CloudFlare suffered a massive security issue affecting all of its customers, including Rosetta Code. All passwords not changed since February 19th 2017 have been expired, and session cookie longevity will be reduced until late March.--Michael Mol (talk) 05:15, 25 February 2017 (UTC)

Runtime evaluation/In an environment

From Rosetta Code
Runtime evaluation/In an environment
You are encouraged to solve this task according to the task description, using any language you may know.
Given a program in the language (as a string or AST) with a free variable named x (or another name if that is not valid syntax), evaluate it with x bound to a provided value, then evaluate it again with x bound to another provided value, then subtract the result of the first from the second and return or print it.

Do so in a way which:

  • does not involve string manipulation of the input source code
  • is plausibly extensible to a runtime-chosen set of bindings rather than just x
  • does not make x a global variable

or note that these are impossible.

See also[edit]

ALGOL 68[edit]

Works with: ALGOL 68G version Any - tested with release mk15-0.8b.fc9.i386 - this implementation is an interpretor, and evaluate is an extension to the standard

Variable names are generally not visible at run time with classic compilers. However ALGOL 68G is an interpretor and it retains this ability. Note that evaluate returns a string.

PROC eval_with_x = (STRING code, INT a, b)STRING:
(INT x=a; evaluate(code) ) + (INT x=b; evaluate(code));
print((eval_with_x("2 ** x", 3, 5), new line))


         +8        +32


AppleScript's run script command allows to interpret and execute the string passed to it as an arbitrarily complex (or simple) script; such a string may thus be viewed as the "program" considered in the task description.
Each invocation of the run script command dynamically happens in a separate execution context, so there are no side-effects; on the other hand, this means that such an invocation is quite costly.
Arguments may be passed as a list of arbitrary values; this however requires the program to be written with an explicit run handler.
The result is the value (if any) returned by the program; any valid AppleScript value may be returned.

Given the above, the task may easily be implemented along these lines:

on task_with_x(pgrm, x1, x2)
local rslt1, rslt2
set rslt1 to run script pgrm with parameters {x1}
set rslt2 to run script pgrm with parameters {x2}
rslt2 - rslt1
end task_with_x

Example usage (for legibility purposes, the program is stored into an intermediate variable):

set pgrm_with_x to "
on run {x}

task_with_x(pgrm_with_x, 3, 5)

The result is 24.0 (a real number).


Works with: AutoHotkey_H

AutoHotkey does not provide an API to the local symbol table. Local variables are also not supported within scopes outside functions. However, a local environment can be simulated by wrapping code in a temporary function.

msgbox % first := evalWithX("x + 4", 5)
msgbox % second := evalWithX("x + 4", 6)
msgbox % second - first
evalWithX(expression, xvalue)
global script
script =
x = %xvalue% ; := would need quotes
return %expression%
renameFunction("expression", "") ; remove any previous expressions
gosub load ; cannot use addScript inside a function yet
exp := "expression"
return %exp%()
DllCall(A_AhkPath "\addScript","Str",script,"Uchar",0,"Cdecl UInt")
renameFunction(funcName, newname){
x%newname% := newname ; store newname in a static variable so its memory is not freed
strput(newname, &x%newname%, strlen(newname) + 1)
if fnp := FindFunc(funcName)
numput(&x%newname%, fnp+0, 0, "uint")


      expression$ = "x^2 - 7"
one = FN_eval_with_x(expression$, 1.2)
two = FN_eval_with_x(expression$, 3.4)
PRINT two - one
DEF FN_eval_with_x(expr$, x)
= EVAL(expr$)


( ( eval-with-x
= code a b argument
.  !arg:((=?code),?a,?b,?argument)
& (!b:?x&!code$!argument)
+ -1*(!a:?x&!code$!argument)
& out$(eval-with-x$((='(.$x^!arg)),3,5,2))
& out$(eval-with-x$((='(.$x^!arg)),12,13,2))




We must define x as global, but we use dynamic bindings. Only functions within the binding will see the newly bound value of x, before it re-establishes the bindings that existed before.

(def ^:dynamic x nil)
(defn eval-with-x [program a b]
(- (binding [x b] (eval program))
(binding [x a] (eval program))))
(eval-with-x '(* x x) 4 9)
=> 65

Common Lisp[edit]

(defun eval-with-x (program a b)
(let ((at-a (eval `(let ((x ',a)) ,program)))
(at-b (eval `(let ((x ',b)) ,program))))
(- at-b at-a)))
(eval-with-x '(exp x) 0 1)
=> 1.7182817

This version ensures that the program is compiled, once, for more efficient execution:

(defun eval-with-x (program a b)
(let* ((f (compile nil `(lambda (x) ,program)))
(at-a (funcall f a))
(at-b (funcall f b)))
(- at-b at-a)))

Déjà Vu[edit]

local fib n:
if <= n 1:
+ fib - n 1 fib - n 2
local :code !compile-string dup "-- fib x" #one less than the xth fibonacci number
!run-blob-in { :fib @fib :x 4 } code
!run-blob-in { :fib @fib :x 6 } code
!. -


# Constructing an environment has to be done by way of evaluation
#for historical reasons which will hopefully be entirely eliminated soon.
def bindX(value) {
def [resolver, env] := e` # bind x and capture its resolver and the
def x # resulting environment
resolver.resolve(value) # set the value
return env
def evalWithX(program, a, b) {
def atA := program.eval(bindX(a))
def atB := program.eval(bindX(b))
return atB - atA
? evalWithX(e`(x :float64).exp()`, 0, 1)
# value: 1.7182818284590455


We evaluate prog in a new environment which is an association list ((x x-value)), and could be ((x x-value) (y y-value)....)

(define (eval-with-x prog x)
(eval prog (environment-new (list (list 'x x)))))
(define prog '( + 1 (* x x)))
(eval-with-x prog 10)101
(eval-with-x prog 1000)1000001
(- (eval-with-x prog 1000) (eval-with-x prog 10))999900
;; check x is unbound (no global)
😖️ error: #|user| : unbound variable : x


Functions below are used by dynamic variable names. Any changes here needs to be backwards compatible, or dynamic variable names must also be changed.

-module( runtime_evaluation ).
-export( [evaluate_form/2, form_from_string/1, task/0] ).
evaluate_form( Form, {Variable_name, Value} ) ->
Bindings = erl_eval:add_binding( Variable_name, Value, erl_eval:new_bindings() ),
{value, Evaluation, _} = erl_eval:expr( Form, Bindings ),
form_from_string( String ) ->
{ok, Tokens, _} = erl_scan:string( String ),
{ok, [Form]} = erl_parse:parse_exprs( Tokens ),
task() ->
Form = form_from_string( "X." ),
Variable1 = evaluate_form( Form, {'X', 1} ),
io:fwrite( "~p~n", [Variable1] ),
Variable2 = evaluate_form( Form, {'X', 2} ),
io:fwrite( "~p~n", [Variable2] ),
io:fwrite( "~p~n", [Variable2 - Variable1] ).
14> runtime_evaluation:task().


EVALUATE invokes the Forth interpreter on the given string.

: f-" ( a b snippet" -- )
[char] " parse ( code len )
2dup 2>r evaluate
swap 2r> evaluate
- . ;
2 3 f-" dup *" \ 5 (3*3 - 2*2)

This can be used to treat a data stream as code, or to provide a lightweight macro facility when used in an IMMEDIATE word.

: :macro ( "name <char> ccc<char>" -- )
:macro times 0 do ;
: test 8 times ." spam " loop ;
see test
: test
8 0
DO .\" spam "
 ; ok


One way is to use a macro. In genyris, macros are lazy functions which execute twice, the return value is also evaluated in the caller's environment:

defmacro add100() (+ x 100)
var x 23
var firstresult (add100)
x = 1000
+ firstresult (add100)

This prints 1223.

Another way is to use dynamically scoped variables. In Genyris, symbols prefixed with a period are looked up in the caller's environment, not the lexical environment of the closure. When a dictionary is the first element of the expression, an environment is created and the &rest is evaluated.

def add100() (+ .x 100)
(dict) # create an environment capable of holding dynamic bindings
var .x 23 # create a binding in the dictionary
var firstresult (add100)
.x = 1000
+ firstresult (add100)

Dictionaries can hold bindings to dynamic symbols. To minimize the danger of dynamic scope there is no recursive ascent in the binding lookup.

var .x 23
print .x # fails


package main
import (
func main() {
// an expression on x
squareExpr := "x*x"
// parse to abstract syntax tree
fset := token.NewFileSet()
squareAst, err := parser.ParseExpr(squareExpr)
if err != nil {
// create an environment or "world"
w := eval.NewWorld()
// allocate a variable
wVar := new(intV)
// bind the variable to the name x
err = w.DefineVar("x", eval.IntType, wVar)
if err != nil {
// bind the expression AST to the world
squareCode, err := w.CompileExpr(fset, squareAst)
if err != nil {
// directly manipulate value of variable within world
*wVar = 5
// evaluate
r0, err := squareCode.Run()
if err != nil {
// change value
// revaluate
r1, err := squareCode.Run()
if err != nil {
// print difference
fmt.Println(r0.(eval.IntValue).Get(nil) - r1.(eval.IntValue).Get(nil))
// int value implementation.
type intV int64
func (v *intV) String() string { return fmt.Sprint(*v) }
func (v *intV) Get(*eval.Thread) int64 { return int64(*v) }
func (v *intV) Set(_ *eval.Thread, x int64) { *v = intV(x) }
func (v *intV) Assign(t *eval.Thread, o eval.Value) {
*v = intV(o.(eval.IntValue).Get(t))




The solution:

def cruncher = { x1, x2, program ->
Eval.x(x1, program) - Eval.x(x2, program)

Test Program:

def fibonacciProgram = '''
x < 1 ? 0 : x == 1 ? 1 : (2..x).inject([0,1]){i, j -> [i[1], i[0]+i[1]]}[1]

println "F(${10}) - F(${5}) = ${Eval.x(10, fibonacciProgram)} - ${Eval.x(5, fibonacciProgram)} = " + cruncher(10, 5, fibonacciProgram)


F(10) - F(5) = 55 - 5 = 50



The following satisfies the requirements:

   EvalWithX=. monad : 0
'CODE V0 V1'=. y
(". CODE [ x=. V1) - (". CODE [ x=. V0)
EvalWithX '^x';0;1


However, it is easier via point-free coding:

   (0&({::) -~&>/@:(128!:2&.>) 1 2&{) '^';0;1

Explicit again[edit]

Or, using y as the free variable, instead of x:

EvalDiffWithY=: dyad define
-~/verb def x"_1 y

Example use:

   '^y' EvalDiffWithY 0 1

This can be extended to support a user declared argument name:

EvalDiffWithName=: adverb define
-~/m adverb def ('(m)=.y';x)"_1 y

This works by preceding the user provided expression with a statement which assigns the argument value to a local variable whose name was provided by the user. [Note that this implementation skirts the requirement that the implementation does not manipulate strings -- instead we manipulate a structure containing strings.]

Example use:

  '^George' 'George' EvalDiffWithName 0 1
'Z + 2^Z' 'Z' EvalDiffWithName 2 3

Of course this could be re-defined such that the free variable declaration appears to the left of the expression ('Z' 'Z + 2^Z' Example 2 3). However, J's currying and precedence rules might make that less convenient to use, if this were ever used in a real program.


Works with: Java version 6+

Although Java is a compiled static language and expression evaluation is not intrinsic part of Java, we can still generate a class at run time and so emulate evaluation of string expressions. Java 1.6 provides some APIs for this sort of thing.


  • this is not thread-safe because it writes a generated class to the file system, then loads it
  • the supplied code to evaluate is assumed to be an expression rather than a series of statements
  • the supplied expression should evaluate to a number
  • the same class is generated twice - never mind
  • it's painfully verbose, but we're bending the language quite a bit
  • the exception handling is minimal, but if something goes wrong you should get a stack dump and the exception might be helpful...
import java.lang.reflect.Method;
import java.util.Arrays;
public class Eval {
private static final String CLASS_NAME = "TempPleaseDeleteMe";
private static class StringCompiler
extends SimpleJavaFileObject {
final String m_sourceCode;
private StringCompiler( final String sourceCode ) {
super( URI.create( "string:///" + CLASS_NAME + Kind.SOURCE.extension ), Kind.SOURCE );
m_sourceCode = sourceCode;
public CharSequence getCharContent( final boolean ignoreEncodingErrors ) {
return m_sourceCode;
private boolean compile() {
final JavaCompiler javac = ToolProvider.getSystemJavaCompiler();
return javac.getTask( null, javac.getStandardFileManager( null, null, null ),
null, null, null, Arrays.asList( this )
private double callEval( final double x )
throws Exception {
final Class<?> clarse = Class.forName( CLASS_NAME );
final Method eval = clarse.getMethod( "eval", double.class );
return ( Double ) eval.invoke( null, x );
public static double evalWithX( final String code, final double x )
throws Exception {
final StringCompiler sc = new StringCompiler(
"class "
+ "{public static double eval(double x){return ("
+ code
+ ");}}"
if ( ! sc.compile() ) throw new RuntimeException( "Compiler error" );
return sc.callEval( x );
public static void main( final String [] args )
throws Exception /* lazy programmer */ {
final String expression = args [ 0 ];
final double x1 = Double.parseDouble( args [ 1 ] );
final double x2 = Double.parseDouble( args [ 2 ] );
evalWithX( expression, x1 )
- evalWithX( expression, x2 )

Example usage - calculating the difference of two squares (i.e. 9 - 2 = 7):

java Eval "Math.pow(x,2)" 3 1.414213562373095




eval uses the environment from the calling function.

function evalWithX(expr, a, b) {
var x = a;
var atA = eval(expr);
x = b;
var atB = eval(expr);
return atB - atA;
evalWithX('Math.exp(x)', 0, 1) // returns 1.718281828459045

Liberty BASIC[edit]

expression$ = "x^2 - 7"
Print (EvaluateWithX(expression$, 5) - EvaluateWithX(expression$, 3))
Function EvaluateWithX(expression$, x)
EvaluateWithX = Eval(expression$)
End Function


code = loadstring"return x^2" --this doesn't really need to be input, does it?
val1 = setfenv(code, {x = + 0})()
val2 = setfenv(code, {x = + 0})()
print(val2 - val1)

In Lua 5.2 one can use the new load function to evaluate a string as Lua code and specify its environment:

Works with: Lua version 5.2
env = {}
f = load("return x", nil, nil, env)
env.x = tonumber( -- user enters 2
a = f()
env.x = tonumber( -- user enters 3
b = f()
print(a + b) --> outputs 5


Input source code is "10 x" , X is locally bound to 3 & 2 and the resulting expressions evaluated.
(10 x /. x -> 3 ) - (10 x /. x -> 2 )
-> 10

MATLAB / Octave[edit]

In Octave, undeclared variables are local.

function r = calcit(f, val1, val2)
x = val1;
a = eval(f);
x = val2;
b = eval(f);
r = b-a;


p = 'x .* 2';
disp(calcit(p, [1:3], [4:6]));


6   6   6


vardef evalit(expr s, va, vb) =
save x,a,b; x := va; a := scantokens s;
x := vb; b := scantokens s; a-b
show(evalit("2x+1", 5, 3));


The ooRexx interpret instruction executes dynamically created ooRexx code in the current variable context.

say evalWithX("x**2", 2)
say evalWithX("x**2", 3.1415926)
::routine evalWithX
use arg expression, x
-- X now has the value of the second argument
interpret "return" expression




fun {EvalWithX Program A B}
{Compiler.evalExpression Program env('X':B) _}
{Compiler.evalExpression Program env('X':A) _}
{Show {EvalWithX "{Exp X}" 0.0 1.0}}


There are many ways of doing this depending on the particular interpretation of the requirements. This code assumes that f is a string representing a GP closure.



sub eval_with_x
{my $code = shift;
my $x = shift;
my $first = eval $code;
$x = shift;
return eval($code) - $first;}
print eval_with_x('3 * $x', 5, 10), "\n"; # Prints "15".

Perl 6[edit]

Works with: rakudo version 2015-12-22

For security, you must explicitly allow use of 'EVAL'.

sub eval_with_x($code, *@x) { [R-] -> \x { EVAL $code } }
say eval_with_x('3 * x', 5, 10); # Says "15".
say eval_with_x('3 * x', 5, 10, 50); # Says "105".


function eval_with_x($code, $a, $b) {
$x = $a;
$first = eval($code);
$x = $b;
$second = eval($code);
return $second - $first;
echo eval_with_x('return 3 * $x;', 5, 10), "\n"; # Prints "15".


(let Expression '(+ X (* X X))            # Local expression
(let X 3
(eval Expression) )
(let X 4
(eval Expression) ) ) )
(let Function (list '(X) Expression) # Build a local function
(Function 3)
(Function 4) ) ) ) )




Pike can only compile complete classes. therefore binding a value to a variable is only possible by string manipulation. even Pikes own interactive mode which seemingly evaluates expressions wraps them into a class and replaces variable references before compiling:

> int x=10;
Result: 10
> x * 5;
Result: 50
> dump wrapper
Last compiled wrapper:
001: mapping(string:mixed) ___hilfe = ___Hilfe->variables;
002: # 1
003: mixed ___HilfeWrapper() { return (([mapping(string:int)](mixed)___hilfe)->x) * 5; ; }

___Hilfe is an object which stores all created variables;

to solve the problem in the task i would create a function that can take arguments:

string payload = "x * 5";
program demo = compile_string("string eval(mixed x){ " + payload + "; }");
Result: 50
Result: 100


>>> def eval_with_x(code, a, b):
return eval(code, {'x':b}) - eval(code, {'x':a})
>>> eval_with_x('2 ** x', 3, 5)

A slight change allows the evaluation to take multiple names:

>>> def eval_with_args(code, **kwordargs):
return eval(code, kwordargs)
>>> code = '2 ** x'
>>> eval_with_args(code, x=5) - eval_with_args(code, x=3)
>>> code = '3 * x + y'
>>> eval_with_args(code, x=5, y=2) - eval_with_args(code, x=3, y=1)


We can set up thing so that the "unbound" variable can be any accepted symbol for variables.

evalWithAB <- function(expr, var, a, b) {
env <- new.env() # provide a separate env, so that the choosen
assign(var, a, envir=env) # var name do not collide with symbols inside
# this function (e.g. it could be even "env")
atA <- eval(parse(text=expr), env)
# and then evaluate the expression inside this
# ad hoc env-ironment
assign(var, b, envir=env)
atB <- eval(parse(text=expr), env)
return(atB - atA)
print(evalWithAB("2*x+1", "x", 5, 3))
print(evalWithAB("2*y+1", "y", 5, 3))
print(evalWithAB("2*y+1", "x", 5, 3)) # error: object "y" not found


Same hack as the on in the CL/Scheme entries:

#lang racket
(define ns (make-base-namespace))
(define (eval-with-x code a b)
(define (with v) (eval `(let ([x ',v]) ,code) ns))
(- (with b) (with a)))

Better: a more direct use of eval with just the code (for example, this won't break if we use a namespace with a different meaning for let, which is very possible in Racket):

#lang racket
(define ns (make-base-namespace))
(define (eval-with-x code a b)
(define (with v)
(namespace-set-variable-value! 'x v #f ns)
(eval code ns))
(- (with b) (with a)))


prog: [x * 2]
fn: func [x] [do bind prog 'x]
a: fn 2
b: fn 4
subtract b a




/*REXX program to demonstrate some run-time evaulations.                */
say b-a
exit /*stick a fork in it, we're done.*/
/*───────────────────────────────────FACT subroutine────────────────────*/
fact: procedure; parse arg n; !=1; do j=2 to n;  !=!*j; end; return !




expression = "return pow(x,2) - 7"
one = evalwithx(expression, 1.2)
two = evalwithx(expression, 3.4)
see "one = " + one + nl + "two = " + two + nl
func evalwithx expr, x
return eval(expr)


one = -5.56
two = 4.56


def bind_x_to_value(x)
def eval_with_x(code, a, b)
eval(code, bind_x_to_value(b)) - eval(code, bind_x_to_value(a))
puts eval_with_x('2 ** x', 3, 5) # Prints "24"

The magic here is how the binding method works with the bind_x_to_value(x) method. When bind_x_to_value is called, it sets its local variable x to the value passed. The binding method then returns a reference to the current context (or stack frame) to the caller. eval can then use the local variable x in this context.


Almost identical to the Common Lisp version above.

(define (eval-with-x prog a b)
(let ((at-a (eval `(let ((x ',a)) ,prog)))
(at-b (eval `(let ((x ',b)) ,prog))))
(- at-b at-a)))


func eval_with_x(code, x, y) {
var f = eval(code);
x = y;
eval(code) - f;
say eval_with_x(x: 3, y: 5, code: '2 ** x'); # => 24


This program defines (at runtime) a function triple(), compiles it, and then executes it twice, with values x = 1 and then x = 3. The program subtracts the returned value from the first call from the value returned from the first call, and prints the result. In this example, the value x is passed as a parameter to the function triple().

     compiled = code(' define("triple(x)") :(a);triple triple = 3 * x :(return)')  :<compiled>
a x = 1
first = triple(x)
x = 3
output = triple(x) - first



If you specifically wanted to not pass x as a parameter but instead use it as a value from the environment, that's easy too:

     compiled = code(' define("triple()") :(a);triple triple = 3 * x :(return)')  :<compiled>
a x = 1
first = triple(x)
x = 3
output = triple(x) - first

The output is the same.


proc eval_twice {func a b} {
set x $a
set 1st [expr $func]
set x $b
set 2nd [expr $func]
expr {$2nd - $1st}
puts [eval_twice {2 ** $x} 3 5] ;# ==> 24

Here's another take, similar to other answers. It passes a code block to be evaled, not just an expression for expr

proc eval_with_x {code val1 val2} {
expr {[set x $val2; eval $code] - [set x $val1; eval $code]}
eval_with_x {expr {2**$x}} 3 5 ;# ==> 24

In 8.5, apply makes environments like this "first class":

Works with: Tcl version 8.5
package require Tcl 8.5
proc eval_with {body a b} {
set lambda [list x $body]
expr {[apply $lambda $b] - [apply $lambda $a]}
eval_with {expr {2**$x}} 3 5 ;# ==> 24


Translation of: Common Lisp

In TXR's embedded Lisp dialect, we can implement the same solution as Lisp or Scheme: transform the code fragment by wrapping a let around it which binds a variable, and then evaluating the whole thing:

(defun eval-subtract-for-two-values-of-x (code-fragment x1 x0)
(- (eval ^(let ((x ,x1)) ,code-fragment))
(eval ^(let ((x ,x0)) ,code-fragment))))
(eval-subtract-for-two-values-of-x 1 2) ;; yields -4.67077427047161

Cutting edge TXR code provides access to the environment manipulation functions, making this possible:

(defun eval-subtract-for-two-values-of-x (code-fragment x1 x0)
(let ((e1 (make-env (list (cons 'x x1))))  ;; create two environments stuffed with binding for x
(e0 (make-env (list (cons 'x x0)))))
(- (eval code-fragment e1)  ;; pass these environment to eval
(eval code-fragment e0))))
(eval-subtract-for-two-values-of-x '(exp x) 1 2)

Alternatively, empty environments can be made and extended with bindings:

(defun eval-subtract-for-two-values-of-x (code-fragment x1 x0)
(let ((e1 (make-env))
(e0 (make-env)))
(env-vbind e1 'x x1)
(env-vbind e0 'x x0)
(- (eval code-fragment e1)
(eval code-fragment e0))))
(eval-subtract-for-two-values-of-x '(exp x) 1 2)

Explicit environment manipulation has the disadvantage of being hostile against compiling. (See notes about compilation in the Common Lisp example.)

there is an eval function which takes an environment parameter. However, currently there isn't any access to the manipulation of environment objects. It's probably a bad idea because run time tricks with lexical environments lead to programs that are not compilable.

Lastly, we can also solve this problem using dynamically scoped (a.k.a "special") variables. The problem description specifically says that the solution is not to use global variables. Though we must define the variables as global, we do not use the global bindings; we use dynamic bindings.

There is a hidden global variable, namely the dynamic environment itself. That's how eval is able to resolve the free-variable x occurring in code-fragment without receiving any environment parameter.

However, our two let constructs carefully save and restore the dynamic environment (and therefore any prior value of x), even in the face of exceptions, and

(defvar x)
(defun eval-subtract-for-two-values-of-x (code-fragment x1 x0)
(- (let ((x x1)) (eval code-fragment))
(let ((x x0)) (eval code-fragment))))
(eval-subtract-for-two-values-of-x '(exp x) 1 2)

TI-89 BASIC[edit]

evalx(prog, a, b)
Local x,eresult1,eresult2
Return eresult2-eresult1
■ evalx("ℯ^x", 0., 1)

There are no facilities for control over the environment; expr() evaluates in the same environment as the caller, including local variables. [Someone please verify this statement.]

UNIX Shell[edit]

The backquotes ` ... ` capture the standard output of a subshell. Changes to parameter x in the subshell will not affect its parent shell.

eval_with_x() {
set -- "`x=$2; eval "$1"`" "`x=$3; eval "$1"`"
expr "$2" - "$1"
eval_with_x '
# compute 2 ** $x
while test $x -gt 0; do
p=`expr $p \* 2`
x=`expr $x - 1`
echo $p
3 5
# Prints '24'


fcn evalWithX(text,x) {
f.x = x; // set free var in compiled blob
f.__constructor(); // run blob
vm.regX // compiler sets the VMs X register for cases like this
const TEXT="var x; x*2"; // variables need to be declared
evalWithX(TEXT,5) - evalWithX(TEXT,3) #--> 4

This is not a complete solution but close.

Another way to do this is to create a class on the fly that contains the code to be run and reusing that class. The class just acts like as a container for x and a function:

var klass=Compiler.Compiler.compileText("var x; returnClass(x*2)");
(klass.__constructor(klass.x=5) - klass.__constructor(klass.x=3)).println();

returnClass(x) is required in a constructor if you want to return something other than self. klass.x=y pokes y into the instance variable x. Running the constructor runs x*2.