEM_OBJECT 00000016>]
    ** [5 <key string>]

The in_vector for-form is designed for the special case where all the
vectors are of the same type, namely standard Pop-11 full vectors.

    vars x,y;
    for x,y in_vector {1 2 3 4}, {a b c} do [^x ^y] => endfor;
    ** [1 a]
    ** [2 b]
    ** [3 c]

If one of the expressions after in_vector does not evaluate to a
vector then a run-time error will result. Otherwise this case is
slightly more efficient (ie. faster) than in_vectorclass.

2.2  in_string

This is designed for the special case where all the vectors are of the
same type, namely standard Pop-11 (d)strings.

    vars x,y;
    for x,y in_string '1234', 'abc' do [^x ^y] => endfor;
    ** [49 97]
    ** [50 98]
    ** [51 99]

If one of the expressions after in_string does not evaluate to a
(d)string then a run-time error will result. Otherwise this case is
slightly more efficient (ie. faster) than in_vectorclass.

2.3  in_dstring

This is the same as in_string when applied to ordinary strings. When
applied to a dstring the full character, include attributes is returned.

For example:

    vars x,y;
    for x,y in_dstring '1234', '\{b}abc' do [^x ^y] => endfor;
    ** [49 1048673]
    ** [50 1048674]
    ** [51 1048675]

See REF * STRINGS for more information on dstrings.

2.4  in_subscripted

    for var1, var2 ... in_subscripted struct1, struct2, ... do

For structures that have a class_apply that handles numerical
subscripts, and for which the procedure length produces an appropriate
result, it is possible to to use the form in_subscripted. This enables
simultaneous iteration over vectors and lists, as in:

    vars a, b, c;
    for a,b,c in_subscripted {1 2 3 4}, 'abcd', [ant bat cow ] do
        [^a ^b ^c] =>
    ** [1 97 ant]
    ** [2 98 bat]
    ** [3 99 cow]

2.5  with_index

For any of the above "vector-type" iteration constructs, it is sometimes
useful to keep a note of the "current index". Ie. the number of the
current slot (its cardinal number). This can be achieved automatically
by using with_index in conjunction with any of the above "vector-type"


    vars a, b, v = {x y z}
    for a with_index b in_vector v do
        [^a ^b] =>
    ** [x 1]
    ** [y 2]
    ** [z 3]

Assigning to the index variable will alter the behaviour of the loop.

See below for details of adding with_index hooks to user defined
"vector-type" iteration constructs.

2.6  in_list

    for var1, var2, ... in_list list1, list2, ... do

This construct is exactly equivalent to

    for var1, var2, ... in list1, list2, ... do

which can be thought of as syntactic sugar for the former. Also,
using_subscriptor cannot be used with this form

    vars x, y;

    for x, y in_list [a b c], [A B C] do [^x ^y] => endfor;
    ** [a A]
    ** [b B]
    ** [c C]

The length of the shortest list determines the number of iterations.

fast_for can also be used to increase speed with reduced checking.
This can lead to unpredictable results if one of this list has a
non-list tail. fast_for cannot be used with dynamic lists.

2.7  on_list

    for var1, var2, ... on_list list1, list2, ... do

This construct is equivalent to

    for var1, var2, ... on list1, list2, ... do


    vars x, y;
    for x, y on_list [a b c d], [A B C] do [^x ^y] => endfor;
    ** [[a b c d] [A B C]]
    ** [[b c d] [B C]]
    ** [[c d] [C]]

Again, the length of the shortest list determines the number of

fast_for can also be used, with the same caveats as for in_list.

2.8  in_property

    for key, val in_property property do

This form is used for iterating over key/value pairs in a property (also


    vars key, value;
    vars foo = newproperty([[a 1] [b 2] [c 3]], 3, false, true);
    for key,value in_property foo do
        [key = ^key , value = ^value] =>
    ** [key = c , value = 3]
    ** [key = a , value = 1]
    ** [key = b , value = 2]

The form:

    for key, val in_property property do

does approximately the same as

        procedure(key, val); ....key....val....endprocedure

However, the "for ... in_property.." version is more efficient as there
is no need to call a procedure on each iteration. quitloop and return
behave as you would expect for a looping construct.

The form

    fast_for key, val in_property property do

does approximately the same as the following

        procedure (key, val); ....key....val....endprocedure

and has similar dangers and restrictions (i.e. the loop body should not
access or update the contents of the property in case a garbage
collection has occurred and they had to be re-located).

See REF * FAST_APPPROPERTY for more information.

In both cases, the order in which the property entries are bound to the
two iteration variables is arbitrary.

Therefore, only one property at a time is allowed with this form as it
is meaningless to consider stepping through two properties in synchrony.

2.9  on_property

    fast_for entry on_property property do

This is similar to in_property except that, instead of the key and value
being returned, each time round the loop a full property entry is put
into entry. This entry is the same as that returned by
fast_get_prop_entry (REF * FASTPROCS/fast_get_prop_entry)


Say that nums is a property matching keys against numbers and that you
want to increment each of these numbers as long as the predicate pred
returns true when applied to the key. The following loop will do that:

    vars e;
    fast_for e on_property nums do
        if pred(fast_prop_entry_arg(e)) then
            fast_prop_entry_value(e) + 1
                -> fast_prop_entry_value(e);

The behaviour of this construct is very rude, and therefore there is no
slow version. The value of entry is a pointer to the entry cell which
is still in the property. Therefore, if the property itself is updated
within the loop, the resulting behaviour of the loop will be undefined.

2.10  from_repeater

    for var from_repeater repeater do

This construct can be used with the repeaters returned from, eg., discin
or incharitem. Each time round the loop var is given the next item
returned from the repeater. The loop exits, without executing the body
of the loop, when the repeater returns termin.

2.11  in_region

    for var in_region list do

This can be used in place of nested numerical for loops, even if the
depth of nesting is not known at compile-time. See HELP * in_region.

2.12  in_array

    for var1, var2, ... in_array array1, array2, ... do

This is used for iterating over data in arrays, much as in_vector can be
used for iterating over data in vectors. The full form is

    for var1, var2, ...
        with_index ivar
        in_array array1, array2, ...
        updating_last n
        in_region list
        of_dimension d

For details, see HELP * in_array.

2.13  equalto and allequalto

    for var equalto pattern [in list] do
    for var allequalto pattern_list [in list] do

The equalto construct iterates over all the elements of list which are
equal to the item pattern (i.e. element = pattern). See HELP * equalto.

The allequalto construct iterates over all possible lists of elements
from list which are equal to the list pattern_list (i.e. element_list =
pattern_list). See HELP * allequalto.

With both these constructs, list defaults to the value of the variable
database if the 'in list' clause is omitted.

3  How to Write Extensions

New extensions can be written using the for_extension define form.


    define :for_extension grum(varlist, isfast);

In this example the extension grum has been declared. Ie. the

    for var1, var2, ... varN grum ....

    fast_for var1, var2, ... varN grum ....

Is now valid. The need for the variables varlist and isfast is described
below, though these are ordinary variables and can be given any name.

Reading in the rest of the loop construct and deciding what Poplog VM
code to plant is then handled by the code that is in the body of the
for_extension define form. Autoloading is supported which means that
these extensions need not take up space until required.

Two arguments are placed on the stack before this code is called:

    varlist a list of the variables following for or fast_for

    isfast a boolean which is true if fast_for was used, false

If isfast is true, code that runs quickly without doing normal
checking may be planted. It is up to the user to ensure that such uses
involve no serious risks.

(See HELP * SLOWPROCS, REF * FASTPROCS for more information on checking
and non-checking procedures)

For example, at compile time the construct

    fast_for item1, item2 grum | struct1, struct2 do

invokes the code with the list

    [item1 item2]

and the boolean value


on the stack, with proglist set to the marked position.

It is up to the user to decide whether to make the fast_ version (ie.
when isfast is true) do anything different.

4  Predefined General Utilities

Several keywords are catered for without users having to define them,
these were described in section 2.

Additional predefined facilities are described below. In particular
there is a general subscriptor mechanism from which in_vectorclass and
in_subscripted etc. are derived.

4.1  Using class_subscr_loop to Define New for-forms

The forms in_vectorclass and in_subscripted described above are
defined as closures of the more general procedure:

    class_subscr_loop(varlist, isfast, index, subscrpdr);

The first two arguments are as before.

    index is either a word or false. If it is a word then this is
    taken as the name of the variable to be used as the index
    counter. This can make the index counter visible to user
    programs. If it is false then the compiler will use a hidden
    variable as the counter.

    subscrpdr is the procedure for getting the class subscripting
    procedure out of the datakey of the vectorclass in question.

Some form of closure of this procedure is all that is needed to form an
efficient vector-type iteration construct.

The form in_vectorclass can be defined thus:

    uses class_subscr_loop;

    define :for_extension in_vectorclass with_nargs 2;
        class_subscr_loop(false, class_fast_subscr);

and in_subscripted can be defined:

    define :for_extension in_subscripted with_nargs 2;
        class_subscr_loop(false, class_apply);

These two are probably the only instantiations that make sense for

If fast_for is used (isfast is therefore true,) then only the length of
the first vector is considered when finding out how many times to
iterate, as illustrated previously.

4.2  Using subscr_loop

The in_vector form, more efficient than in_vectorclass, is defined using
a closure of the more general procedure:

    subscr_loop(varlist, isfast, index, subscrpdr, checkpdr)

The first four arguments are as before. checkpdr is a checking procedure
which must mishap if the data to be subscripted is of the wrong class.

A consequence of this is that the subscripting procedure subscrpdr can
be of the fast variety.

A closure of this procedure is all that is needed to form a vector-type
iteration construct which is runs slightly faster than the
in_vectorclass form.

So in_vector could be defined as:

    uses subscr_loop;

    define lconstant check_vector(item);
        lvars item;
        unless isvector(item) then
            mishap(item, 1, 'VECTOR NEEDED')

    define :for_extension in_vector with_nargs 2;
        subscr_loop(false, fast_subscrv, check_vector);

When isfast is true, the checking procedure does not get run, and
there are the same consequences that applied to class_subscr_loop.

4.3  Adding with_index Hooks

The define form with_index_hook can be used to define hooks for the
with_index style of vector type iteration.


    define :with_index_hook in_vector with_nargs 3;
        subscr_loop(fast_subscrv, check_vector);

is used to define the construct:
    for vars with_index ivar in_vector | exprs do
        ...                            -

Simply the body of the define-form must define a procedure of three
arguments: varlist, isfast and index.

The functions of these variable are as described above in the
description of class_subscr_loop. This procedure will be called with
varlist set to the list of loop variables vars, isfast will be set to
true or false depending on whether fast_for or for respectively was used
to open the construct, and index is set to the index variable ivar.
proglist will be set to the marked position.

5  fast_for Details

Using fast_for often gains speed just by switching off checking.

Also, for vector-type iteration, the non-fast forms compare the lengths
of all the vectors and uses the shortest one to determine the number of
iterations required. This comparison is avoided if fast_for is used.
Instead only the length of the first vector is examined.

The rules that apply to fast_appproperty apply to the fast version of

--- C.all/help/for_form
--- Copyright University of Sussex 1996. All rights reserved.