Pclass
Pclass
com
class — Class programming
Description
Stata’s two programming languages, ado and Mata, each support object-oriented programming. This
manual entry explains object-oriented programming in ado. Most users interested in object-oriented
programming will wish to do the programming in Mata. See [M-2] class to learn about object-oriented
programming in Mata.
Ado classes are a programming feature of Stata that are especially useful for dealing with graphics
and GUI problems, although their use need not be restricted to those topics. Ado class programming
is an advanced programming topic and will not be useful to most programmers.
1
2 class — Class programming
1. Introduction
A class is a collection of member variables and member programs. The member programs of a
class manipulate or make calculations based on the member variables. Classes are defined in .class
files. For instance, we might define the class coordinate in the file coordinate.class:
begin coordinate.class
version 18.0
class coordinate {
double x
double y
}
program .set
args x y
.x = ‘x’
.y = ‘y’
end
end coordinate.class
The above file does not create anything. It merely defines the concept of a “coordinate”. Now that
the file exists, however, you could create a “scalar” variable of type coordinate by typing
.coord = .coordinate.new
.coord is called an instance of coordinate; it contains .coord.x (a particular x coordinate)
and .coord.y (a particular y coordinate). Because we did not specify otherwise, .coord.x and
.coord.y contain missing values, but we could reset .coord to contain (1,2) by typing
.coord.x = 1
.coord.y = 2
Here we can do that more conveniently by typing
.coord.set 1 2
because coordinate.class provides a member program called .set that allows us to set the
member variables. There is nothing especially useful about .set; we wrote it mainly to emphasize
that classes could, in fact, contain member programs. Our coordinate.class definition would be
nearly as good if we deleted the .set program. Classes are not required to have member programs,
but they may.
If we typed
.coord2 = .coordinate.new
.coord2.set 2 4
we would now have a second instance of a coordinate, this one named .coord2, which would
contain (2,4).
class — Class programming 3
What are the values of these variables? Because we did not specify otherwise, .li.c0 and .li.c1
will receive default values for their type, coordinate. That default is (.,.) because we did not specify
otherwise when we defined lines or coordinates. Therefore, the default values are (.,.) and (.,.),
and we have a missing line.
As with coordinate, we included the member function .set to make setting the line easier. We
can type
.li.set 1 2 2 4
and we will have a line going from (1,2) to (2,4).
4 class — Class programming
.set, .length, and .midpoint came from line.class. .c0.set and .c1.set came from
coordinate.class.
Member program .length returns the length of the line.
.len = .li.length
would create .len containing the result of .li.length. The result of running the program .length
on the object .li. .length returns a double, and therefore, .len will be a double.
.midpoint returns the midpoint of a line.
.mid = .li.midpoint
would create .mid containing the result of .li.midpoint, the result of running the program
.midpoint on the object .li. .midpoint returns a coordinate, and therefore, .mid will be a
coordinate.
2. Definitions
After that, .name is variously called an identifier, class variable, class instance, object, object
instance, or sometimes just an instance. Call it what you will, the above creates new .name—or
replaces existing .name—to contain the result of an application of the definition of classname. And,
just as with any variable, you can have many different variables with many different names all the
same type.
.name is called a first-level or top-level identifier. .name1.name2 is called a second-level identifier,
and so on. Assignment into top-level identifiers is allowed if the identifier does not already exist or
if the identifier exists and is of type classname. If the top-level identifier already exists and is of a
different type, you must drop the identifier first and then re-create it; see 11. Object destruction.
Consider the assignment
.name1.name2 = .classname.new
The above statement is allowed if .name1 already exists and if .name2 is declared, in .name1’s class
definition, to be of type classname. In that case, .name1.name2 previously contained a classname
instance and now contains a classname instance, the difference being that the old contents were
discarded and replaced with the new ones. The same rule applies to third-level and higher identifiers.
Classes, and class instances, may also contain member programs. Member programs are identified
in the same way as class variables. .name1.name2 might refer to a member variable or to a member
program.
Assume that we type “.mycoord.set 2 4”. When .set executes, it executes in the context of
.mycoord. In the program, the references to .x and .y are assumed to be to .mycoord.x and
.mycoord.y. If we typed “.other.set”, the references would be to .other.x and .other.y.
Look at the statement “.x = ‘x’” in .set. Pretend that ‘x’ is 2 so that, after macro substitution,
the statement reads “.x = 2”. Is this a statement that the first-level identifier .x is to be set to 2?
No, it is a statement that .impliedcontext.x is to be set to 2. The same would be true whether .x
appeared to the right of the equal sign or anywhere else in the program.
The rules for resolving things like .x and .y are actually more complicated. They are resolved to
the implied context if they exist in the implied context, and otherwise they are interpreted to be in
the global context. Hence, in the above examples, .x and .y were interpreted as being references to
.impliedcontext.x and .impliedcontext.y because .x and .y existed in .impliedcontext. If, however,
our program made a reference to .c, that would be assumed to be in the global context (that is, to
be just .c), because there is no .c in the implied context. This is discussed at length in 9. Prefix
operators.
If a member program calls a regular program—a regular ado-file—that program will also run in
the same class context; for example, if .set included the lines
6 class — Class programming
move_to_right
.x = r(x)
.y = r(y)
and program move to right.ado had lines in it referring to .x and .y, they would be interpreted
as .impliedcontext.x and .impliedcontext.y.
In all programs—member programs or ado-files—we can explicitly control whether we want
identifiers in the implied context or globally with the .Local and .Global prefixes; see 9. Prefix
operators.
3. Version control
The first thing that should appear in a .class file is a version statement; see [P] version. For
example, coordinate.class reads
begin coordinate.class
version 18.0
[ class statement defining member variables omitted ]
program .set
args x y
.x = ‘x’
.y = ‘y’
end
end coordinate.class
The version 18.0 at the top of the file specifies not only that, when the class definition is read, it
be interpreted according to version 18.0 syntax, but also that when each of the member programs runs,
it be interpreted according to version 18.0. Thus you do not need to include a version statement
inside the definition of each member program, although you may if you want that one program to
run according to the syntax of a different version of Stata.
Including the version statement at the top, however, is of vital importance. Stata is under continual
development, and so is the class subsystem. Syntax and features can change. Including the version
command ensures that your class will continue to work as you intended.
4. Member variables
4.1 Types
The second thing that appears in a .class file is the definition of the member variables. We have
seen two examples:
begin coordinate.class
version 18.0
class coordinate {
double x
double y
}
[ member programs omitted ]
end coordinate.class
and
class — Class programming 7
begin line.class
version 18.0
class line {
coordinate c0
coordinate c1
}
[ member programs omitted ]
end line.class
In the first example, the member variables are .x and .y, and in the second, .c0 and .c1. In the
first example, the member variables are of type double, and in the second, of type coordinate,
another class.
The member variables may be of type
You may also specify a string expression, but you must enclose it in parentheses. For
example,
string name = ("no" + "body")
string b = (char(11))
Or you may specify the identifier of a member variable to be copied or a member program
to be run, as long as the member variable is a string or the program returns a string.
If a member program is specified that requires arguments, they must be specified following
the identifier. Examples include
string n = .defaultname
string a = .recapitalize "john smith"
string b = .names.defaults, category(null)
The identifiers are interpreted in terms of the global context, not the class context being
defined. Thus .defaultname, .recapitalize, and .names.defaults must exist in the
global context.
array membervarname = {. . . }
After the equal sign, you type the set of elements in braces ({ and }), with each element
separated from the next by a comma.
If an element is enclosed in quotes (simple or compound), the corresponding array element
is defined to be string with the contents specified.
If an element is a literal number excluding ., .a, . . . , and .z, the corresponding array
element is defined to be double and filled in with the number specified.
If an element is enclosed in parentheses, what appears inside the parentheses is evaluated
as an expression. If the expression evaluates to a string, the corresponding array element is
defined to be string and the result is filled in. If the expression evaluates to a number,
the corresponding array element is defined to be double and the result is filled in. Missing
values may be assigned to array elements by being enclosed in parentheses.
An element that begins with a period is interpreted as an object identifier in the global
context. That object may be a member variable or a member program. The corresponding
array element is defined to be of the same type as the specified member variable or of the
same type as the member program returns. If a member program is specified that requires
arguments, the arguments must be specified following the identifier, but the entire syntactical
elements must be enclosed in square brackets ([ and ]).
If the element is nothing, the corresponding array element is left undefined.
Examples include
array mixed = {1, 2, "three", 4}
array els = {.box.new, , .table.new}
array rad = {[.box.new 2 3], , .table.new}
Note the double commas in the last two initializations. The second element is left undefined.
Some programmers would code
array els = {.box.new, /*nothing*/, .table.new}
array rad = {[.box.new 2 3], /*nothing*/, .table.new}
to emphasize the null initialization.
10 class — Class programming
classname membervarname = . . .
After the equal sign, you specify the identifier of a member variable to be copied or a
member program to be run, as long as the member variable is of type classname or the
member program returns something of type classname. If a member program is specified
that requires arguments, they must be specified following the identifier. In either case, the
identifier will be interpreted in the global context. Examples include
box mybox1 = .box.new
box mybox2 = .box.new 2 4 7 8, tilted
All the types can be initialized by copying other member variables or by running other member
programs. These other member variables and member programs must be defined in the global context
and not the class context. In such cases, each initialization value or program is, in fact, copied or
run only once—at the time the class definition is read—and the values are recorded for future use.
This makes initialization fast. This also means, however, that
• If, in a class definition called, say, border.class, you defined a member variable that was
initialized by .box.new, and if .box.new counted how many times it is run, then even if
you were to create 1,000 instances of border, you would discover that .box.new was run
only once. If .box.new changed what it returned over time (perhaps because of a change
in some state of the system being implemented), the initial values would not change when
a new border object was created.
• If, in border.class, you were to define a member variable that is initialized as .sys-
tem.curvals.no of widgets, which we will assume is another member variable, then even
if .system.curvals.no of widgets were changed, the new instances of border.class
would always have the same value—the value of .system.curvals.no of widgets
current at the time border.class was read.
In both of the above examples, the method just described—the prerecorded assignment method of
specifying initial values—would be inadequate. The method just described is suitable for specifying
constant initial values only.
begin coordinate.class
version 18.0
class coordinate {
double x
double y
}
program .new
if "‘0’" != "" {
.set ‘0’
}
end
program .set
args x y
.x = ‘x’
.y = ‘y’
end
end coordinate.class
With this addition, we could type
.coord = .coordinate.new
.coord.set 2 4
or we could type
.coord = .coordinate.new 2 4
We have arranged .new to take arguments—optional ones here—that specify where the new point
is to be located. We wrote the code so that .new calls .set, although we could just as well have
written the code so that the lines in .set appeared in .new and then deleted the .set program. In
fact, the two-part construction can be desirable because then we have a function that will reset the
contents of an existing class as well.
In any case, by defining your own .new, you can arrange for any sort of complicated initialization
of the class, and that initialization can be a function of arguments specified if that is necessary.
The .new program need not return anything; see 6. Member programs’ return values.
.new programs are not restricted just to filling in initial values. They are programs that you
can code however you wish. .new is run every time a new instance of a class is created with one
exception: when an instance is created as a member of another instance (in which case, the results
are prerecorded).
4.6 Scope
In the examples we have seen so far, the member variables are unique to the instance. For example,
if we have
.coord1 = .coordinate.new
.coord2 = .coordinate.new
then the member variables of .coord1 have nothing to do with the member variables of .coord2.
If we were to change .coord1.x, then .coord2.x would remain unchanged.
Classes can also have variables that are shared across all instances of the class. Consider
begin coordinate2.class
version 18.0
class coordinate2 {
classwide:
double x_origin = 0
double y_origin = 0
instancespecific:
double x = 0
double y = 0
}
end coordinate2.class
In this class definition, .x and .y are as they were in coordinate.class—they are unique to
the instance. .x origin and .y origin, however, are shared across all instances of the class. That
is, if we were to type
.ac = .coordinate2.new
.bc = .coordinate2.new
there would be only one copy of .x origin and of .y origin. If we changed .x origin in .ac,
.ac.x_origin = 2
we would find that .bc.x origin had similarly been changed. That is because .ac.x origin and
.bc.x origin are, in fact, the same variable.
The effects of initialization are a little different for classwide variables. In coordinate2.class,
we specified that .origin x and .origin y both be initialized as 0, and so they were when we typed
“.ac = .coordinate2.new”, creating the first instance of the class. After that, however, .origin x
and .origin y will never be reinitialized because they need not be re-created, being shared. (That
is not exactly accurate because, once the last instance of a coordinate2 has been destroyed, the
variables will need to be reinitialized the next time a new first instance of coordinate2 is created.)
Classwide variables, just as with instance-specific variables, can be of any type. We can define
begin supercoordinate.class
version 18.0
class supercoordinate {
classwide:
coordinate origin
instancespecific:
coordinate pt
}
end supercoordinate.class
The qualifiers classwide: and instancespecific: are used to designate the scope of the
member variables that follow. When neither is specified, instancespecific: is assumed.
class — Class programming 13
Technical note
Just as with the declaration of member variables inside the class {} statement, you can omit
specifying the type when you specify the initialization. In the above, the following would also be
allowed:
.coord.Declare mycolor = .color.new, color(default)
14 class — Class programming
time a new instance is created, unique x- and y -coordinate variables are created and filled in with
missing. This work is done by .nextnames.
The .set looks similar to the one from .varcoordinates except that now we are holding
variable names in ‘.x’ and ‘.y’, and we use replace to store the values from the specified variables
into our coordinate variables.
The .oncopy member function creates unique names to hold the variables, using .nextnames,
and then copies the contents of the coordinate variables from the source object, using .set.
Now, when we type
.coordcopy = .coord
the x- and y -coordinate variables in .coordcopy will be different variables from those in .coord
with copies of their values.
The varcoordinate class does not yet do anything interesting, and other than the example in
the following section, we will not develop it further.
5. Inheritance
One class definition can inherit from other class definitions. This is done by including the
inherit(classnamelist) option:
begin newclassname.class
version 18.0
class newclassname {
...
}, inherit(classnamelist)
program . . .
...
end
...
end newclassname.class
16 class — Class programming
newclassname inherits the member variables and member programs from classnamelist. In general,
classnamelist contains one class name. When classnamelist contains more than one class name, that
is called multiple inheritance.
To be precise, newclassname inherits all the member variables from the classes specified ex-
cept those that are explicitly defined in newclassname, in which case the definition provided in
newclassname.class takes precedence. It is considered bad style to name member variables that
conflict.
For multiple inheritance, it is possible that, although a member variable is not defined in newclass-
name, it is defined in more than one of the “parents” (classnamelist). Then it will be the definition
in the rightmost parent that is operative. This too is to be avoided, because it almost always results
in programs’ breaking.
newclassname also inherits all the member programs from the classes specified. Here name conflicts
are not considered bad style, and in fact, redefinition of member programs is one of the primary
reasons to use inheritance.
newclassname inherits all the programs from classnamelist—even those with names in common—
and a way is provided to specify which of the programs you wish to run. For single inheritance,
if member program .zifl is defined in both classes, then .zifl is taken as the instruction to run
.zifl as defined in newclassname, and .Super.zifl is taken as the instruction to run .zifl as
defined in the parent.
For multiple inheritance, .zifl is taken as the instruction to run .zifl as defined in newclassname,
and .Super(classname).zifl is taken as the instruction to run .zifl as defined in the parent
classname.
A good reason to use inheritance is to “steal” a class and to modify it to suit your purposes.
Pretend that you have alreadyexists.class and from that you want to make alternative.class,
something that is much like alreadyexists.class—so much like it that it could be used wherever
alreadyexists.class is used—but it does one thing a little differently. Perhaps you are writing a
graphics system, and alreadyexists.class defines everything about the little circles used to mark
points on a graph, and now you want to create alternate.class that does the same, but this time
for solid circles. Hence, there is only one member program of alreadyexists.class that you want
to change: how to draw the symbol.
In any case, we will assume that alternative.class is to be identical to alreadyexists.class,
except that it has changed or improved member function .zifl. In such a circumstance, it would
not be uncommon to create
begin alternative.class
version 18.0
class alternative {
}, inherit(alreadyexists)
program .zifl
...
end
end alternative.class
Moreover, in writing .zifl, you might well call .Super.zifl so that the old .zifl performed its
tasks, and all you had to do was code what was extra (filling in the circles, say). In the example
above, we added no member variables to the class.
class — Class programming 17
Perhaps the new .zifl needs a new member variable—a double—and let’s call it .sizeofresult.
Then we might code
begin alternative.class
version 18.0
class alternative {
double sizeofresult
}, inherit(alreadyexists)
program .zifl
...
end
end alternative.class
Now let’s consider initialization of the new variable, .sizeofresult. Perhaps having it initialized
as missing is adequate. Then our code above is adequate. Suppose that we want to initialize it to 5.
Then we could include an initializer statement. Perhaps we need something more complicated that
must be handled in a .new. In this final case, we must call the inherited classes’ .new programs by
using the .Super modifier:
begin alternative.class
version 18.0
class alternative {
double sizeofresult
}, inherit(alreadyexists)
program .new
...
.Super.new
...
end
program .zifl
...
end
end alternative.class
Member programs may optionally return “values”, and those can be doubles, strings, arrays,
or class instances. These return values can be used in assignment, and thus you can code
.len = .li.length
.coord3 = .li.midpoint
Just because a member program returns something, it does not mean it has to be consumed. The
programs .li.length and .li.midpoint can still be executed directly,
.li.length
.li.midpoint
and then the return value is ignored. (.midpoint and .length are member programs that we included
in line.class. .length returns a double, and .midpoint returns a coordinate.)
You cause member programs to return values by using the class exit command; see [P] class
exit.
Do not confuse returned values with return codes, which all Stata programs set, even member
programs. Member programs exit when they execute.
18 class — Class programming
Any of the preceding are valid ways of exiting a member program, although the last is perhaps
best avoided. class exit without arguments has the same effect as exit without arguments; it does
not matter which you code.
If a member program returns nothing, the result is as if it returned string containing "" (nothing).
Member programs may also return values in r(), e(), and s(), just like regular programs. Using
class exit to return a class result does not prevent member programs from also being r-class,
e-class, or s-class.
7. Assignment
Consider .coord defined
.coord = .coordinate.new
That is an example of assignment. A new instance of class coordinate is created and assigned
to .coord. In the same way,
.coord2 = .coord
is another example of assignment. A copy of .coord is made and assigned to .coord2.
Assignment is not allowed just with top-level names. The following are also valid examples of
assignment:
.coord.x = 2
.li.c0 = .coord
.li.c0.x = 2+2
.todo.name = "Jane Smith"
.todo.n = 2
.todo.list[1] = "Turn in report"
.todo.list[2] = .li.c0
In each case, what appears on the right is evaluated, and a copy is put into the specified place.
Assignment based on the returned value of a program is also allowed, so the following are also valid:
.coord.x = .li.length
.li.c0 = .li.midpoint
.length and .midpoint are member programs of line.class, and .li is an instance of line. In
the first example, .li.length returns a double, and that double is assigned to .coord.x. In the
second example, .li.midpoint returns a coordinate, and that coordinate is assigned to li.c0.
Also allowed would be
.todo.list[3] = .color.cvalue, color(green)
.todo.list = {"Turn in report", .li.c0, [.color.cvalue, color(green)]}
class — Class programming 19
In both examples, the result of running .color.cvalue, color(green) is assigned to the third
array element of .todo.list.
The first line is valid because .newthing did not previously exist. After the first assignment, however,
.newthing did exist and was of type double. That caused the second assignment to be invalid, the
error being “type mismatch”; r(109).
The following are also invalid:
.coord.x = .li.midpoint
.li.c0 = .li.length
They are invalid because .li.midpoint returns a coordinate, and .coord.x is a double, and
because .li.length returns a double, and .li.c0 is a coordinate.
and
.todo.list = {"Turn in report", .li.c0, [.color.cvalue, color(green)]}
do not have the same effect. The first set of statements reassigns elements 1, 2, and 3 and leaves any
other defined elements unchanged. The second statement replaces the entire array with an array that
has only elements 1, 2, and 3 defined.
After an element has been assigned, it may be unassigned (cleared) using .Arrdropel. For
example, to unassign .todo.list[1], you would type
.todo.list[1].Arrdropel
Clearing an element does not affect the other elements of the array. In the above example,
.todo.list[2] and .todo.list[3] continue to exist.
New and existing elements may be assigned and reassigned freely, except that if an array element
already exists, it may be reassigned only to something of the same type.
.todo.list[2] = .coordinate[2]
would not be allowed because .todo.list[2] is a coordinate and "Clear the coordinate"
is a string. If you wish to reassign an array element to a different type, you first drop the existing
array element and then assign it.
.todo.list[2].Arrdropel
.todo.list[2] = "Clear the coordinate"
#
If the rvalue is a number excluding missing values ., .a, . . . , and .z, a double equal to
the number specified will be returned.
exp and (exp)
If the rvalue is an expression, the expression will be evaluated and the result returned. A
double will be returned if the expression returns a numeric result and a string will be
returned if expression returns a string. Expressions returning matrices are not allowed.
The expression need not be enclosed in parentheses if the expression does not begin with
simple or compound double quotes and does not begin with a period followed by nothing
or a letter. In the cases just mentioned, the expression must be enclosed in parentheses. All
expressions may be enclosed in parentheses.
An implication of the above is that missing value literals must be enclosed in parentheses:
lvalue = (.).
.id .id . . . program arguments
If the rvalue begins with a period,
it is interpreted as an object reference. The object is
evaluated and returned. .id .id . . . may refer to a member variable or a member program.
If .id .id . . . refers to a member variable, the value of the variable will be returned.
If .id .id . . . refers to a member program, the program will be executed and the result
returned. If the member program returns nothing, a string containing "" (nothing) will be
returned.
If .id .id . . . refers to a member program, arguments may be specified following the
program name.
{} and {el ,el ,. . . }
If the rvalue begins with an open brace, an array will be returned.
If the rvalue is {}, an empty array will be returned.
If the rvalue is {el ,el ,. . . }, an array containing the specified elements will be returned.
If an el is nothing, the corresponding array element will be left undefined.
If an el is " string " or ‘" string "’, the corresponding array element will be defined as a
string containing string.
If an el is # excluding missing values ., .a, . . . , .z, the corresponding array element will
be defined as a double containing the number specified.
If an el is (exp), the expression is evaluated, and the corresponding array element will
be defined as a double if the expression returns a numeric result or as a string if the
expression returns a string. Expressions returning matrices are not allowed.
If an el is .id .id . . . or [.id .id . . . program arguments ], the object is evaluated,
and the corresponding array element will be defined according to what was returned. If the
object is a member program and arguments need to be specified, the el must be enclosed in
square brackets.
Recursive array definitions are not allowed.
Finally, in 4.3 Specifying initialization—where we discussed member variable initialization—what
actually appears to the right of the equal sign is an rvalue, and everything just said applies. The
previous discussion was incomplete.
22 class — Class programming
8. Built-ins
.new and .ref are examples of built-in member programs that are included in every class. There
are other built-ins as well.
Built-ins may be used on any object except programs and other built-ins. Let .B refer to a built-in.
Then
• If .a.b.myprog refers to a program, .a.b.myprog.B is an error (and, in fact,
.a.b.myprog.anything is also an error).
• .a.b.B.anything is an error.
class — Class programming 23
Built-ins come in two forms: built-in functions and built-in modifiers. Built-in functions return
information about the class or class instance on which they operate but do not modify the class or
class instance. Built-in modifiers might return something—in general they do not—but they modify
(change) the class or class instance.
Except for .new (and that was covered in 4.4 Specifying initialization 2, .new), built-ins may not
be redefined.
.ref n
returns a double. Returned is the total number of identifiers sharing object. Returned is 1
if the object is unshared. See 7.4 Assignment of reference.
.arrnels
returns a double. .arrnels is for use with arrays; it returns the largest index of the array
that has been assigned data. If object is not an array, it returns an error.
.arrindexof "string"
returns a double. .arrindexof is for use with arrays; it searches the array for the
first element equal to string and returns the index of that element. If string is not found,
.arrindexof returns 0. If object is not an array, it returns an error.
.classmv
returns an array containing the .refs of each classwide member variable in object. See
12.3 Arrays of member variables.
.instancemv
returns an array containing the .refs of each instance-specific member variable in object.
See 12.3 Arrays of member variables.
.dynamicmv
returns an array containing the .refs of each dynamically allocated member variable in
object. See 12.3 Arrays of member variables.
.superclass
returns an array containing the .refs of each of the classes from which the specified
object inherited. See 12.3 Arrays of member variables.
9. Prefix operators
There are three prefix operators:
.Global
.Local
.Super
Prefix operators determine how object names such as .a, .a.b, .a.b.c, . . . are resolved.
Consider a program invoked by typing .alpha.myprog. In program .myprog, any lines such as
.a = .b
are interpreted according to the implied context, if that is possible. .a is interpreted to mean .alpha.a
if .a exists in .alpha; otherwise, it is taken to mean .a in the global context, meaning that it is
taken to mean just .a. Similarly, .b is taken to mean .alpha.b if .b exists in .alpha; otherwise,
it is taken to mean .b.
What if .myprog wants .a to be interpreted in the global context even if .a exists in .alpha?
Then the code would read
.Global.a = .b
If instead .myprog wanted .b to be interpreted in the global context (and .a to be interpreted in
the implied context), the code would read
.a = .Global.b
Obviously, if the program wanted both to be interpreted in the global context, the code would read
.Global.a = .Global.b
.Local is the reverse of .Global: it ensures that the object reference is interpreted in the implied
context. .Local is rarely specified because the local context is searched first, but if there is a
circumstance where you wish to be certain that the object is not found in the global context, you may
specify its reference preceded by .Local. Understand, however, that if the object is not found, an
error will result, so you would need to precede commands containing such references with capture;
see [P] capture.
In fact, if it is used at all, .Local is nearly always used in a macro-substitution context—something
discussed in the next section—where errors are suppressed and where nothing is substituted when
errors occur. Thus in advanced code, if you were trying to determine whether member variable
.addedvar exists in the local context, you could code
if "‘Local.addedvar.objtype’" == "" {
/* it does not exist */
}
else {
/* it does */
}
The .Super prefix is used only in front of program names and concerns inheritance when one
program occults another. This was discussed in 5. Inheritance.
26 class — Class programming
When a class object is quoted, its printable form is substituted. This is defined as
Any object may be quoted, including programs. If the program takes arguments, they are included
inside the quotes:
scalar len = ‘.coord.length’
local clr "‘.color.cvalue, color(green)’"
If the quoted reference results in an error, the error message is suppressed, and nothing is substituted.
Similarly, if a class instance is quoted—or a program returning a class instance is quoted—nothing
is substituted. That is, nothing is substituted, assuming that the member program .macroexpand has
not been defined for the class, as is usually the case. If .macroexpand has been defined, however, it
is executed, and what macroexpand returns—which may be a string or a double—is substituted.
For example, say that we wanted to make all objects of type coordinate substitute (#,#) when
they were quoted. In the class definition for coordinate, we could define .macroexpand,
class — Class programming 27
begin coordinate.class
version 18.0
class coordinate {
[ declaration of member variables omitted ]
}
[ definitions of class programs omitted ]
program .macroexpand
local tosub : display "(" ‘.x’ "," ‘.y’ ")"
class exit "‘tosub’"
end
end coordinate.class
and now coordinates will be substituted. Say that .mycoord is a coordinate currently set to
(2,3). If we did not include .macroexpand in the coordinate.class file, typing
. . . ‘.mycoord’. . .
would not be an error but would merely result in
......
Having defined .macroexpand, it will result in
. . . (2,3). . .
A .macroexpand member function is intended as a utility for returning the printable form of a class
instance and nothing more. In fact, the class system prevents unintended corruption of class-member
variables by making a copy, returning the printable form, and then destroying the copy. These steps
ensure that implicitly calling .macroexpand has no side effects on the class instance.
The program creates two new class instances of bubbles in the global context, both with temporary
names. We can be assured that .‘a’ and .‘b’ are global because the names ‘a’ and ‘b’ were
obtained from tempname and therefore cannot already exist in whatever context in which .tension
runs. Therefore, when the program ends, .‘a’ and .‘b’ will be automatically dropped. Even so,
.tension can return .‘a’. It can do that because, at the time class exit is executed, the program
has not yet concluded and .‘a’ still exists. You can even code
program .tension
...
tempname a b
.‘a’ = .bubble.new
.‘b’ = .bubble.new
...
class exit .‘a’.ref
end
and that also will return .a and, in fact, will be faster because no extra copy will be made. This form
is recommended when returning an object stored in a temporary name. Do not, however, add .refs
on the end of “real” (nontemporary) objects being returned because then you would be returning not
just the same values as in the real object but the object itself.
You can clear the entire class system by typing discard; see [P] discard. There is no classutil
drop all command: Stata’s graphics system also uses the class system, and dropping all the class
definitions and instances would cause graph difficulty. discard also clears all open graphs, so the
disappearance of class definitions and instances causes graph no difficulty.
During the development of class-based systems, you should type discard whenever you make a
change to any part of the system, no matter how minor or how certain you are that no instances of
the definition modified yet exist.
12.1 Keys
The .objkey built-in function returns a string called a key that can be used to reference the
object as an rvalue but not as an lvalue. This would typically be used in
local k = ‘.a.b.objkey’
or
.c.k = .a.b.objkey
where .c.k is a string. Thus the keys stored could be then used as follows:
It does not matter if the key is stored in a macro or a string member variable—it can be used
equally well—and you always use the key by macro quoting.
A key is a special string that stands for the object. Why not, you wonder, simply type .a.b rather
than .‘.c.k’ or .‘k’? The answer has to do with implied context.
class — Class programming 29
12.2 Unames
The built-in function .uname returns a name that can be used throughout Stata that uniquely
corresponds to the object. The mapping is one way. Unames can be obtained for objects, but the
original object’s name cannot be obtained from the uname.
Pretend that you have object .a.b.c, and you wish to obtain a name you can associate with that
object because you want to create a variable in the current dataset, or a value label, or whatever else,
to go along with the object. Later, you want to be able to reobtain that name from the object’s name.
.a.b.c.uname will provide that name. The name will be ugly, but it will be unique. The name is
not temporary: you must drop whatever you create with the name later.
Unames are, in fact, based on the object’s .ref. That is, consider two objects, .a.b.c and .d.e,
and pretend that they refer to the same data; that is, you have previously executed
.a.b.c.ref = .d.e.ref
or
.d.e.ref = .a.b.c.ref
Then .a.b.c.uname will equal .d.e.uname. The names returned are unique to the data being
recorded, not the identifiers used to arrive to the data.
As an example of use, within Stata’s graphics system sersets are used to hold the data behind a
graph; see [P] serset. An overall graph might consist of several graphs. In the object nesting for a
graph, each individual graph has its own object holding a serset for its use. The individual objects,
however, are shared when the same serset will work for two or more graphs, so that the same data
are not recorded again and again. That is accomplished by simply setting their .refs equal. Much
later in the graphics code, when that code is writing a graph out to disk for saving, it needs to figure
out which sersets need to be saved, and it does not wish to write shared sersets out multiple times.
Stata finds out what sersets are shared by looking at their unames and, in fact, uses the unames to
help it keep track of which sersets go with which graph.
30 class — Class programming
begin coordinate2.class
version 18.0
class coordinate2 {
classwide:
double x_origin = 0
double y_origin = 0
instancespecific:
double x = 0
double y = 0
}
end coordinate2.class
Then
referring to . . . is equivalent to referring to . . .
.my.c.classmv[1] .my.c.c.x origin
.my.c.classmv[2] .my.c.c.y origin
.my.c.instancemv[1] .my.c.c.x
.my.c.instancemv[2] .my.c.c.y
class — Class programming 31
If any member variables were added dynamically using .Dynamic, they could equally well be
accessed via .my.c.dynamicmv[] or their names. Either of the above could be used on the left or
right of an assignment.
If coordinate2.class inherited from another class (it does not), referring to .coor-
dinate2.superclass[1] would be equivalent to referring to the inherited class; .coordi-
nate2.superclass[1].new, for instance, would be allowed.
These “functions” are mainly of interest to those writing utilities to act on class instances as a
general structure.
Appendix B. Jargon
built-in: a member program that is automatically defined, such as .new. A built-in function is a
member program that returns a result without changing the object on which it was run. A built-in
modifier is a member program that changes the object on which it was run and might return a
result as well.
class: a name for which there is a class definition. If we say that coordinate is a class, then
coordinate.class is the name of the file that contains its definition.
class instance: a “variable”; a specific, named copy (instance) of a class with its member values filled
in; an identifier that is defined to be of type classname.
classwide variable: a member variable that is shared by all instances of a class. Its alternative is an
instance-specific variable.
inheritance: the ability to define a class in terms of one (single inheritance) or more (multiple
inheritance) existing classes. The existing class is typically called the base or super class, and by
default, the new class inherits all the member variables and member programs of the base class.
identifier: the name by which an object is identified, such as .mybox or .mybox.x.
implied context: the instance on which a member program is run. For example, in .a.b.myprog,
.a.b is the implied context, and any references to, say, .x within the program, are first assumed
to, in fact, be references to .a.b.x.
instance: a class instance.
instance-specific variable: a member variable that is unique to each instance of a class; each instance
has its own copy of the member variable. Its alternative is a classwide variable.
lvalue: an identifier that may appear to the left of the = assignment operator.
member program: a program that is a member of a class or of an instance.
member variable: a variable that is a member of a class or of an instance.
32 class — Class programming
object: a class or an instance; this is usually a synonym for an instance, but in formal syntax
definitions, if something is said to be allowed to be used with an object, that means it may be
used with a class or with an instance.
polymorphism: when a system allows the same program name to invoke different programs according
to the class of the object. For example, .draw might invoke one program when used on a star
object, .mystar.draw, and a different program when used on a box object, .mybox.draw.
reference: most often the word is used according to its English-language definition, but a .ref
reference can be used to obtain the data associated with an object. If two identifiers have the same
reference, then they are the same object.
return value: what an object returns, which might be of type double, string, array, or classname.
Generally, return value is used in discussions of member programs, but all objects have a return
value; they typically return a copy of themselves.
rvalue: an identifier that may appear to the right of the = assignment operator.
scope: how it is determined to what object an identifier references. .a.b might be interpreted in the
global context and literally mean .a.b, or it might be interpreted in an implied context to mean
.impliedcontext.a.b.
shared object: an object to which two or more different identifiers refer.
type: the type of a member variable or of a return value, which is double, string, array, or
classnam.
where
mvname stands for member variable name;
rvalue is defined in Appendix C.2 Assignment; and
type is classname | double | string | array .
class — Class programming 33
The .Declare built-in may be used to add a member variable to an existing class instance,
.id[.id[. . . ]] .Declare type newmvname = rvalue
.id[.id[. . . ]] .Declare newmvname = rvalue
where id is name | name[exp] , the latter being how you refer to an array element; exp must
evaluate to a number. If exp evaluates to a noninteger number, it is truncated.
where
lvalue is .id .id . . .
rvalue is
" string "
‘" string "’
#
exp
(exp)
.id .id . . .
.id .id . . . .pgmname pgm arguments
.id .id . . . .Super (classname) .pgmname pgm arguments
{}
{el ,el ,. . . }
When exp evaluates to a string, the result will contain at most 2045 characters and will be terminated
early if it contains a binary 0.
The last two syntaxes concern assignment to arrays; el may be
nothing
" string "
‘" string "’
#
(exp)
.id .id . . .
.id .id . . . .pgmname
[ .id .id . . . .pgmname pgm arguments ]
[ .id .id . . . .Super (classname) .pgmname pgm arguments ]
id is name | name[exp] , the latter being how you refer to an array element; exp must evaluate to
a number. If exp evaluates to a noninteger number, it is truncated.
34 class — Class programming
. . . ‘‘ref’’. . .
In the above, perhaps local tmpname was obtained from tempname (see [P] macro), and perhaps
local ref contains ‘‘.myobj.cvalue’’.
When a class object is quoted, its printable form is substituted. This is defined as
If the quoted reference results in an error, the error message is suppressed and nothing is substituted.
Built-in modifiers
Also see
[P] class exit — Exit class-member program and return result
[P] classutil — Class programming utility
[P] sysdir — Query and set system directories
[M-2] class — Object-oriented programming (classes)
[U] 17.5 Where does Stata look for ado-files?
Stata, Stata Press, and Mata are registered trademarks of StataCorp LLC. Stata and
®
Stata Press are registered trademarks with the World Intellectual Property Organization
of the United Nations. Other brand and product names are registered trademarks or
trademarks of their respective companies. Copyright c 1985–2023 StataCorp LLC,
College Station, TX, USA. All rights reserved.