| Jump to: | OMake Home • Guide Home • Guide (single-page) • Contents (short) • Contents (long) | |
| Index: | All • Variables • Functions • Objects • Targets • Options |
omake provides a full programming-language including many system and IO functions. The language is object-oriented – everything is an object, including the base values like numbers and strings. The omake language can be characterized as follows:
export directive, instead of
directly mutating variables
To illustrate these features, we will use the osh(1) omake program shell.
The osh(1) program provides a toploop, where expressions can be entered
and the result printed. osh(1) normally interprets input as command text
to be executed by the shell, so in many cases we will use the value
form to evaluate an expression directly.
osh> 1
*** omake error: File -: line 1, characters 0-1 command not found: 1
osh> value 1
- : "1" : Sequence
osh> ls -l omake
-rwxrwxr-x 1 jyh jyh 1662189 Aug 25 10:24 omake*
Dynamic scoping means that the value of a variable is determined by the most recent binding of the variable in scope at runtime. Consider the following program.
OPTIONS = a b c
f() =
println(OPTIONS = $(OPTIONS))
g() =
OPTIONS = d e f
f()
If f() is called without redefining the OPTIONS variable,
the function should print the string OPTIONS = a b c.
In contrast, the function g() redefines the OPTIONS
variable and evaluates f() in that scope, which now prints the
string OPTIONS = d e f.
The body of g defines a local scope – the redefinition of the
OPTIONS variable is local to g and does not persist
after the function terminates.
osh> g()
OPTIONS = d e f
osh> f()
OPTIONS = a b c
Dynamic scoping can be tremendously helpful for simplifying the code
in a project. For example, the OMakeroot file defines a set of
functions and rules for building projects using such variables as
CC, CFLAGS, etc. However, different parts of a project
may need different values for these variables. For example, we may
have a subdirectory called opt where we want to use the
-03 option, and a subdirectory called debug where we
want to use the -g option. Dynamic scoping allows us to redefine
these variables in the parts of the project without having to
redefine the functions that use them.
section
CFLAGS = -O3
.SUBDIRS: opt
section
CFLAGS = -g
.SUBDIRS: debug
However, dynamic scoping also has drawbacks. First, it can become confusing: you might have a variable that is intended to be private, but it is accidentally redefined elsewhere. For example, you might have the following code to construct search paths.
PATHSEP = :
make-path(dirs) =
return $(concat $(PATHSEP), $(dirs))
make-path(/bin /usr/bin /usr/X11R6/bin)
- : "/bin:/usr/bin:/usr/X11R6/bin" : String
However, elsewhere in the project, the PATHSEP variable is
redefined as a directory separator /, and your function
suddenly returns the string /bin//usr/bin//usr/X11R6/bin,
obviously not what you want.
The private block is used to solve this problem. Variables
that are defined in a private block use static scoping – that
is, the value of the variable is determined by the most recent
definition in scope in the source text.
private
PATHSEP = :
make-path(dirs) =
return $(concat $(PATHSEP), $(dirs))
PATHSEP = /
make-path(/bin /usr/bin /usr/X11R6/bin)
- : "/bin:/usr/bin:/usr/X11R6/bin" : String
This has two aspects: First of all, functions are values like other values:
p(f, x) =
y = $(f $(x), 1)
println($"The value is $(y)")
p($(add), 5) # prints 6
p($(sub), 5) # prints 4
The other aspect is that variables (and thus the whole environment) can exist in several versions: The assignment to a variable creates first a second version in the current block, and is not directly applied to the orignal variable, unless it is finally “exported”.
(Note that in previous versions of the manual you could read here that there “is no assignment operator”. On the surface this is of course not true, as we provide such an operator. This comment referred to the implementation, which represents environments as functional maps from names to values, and reduces assignment to a functional update of the current environment, yielding to a new version.)
The export directive can be used to propagate all or part of an inner scope back to its
parent. If used without
arguments, the entire scope is propagated back to the parent; otherwise the arguments specify which
part of the environment to propagate. The most common usage is to export some or all of the definitions in a
conditional block. In the following example, the variable B is bound to 2 after the
conditional. The A variable is not redefined.
if $(test)
A = 1
B = $(add $(A), 1)
export B
else
B = 2
export
If the export directive is used without an argument, all of the following is exported:
If the export directive is used with an argument, the argument expression is evaluated
and the resulting value is interpreted as follows:
export function, then the corresponding environment or partial
environment is exported.
For example, in the following (somewhat artificial) example, the variables A and B
will be exported, and the implicit rule will remain in the environment after the section ends, but
the variable TMP and the target tmp_phony will remain unchanged.
section
A = 1
B = 2
TMP = $(add $(A), $(B))
.PHONY: tmp_phony
tmp_phony:
prepare_foo
%.foo: %.bar tmp_phony
compute_foo $(TMP) $< $@
export A B .RULE
This feature was introduced in version 0.9.8.5.
The export directive does not need to occur at the end of a block. An export is valid from
the point where it is specified to the end of the block in which it is contained. In other words,
the export is used in the program that follows it. This can be especially useful for reducing the
amount of code you have to write. In the following example, the variable CFLAGS is exported
from the both branches of the conditional.
export CFLAGS
if $(equal $(OSTYPE), Win32)
CFLAGS += /DWIN32
else
CFLAGS += -UWIN32
This feature was introduced in version 0.9.8.5.
The use of export does not affect the value returned by a block. The value is computed as usual, as the value of the last statement in the block, ignoring the export. For example, suppose we wish to implement a table that maps strings to unique integers. Consider the following program.
# Empty map
table = $(Map)
# Add an entry to the table
intern(s) =
export
if $(table.mem $s)
table.find($s)
else
private.i = $(table.length)
table = $(table.add $s, $i)
value $i
intern(foo)
intern(boo)
intern(moo)
# Prints "boo = 1"
println($"boo = $(intern boo)")
Given a string s, the function intern returns either the value already associated with
s, or assigns a new value. In the latter case, the table is updated with the new value. The
export at the beginning of the function means that the variable table is to be
exported. The bindings for s and i are not exported, because they are private.
Evaluation in omake is eager. That is, expressions are evaluated as soon as they are encountered by the evaluator. One effect of this is that the right-hand-side of a variable definition is expanded when the variable is defined.
osh> A = 1
- : "1"
osh> A = $(A)$(A)
- : "11"
In the second definition, A = $(A)$(A), the right-hand-side is evaluated first, producing the
sequence 11. Then the variable A is redefined as the new value. When combined
with dynamic scoping, this has many of the same properties as conventional imperative programming.
osh> A = 1
- : "1"
osh> printA() =
println($"A = $A")
osh> A = $(A)$(A)
- : "11"
osh> printA()
11
In this example, the print function is defined in the scope of A. When it is called on
the last line, the dynamic value of A is 11, which is what is printed.
However, dynamic scoping and imperative programming should not be confused. The following example
illustrates a difference. The second printA is not in the scope of the definition
A = x$(A)$(A)x, so it prints the original value, 1.
osh> A = 1
- : "1"
osh> printA() =
println($"A = $A")
osh> section
A = x$(A)$(A)x
printA()
x11x
osh> printA()
1
See also Section 7.7 for further ways to control the evaluation order through the use of “lazy” expressions.
omake is an object-oriented language. Everything is an object, including
base values like numbers and strings. In many projects, this may not be so apparent
because most evaluation occurs in the default toplevel object, the Pervasives
object, and few other objects are ever defined.
However, objects provide additional means for data structuring, and in some cases judicious use of objects may simplify your project.
Objects are defined with the following syntax. This defines name
to be an object with several methods an values.
name. = # += may be used as well
extends parent-object # optional
class class-name # optional
# Fields
X = value
Y = value
# Methods
f(args) =
body
g(arg) =
body
An extends directive specifies that this object inherits from
the specified parent-object. The object may have any number of
extends directives. If there is more than on extends
directive, then fields and methods are inherited from all parent
objects. If there are name conflicts, the later definitions override
the earlier definitions.
The class directive is optional. If specified, it defines a name
for the object that can be used in instanceof operations, as well
as :: scoping directives discussed below.
The body of the object is actually an arbitrary program. The variables defined in the body of the object become its fields, and the functions defined in the body become its methods.
The fields and methods of an object are named using object.name notation.
For example, let’s define a one-dimensional point value.
Point. =
class Point
# Default value
x = $(int 0)
# Create a new point
new(x) =
x = $(int $(x))
return $(this)
# Move by one
move() =
x = $(add $(x), 1)
return $(this)
osh> p1 = $(Point.new 15)
osh> value $(p1.x)
- : 15 : Int
osh> p2 = $(p1.move)
osh> value $(p2.x)
- : 16 : Int
The $(this) variable always represents the current object.
The expression $(p1.x) fetches the value of the x field
in the p1 object. The expression $(Point.new 15)
represents a method call to the new method of the Point
object, which returns a new object with 15 as its initial value. The
expression $(p1.move) is also a method call, which returns a
new object at position 16.
Note that objects are functional — it is not possible to modify the fields
or methods of an existing object in place. Thus, the new and move
methods return new objects.
Suppose we wish to create a new object that moves by 2 units, instead of
just 1. We can do it by overriding the move method.
Point2. =
extends $(Point)
# Override the move method
move() =
x = $(add $(x), 2)
return $(this)
osh> p2 = $(Point2.new 15)
osh> p3 = $(p2.move)
osh> value $(p3.x)
- : 17 : Int
However, by doing this, we have completely replaced the old move method.
Suppose we wish to define a new move method that just calls the old one twice.
We can refer to the old definition of move using a super call, which uses the notation
$(classname::name <args>). The classname should be the name of the
superclass, and name the field or method to be referenced. An alternative
way of defining the Point2 object is then as follows.
Point2. =
extends $(Point)
# Call the old method twice
move() =
this = $(Point::move)
return $(Point::move)
Note that the first call to $(Point::move) redefines the
current object (the this variable). This is because the method
returns a new object, which is re-used for the second call.
| Jump to: | OMake Home • Guide Home • Guide (single-page) • Contents (short) • Contents (long) | |
| Index: | All • Variables • Functions • Objects • Targets • Options |
