6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
6.001 Notes,Section 15.6
Slide 15.6.1
The next stage in the evolution of our evaluator is to pull the
environment out as an explicit parameter,Up until now we
could rely on just having a single environment in which to store
bindings for variables,It made sense to have a global
environment in which to hold names for procedures and global
variables,We are about to start moving towards application,in
fact we built a simple version of application but we want to
extend that to procedures that need local environments,Thus
let's pull out the environment explicitly,So we will have to
change our evaluator from simply evaluating expressions to
evaluating expressions with respect to some environment,
What changes to we need? All procedures must now call
eval with that extra argument of an environment,Lookup and define* are going to use that
environment argument to specify where to work,
Slide 15.6.2
This is actually a really boring change! The functionality of
this evaluator is exactly the same as the previous version,All
we do is extend eval to take an expression and an
environment as arguments,and then make sure that all the
dispatch functions also take environments as arguments,Other
than make those changes,everything else is exactly as before,
Of course now when we evaluate an expression we have to
specify what environment,and we will need an initial or global
environment to start with,This will contain bindings for built in
procedures,
Otherwise this is just bookkeeping,It will allow us to extend
our evaluator in a more general way,
Slide 15.6.3
Notice in fact that the only non-trivial case is that dealing with
application inside of eval,Remember that this is the dispatch
that says that we have to apply the object that represents the
procedure to its arguments,Notice the change we need,As
before,we will evaluate the first subexpression of the tree
structure to get the procedure,now including the environment
as an argument,
In order to get the values of the other expressions before we
could just map eval down the list of expressions,Here,
however,we have to evaluate each expression with respect to
the environment so we need a lambda here to capture that
procedure,
We will come back to this shortly,but you can see from a Scheme perspective this makes sense,We are mapping a
procedure that evaluates each subexpression with respect to the environment passed in,This is the only non-trivial
extension in this case,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
6.001 Notes,Section 15.7
Slide 15.7.1
Now we are ready to make our last big extension to our
interpreter,We would like to be able to add new procedures,If
you think about it,once we add this ability we have all the basic
components of an interpreter,we can deal with primitive
expressions,with special forms,and with applications,and as
along as we can create our own procedures,those applications
could include arbitrary nesting of expressions,
How do we add new procedures? We would like to create a
name for something we make with the equivalent of a
lambda,We want to capture procedures within lambda's
and then use that procedure as an application,
Slide 15.7.2
To do this,we need to add to our evaluator the ability to capture
a computation in a procedure,Earlier in the term we saw this
with Scheme,that a lambda expression could capture a
common pattern of computation,Now we are going to see how
to add that ability to an evaluator,by creating something that
will deal with lambda-like expressions,
Here is our strategy for accomplishing this,First,we will need
another dispatch in our evaluator,something that deals
explicitly with lambda* expressions,We would like the
value of the lambda* expression to return a compound
procedure,however we decide to define it,Second,given the
ability to create our own procedures,we will need to extend apply to handle such procedures,Right now
apply is just set up to handle built in things we inherit from Scheme,but we need to have it also handle
procedures we create,And then we should go back and finish the environment model,since it will become
essential in the application of procedures,
Slide 15.7.3
Adding the dispatch to eval is straightforward,We'll need a
new tag checker for lambda* expressions,and a new
dispatch clause inside eval to a procedure to handle the
create of lambda* objects,Notice where we add this
dispatch clause,it must come before the application case,since
lambda* is a special form,It should really also come after
the checks for the primitive expressions (numbers or symbols),
The ordering of where within the special forms it falls is not
crucial,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.7.4
So what should eval-lambda do? Remember that
eval takes as input a tree structure representing a
lambda* expression,which it then passes to this procedure,
together with a pointer to the environment,So first,we need to
walk down that tree structure,and grab off the arguments of the
expression,the formal parameters if you like,That's what the
cadr does,And we want to walk further along the tree and
grab the body of the expression,that is the caddr,Having
pulled off those two pieces,without evaluation,we then want to
glue together some representation of a procedure as a set of
arguments,a body,and an environment,This should sound familiar,as it sounds a lot like what we have when we
do evaluation with our "double bubble" notation in the environment model,There we have a similar representation
for a procedure,formal parameters,body and environment,Make-compound is doing a similar thing here,
Note that there is no call to eval inside eval-lambda,None of these expressions is being evaluated; we
are simply manipulating tree structure,
Also notice that we have made a decision about the body of a procedure,By using the caddr of the expression,
we are assuming that the body contains exactly one expression,That is a design choice that we could change,
Slide 15.7.5
So adding the new form to create procedure objects is pretty
straightforward,We will have to eventually work out what
make-compound does,but that is basically an
abstraction issue,
Now the question is,how do I change apply? Remember
that apply is going to evaluate the first subexpression of a
tree structure,get out the appropriate procedure,evaluate all the
arguments getting back a list of those values,and then we want
to apply that procedure to that list,
In the case of primitive things,inherited things from Scheme
like +,* etc.,apply was just a matter of doing the basic
operation,Now,we need to extend apply,
Slide 15.7.6
In this case,when we go to apply an operator to a list of
operands,we will need to check for type,Is it a primitive
operation? If yes,we just apply the underlying Scheme
procedure as before,Is it a compound operator,that is,are we
evaluating an expression in which we are using a procedure that
we created in our evaluator using lambda*? In this case we
need to implement the idea of the environment model,
Thus,we get the body of the operator,that is,we pull out the
piece that we glued together when we made the procedure
object,Now we want to evaluate that body,But we are going to
evaluate that with respect to an environment,created by taking
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
the operands passed in,taking the formal parameters of the procedure object,and binding them together,This new
environment should extend the previous one,in case we need things from there,
Thus when we have a compound application,we will grab the body of the procedure and evaluate it in a new
environment,which has access to everything in the previous environment,plus bindings for the formal parameters
of the procedure to the values passed in,
Note what this does,Given this version of apply we have a wonderfully cyclic structure,Eval of an
expression with respect to an environment in the general case will reduce to applying a procedure to a set of
arguments,and that will reduce in the general case to evaluating a new expression (the body of the operator)
with respect to a new environment,In this way,eval and apply will work hand-in-hand in terms of
unwrapping the abstractions of the procedures down to primitive operations on primitive objects,
Slide 15.7.7
All that is left to do is to implement the "double bubble" idea as
an abstract data type,We can do that by creating a tag for a
compound procedure and having the constructor (make
compound) simply glue things together in a list,with the
appropriate predicate and selectors associated with this data
abstraction,This is not the only way to do this; all we need is
some method for gluing things together and getting them back
apart,
Slide 15.7.8
Looking at the code helps us understand our changes,but it is
probably easier to see the changes by watching the evolution of
the evaluation of an expression,Let's trace through the
evaluation of a lambda* expression,That is,we are
evaluating a tree structure whose first element is the symbol
lambda*,whose second element is a list of parameters,and
whose third element is an expression constituting the body of
the procedure,We are evaluating this with respect to some
environment,say the global environment,
Slide 15.7.9
As before,please follow along in the code (for part 6),
So eval takes in that tree structure representing this
expression and checks it for different types of expressions
(number,symbol,etc.),Eventually eval reaches the case for
handling lambda* expressions,Thus,we will dispatch on
type to eval-lambda,the specific procedure designed to
evaluate lambda*'s,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.7.10
Eval-lambda starts with the same tree structure as an
argument,plus some pointer to the global environment (GE),
What does it do?
It walks down the tree structure of the first argument,and grab
the arguments (that's the cadr),Similarly,walk down the tree
structure and grab the body (that's the caddr) (note that we
are assuming there is only one expression in the body),In this
case,the cadr will return the list (x*) and the caddr
will return the list (plus* x* x*),Let me stress again
that we are simply getting these as tree structure; there is no
evaluation yet,Having acquired those two structures,we then
glue them together using make-compound,
Slide 15.7.11
So the evaluation reduces to this expression,This is simply
using our constructor to glue these pieces together,and in this
case that converts to this form...,
Slide 15.7.12
..,and thus we are down to a basic primitive,We are going to
construct a new list structure,with a type tag,a list of
parameters,a body and the environment in which this
lambda* expression was created,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.7.13
And that says that the evaluation of this expression returns this
list structure,That is actually quite nice! We have created a new
object in our language,we have evaluated a lambda* and
created a procedure object,When we go to do an application (as
we will see shortly) we need an expression represented as list
structure so that we can pull out the pieces,and here they are,
All of the parts we will need for an application are represented
as list structure within this list,so that we easily use those
within the evaluator,
Slide 15.7.14
Thus,this piece of list structure represents a procedure,It is our
particular way of creating a data structure for a procedure,it has
a tag identifying it as a compound procedure,a set of
parameters,and a body,all represented in the format we expect
as a parsed tree,
Slide 15.7.15
Now that we can create procedures in our interpreter,we
naturally want to give them names,and that just uses our
define* expressions,How does this work? First of all,we
will have some initial bindings in our environment,.,
Slide 15.7.16
..,as shown here in our abstract table representation for an
environment,This environment includes prior bindings for
names to numbers and for symbols to Boolean values,We
might also have some other bindings based on built in
procedures,for example,we may have installed a primitive
procedure for plus*,.,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.7.17
..,and as we saw from our earlier work,plus* would be
represented by a primitive procedure,in this case as list
structure with a tag identifying it as a primitive and a pointer to
the actual primitive code,Remember that this was installed in
the global environment when we loaded up our evaluator,
Slide 15.7.18
That is,there is a pointer from plus* in the global
environment (that table) to the actual representation of the
object,
Slide 15.7.19
So what happens when we evaluate the expression shown at the
top? Eval will dispatch this to eval-define which
will create a binding for the symbol twice* in the
environment,to the value returned by evaluating (lambda*
(x*) (plus* x* x*)),We know what that latter
evaluation will return,that is the thing we just saw and it gives
us back this structure,
Slide 15.7.20
We do need to worry about where the environment part of the
structure points to,and we know it should point to the
environment in which the evaluation was done,hence to this
frame..,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.7.21
..,and the binding of the name twice* to this value will
take place in this environment,Thus we add a pairing to the
table of the symbol twice* and a pointer to this structure,
Slide 15.7.22
So notice what has happened,We have created the ability to
create names for procedures in our environment,that point to
one of these procedure objects,Thus we can now refer to that
procedure object by name,thus building in a level of
abstraction,
Slide 15.7.23
So let's look at the use of this abstraction,by looking at the
evaluation of an expression that involves a name for one of
these procedures,We want to see how eval and apply
work together to unwrap the abstraction to more primitive
expressions,
First,let's evaluate the expression shown with respect to the
global environment,
Slide 15.7.24
Eval will deduce that this is an application,thus reducing to
this apply expression,We must evaluate the first
subexpression (which we simply grab from the tree structure)
and apply the result to the list we get by mapping eval down
the list of arguments in the initial expression,Note that all of
these evaluations take place with respect to the same
environment,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.7.25
And what does the evaluation of the pieces do? Twice* is
just a symbol,so eval dispatches to a lookup,which returns
this list structure that captures the pieces of the procedure
(label,argument list,body and inherited environment),
Evaluating the arguments simply returns the list (4) since
numbers are self-evaluating,Thus we are ready to apply this
representation of a compound procedure to a list of argument
values,
Slide 15.7.26
And the code for apply indicates that we should grab (using
list manipulation procedures) the formal parameters of the
procedure,take the list of argument values supplied,and glue
them together with respect to the original environment in which
the procedure was created,to create a new frame,That is just
some list manipulation that we haven't detailed yet,
Conceptually,we know what this should do,create a new table
(or environment) with the formal parameters bound to the
argument list,and which is scoped by the environment in which
the procedure was created,
And then,we walk down the tree representation for the
procedure object (this version of a "double bubble"),grab off the body,and we will evaluate that new expression in
this new environment,Note that this is all just list manipulation,we have not evaluated any of the pieces,we are
simply setting up to do that,
Slide 15.7.27
So extending the environment will be some kind of abstract
data type manipulation that creates a new environment,call it
E1,in which those variables have been bound,We are now
going to evaluate this new expression with respect to that
environment,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.7.28
Conceptually,we have some new table structure,as shown
here,with a new table for the binding of x* and a pointer to
the enclosing environment (GE),We are now evaluating with
respect to this environment E1,
Notice how we have reduced evaluation of one expression with
respect to some environment to evaluation of a simpler
expression with respect to a new,extended environment,There
is a basic cycle of the environment model,and more
specifically,of our evaluator,
Slide 15.7.29
Before we carry on tracing through the evaluation of this
simpler expression with respect to this new environment,let's
stop and look at an important point,Suppose we started off the
overall evaluation of (twice* 4),not in the global
environment,but in some other environment,What would
happen?
The application would have to do an evaluation of the
subexpressions,and those would also be with respect to this
new environment since that is passed down as part of the
evaluation,
Slide 15.7.30
So at this stage we would use that same new environment,
Slide 15.7.31
But when we get to the actual application,notice that here the
environment doesn't change,because the global environment
used here comes from the creation of the procedure,When we
evaluated the lambda* expression that created the
procedure object,part of that creation as a link to the
environment in which the lambda* was evaluated,And that
means that evaluating the body is going to extend this
environment,not the environment we called it in,Thus,
although we can evaluate this expression in any environment,
the environment that gets inherited by the procedure is exactly
that one in which it was created,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.7.32
So now let's finish off tracing out the evaluation of our
expression,We now have a new environment,the one we just
created,In it there is a local binding of the name x* to the
value 4,and which is scoped by the global environment,
Slide 15.7.33
Within that environment we are evaluating the body of that
procedure,(plus* x* x*),Of course,the evaluator
doesn't know that this is a body of a procedure; it has simply
unwrapped previous evaluations to the stage of evaluating this
procedure with respect to this environment,We can now
quickly step through the evolution since it is much like earlier
examples,
This reduces to an application,Getting the first value is simply
a lookup of a binding,and the map operation will simply
lookup the bindings of the other expressions,Note that the first
binding will be found in GE,while the other bindings will be
found in E1,
This reduces to the application of a primitive procedure to a set of values,This then becomes an application of a
built in operation,which just reduces to the expected value,
Slide 15.7.34
The only thing left to be done is to actually implement the
environment model,that is,dealing with frames and
environments,Here,we get to make some design choices,Up
till now we have treated it as an abstract data type,but we can
make it much more concrete,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.7.35
For the purposes of this exposition,we can just choose to
represent an environment as a list of tables,We will use the
table abstraction,as before (which we know we can implement
in terms of list structure),and an environment will just be a list
of these tables,sequenced together,
Slide 15.7.36
So,for example,given this abstract representation for the
environment as a nested chain of tables starting with Frame A
as a table,scoped by the global environment which is also a
table,we can simply represent this much more concretely as,.,
Slide 15.7.37
..,this list structure,Now an environment points to a list,the
first element of which is the first frame,the second element of
which points to the remaining list of tables,which might be
arbitrarily long,Eventually,the list will terminate in the global
environment,which will contain the bindings for all the things
that are established in the installation of the original
environment,This is one way of representing an environment
chain,which clearly illustrates the chaining together of frames
in sequence,
Slide 15.7.38
Here is the code to actually implement environments,Most of
this is bookkeeping detail,using lists,
Creating the initial global environment is simply a matter of
starting with an empty environment,and then building a new
frame that creates pairings of names for built in procedures with
the actual representations of the primitives,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.7.39
What else do we still need? Lookup needs to be a bit more
explicit,To look up a value in an environment,we start by
asking if the environment is empty,If it is,we have a problem,
and we complaing that there cannot be a binding for this
variable,Otherwise we do what we did before,the only
difference is that we do the lookup with a table-get in
the first frame of the sequence,If there is no binding there,we
recursively move on to doing the lookup in the remainder of the
environment,Thus we are just walking down the list of frames,
looking in each one for a binding of the variable,
Slide 15.7.40
The only other change we need to make is to define*,We
know this should change the first frame of the environment
chain,Thus we should do a table-put! into the first
frame in this sequence,Note that this is just list manipulation to
create a proper sequence of frames constituting an environment,
6.001 Notes,Section 15.8
Slide 15.8.1
Thus,we have implemented our interpreter,
The key thing to note is the cycle that occurs with application of
a procedure,Eval and apply form the core of the
evaluator,Eval takes an expression and an environment and
reduces this to applying an operator to a set of values,
which in turn reduces to evaluating a new expression with
respect to a new environment,They continue this cycle,
unwinding the abstractions of the procedures,until it reduces
the expression to primitive components,
In particular,notice that there are no pending operations on
either call,Eval does not have any pending operations,and
is inherently iterative,Apply similarly has no pending operations,Thus,if we are evaluating an iterative
procedure,we will get an iterative behavior,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.8.2
So it appears that we have worked up to a fairly full-fledged
evaluator,What's is still missing from this interpreter? Only a
couple of things,right now we can't evaluated a sequence of
expressions,For example the bodies of our procedures expect
only one expression,It would be fairly easy to add such a
capability into our system,What else are we missing? We only
have a couple of data types right now,So for example we might
add in strings or other data types,
Slide 15.8.3
This then leads to the major punch line of this lecture,
Everything in this lecture would still work if we just removed
the *'s from the names! So what does that mean? It says that
what we have done is describe the process of evaluation and
interpretation in Scheme,We have really built a Scheme
interpreter,Not as complete as the full-blown one,but the
things missing are pretty minor,Literally everything we've done
would still work if we removed the *'s which says we can build
a Scheme interpreter,We happen to have done it on top of
Scheme,but we could have done it on top of a more primitive
language,or we could have done it in some other language,and
this is the central point,By defining an evaluator or an interpreter we are defining the language,We are
specifying what it means to evaluate expressions in this language,what is the legal syntax and the legal semantics,
and once we have done that we have created a language that we can then use to solve other problems,This point
we will return to,but this is the "take home" message of this lecture,
6.001 Notes,Section 15.6
Slide 15.6.1
The next stage in the evolution of our evaluator is to pull the
environment out as an explicit parameter,Up until now we
could rely on just having a single environment in which to store
bindings for variables,It made sense to have a global
environment in which to hold names for procedures and global
variables,We are about to start moving towards application,in
fact we built a simple version of application but we want to
extend that to procedures that need local environments,Thus
let's pull out the environment explicitly,So we will have to
change our evaluator from simply evaluating expressions to
evaluating expressions with respect to some environment,
What changes to we need? All procedures must now call
eval with that extra argument of an environment,Lookup and define* are going to use that
environment argument to specify where to work,
Slide 15.6.2
This is actually a really boring change! The functionality of
this evaluator is exactly the same as the previous version,All
we do is extend eval to take an expression and an
environment as arguments,and then make sure that all the
dispatch functions also take environments as arguments,Other
than make those changes,everything else is exactly as before,
Of course now when we evaluate an expression we have to
specify what environment,and we will need an initial or global
environment to start with,This will contain bindings for built in
procedures,
Otherwise this is just bookkeeping,It will allow us to extend
our evaluator in a more general way,
Slide 15.6.3
Notice in fact that the only non-trivial case is that dealing with
application inside of eval,Remember that this is the dispatch
that says that we have to apply the object that represents the
procedure to its arguments,Notice the change we need,As
before,we will evaluate the first subexpression of the tree
structure to get the procedure,now including the environment
as an argument,
In order to get the values of the other expressions before we
could just map eval down the list of expressions,Here,
however,we have to evaluate each expression with respect to
the environment so we need a lambda here to capture that
procedure,
We will come back to this shortly,but you can see from a Scheme perspective this makes sense,We are mapping a
procedure that evaluates each subexpression with respect to the environment passed in,This is the only non-trivial
extension in this case,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
6.001 Notes,Section 15.7
Slide 15.7.1
Now we are ready to make our last big extension to our
interpreter,We would like to be able to add new procedures,If
you think about it,once we add this ability we have all the basic
components of an interpreter,we can deal with primitive
expressions,with special forms,and with applications,and as
along as we can create our own procedures,those applications
could include arbitrary nesting of expressions,
How do we add new procedures? We would like to create a
name for something we make with the equivalent of a
lambda,We want to capture procedures within lambda's
and then use that procedure as an application,
Slide 15.7.2
To do this,we need to add to our evaluator the ability to capture
a computation in a procedure,Earlier in the term we saw this
with Scheme,that a lambda expression could capture a
common pattern of computation,Now we are going to see how
to add that ability to an evaluator,by creating something that
will deal with lambda-like expressions,
Here is our strategy for accomplishing this,First,we will need
another dispatch in our evaluator,something that deals
explicitly with lambda* expressions,We would like the
value of the lambda* expression to return a compound
procedure,however we decide to define it,Second,given the
ability to create our own procedures,we will need to extend apply to handle such procedures,Right now
apply is just set up to handle built in things we inherit from Scheme,but we need to have it also handle
procedures we create,And then we should go back and finish the environment model,since it will become
essential in the application of procedures,
Slide 15.7.3
Adding the dispatch to eval is straightforward,We'll need a
new tag checker for lambda* expressions,and a new
dispatch clause inside eval to a procedure to handle the
create of lambda* objects,Notice where we add this
dispatch clause,it must come before the application case,since
lambda* is a special form,It should really also come after
the checks for the primitive expressions (numbers or symbols),
The ordering of where within the special forms it falls is not
crucial,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.7.4
So what should eval-lambda do? Remember that
eval takes as input a tree structure representing a
lambda* expression,which it then passes to this procedure,
together with a pointer to the environment,So first,we need to
walk down that tree structure,and grab off the arguments of the
expression,the formal parameters if you like,That's what the
cadr does,And we want to walk further along the tree and
grab the body of the expression,that is the caddr,Having
pulled off those two pieces,without evaluation,we then want to
glue together some representation of a procedure as a set of
arguments,a body,and an environment,This should sound familiar,as it sounds a lot like what we have when we
do evaluation with our "double bubble" notation in the environment model,There we have a similar representation
for a procedure,formal parameters,body and environment,Make-compound is doing a similar thing here,
Note that there is no call to eval inside eval-lambda,None of these expressions is being evaluated; we
are simply manipulating tree structure,
Also notice that we have made a decision about the body of a procedure,By using the caddr of the expression,
we are assuming that the body contains exactly one expression,That is a design choice that we could change,
Slide 15.7.5
So adding the new form to create procedure objects is pretty
straightforward,We will have to eventually work out what
make-compound does,but that is basically an
abstraction issue,
Now the question is,how do I change apply? Remember
that apply is going to evaluate the first subexpression of a
tree structure,get out the appropriate procedure,evaluate all the
arguments getting back a list of those values,and then we want
to apply that procedure to that list,
In the case of primitive things,inherited things from Scheme
like +,* etc.,apply was just a matter of doing the basic
operation,Now,we need to extend apply,
Slide 15.7.6
In this case,when we go to apply an operator to a list of
operands,we will need to check for type,Is it a primitive
operation? If yes,we just apply the underlying Scheme
procedure as before,Is it a compound operator,that is,are we
evaluating an expression in which we are using a procedure that
we created in our evaluator using lambda*? In this case we
need to implement the idea of the environment model,
Thus,we get the body of the operator,that is,we pull out the
piece that we glued together when we made the procedure
object,Now we want to evaluate that body,But we are going to
evaluate that with respect to an environment,created by taking
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
the operands passed in,taking the formal parameters of the procedure object,and binding them together,This new
environment should extend the previous one,in case we need things from there,
Thus when we have a compound application,we will grab the body of the procedure and evaluate it in a new
environment,which has access to everything in the previous environment,plus bindings for the formal parameters
of the procedure to the values passed in,
Note what this does,Given this version of apply we have a wonderfully cyclic structure,Eval of an
expression with respect to an environment in the general case will reduce to applying a procedure to a set of
arguments,and that will reduce in the general case to evaluating a new expression (the body of the operator)
with respect to a new environment,In this way,eval and apply will work hand-in-hand in terms of
unwrapping the abstractions of the procedures down to primitive operations on primitive objects,
Slide 15.7.7
All that is left to do is to implement the "double bubble" idea as
an abstract data type,We can do that by creating a tag for a
compound procedure and having the constructor (make
compound) simply glue things together in a list,with the
appropriate predicate and selectors associated with this data
abstraction,This is not the only way to do this; all we need is
some method for gluing things together and getting them back
apart,
Slide 15.7.8
Looking at the code helps us understand our changes,but it is
probably easier to see the changes by watching the evolution of
the evaluation of an expression,Let's trace through the
evaluation of a lambda* expression,That is,we are
evaluating a tree structure whose first element is the symbol
lambda*,whose second element is a list of parameters,and
whose third element is an expression constituting the body of
the procedure,We are evaluating this with respect to some
environment,say the global environment,
Slide 15.7.9
As before,please follow along in the code (for part 6),
So eval takes in that tree structure representing this
expression and checks it for different types of expressions
(number,symbol,etc.),Eventually eval reaches the case for
handling lambda* expressions,Thus,we will dispatch on
type to eval-lambda,the specific procedure designed to
evaluate lambda*'s,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.7.10
Eval-lambda starts with the same tree structure as an
argument,plus some pointer to the global environment (GE),
What does it do?
It walks down the tree structure of the first argument,and grab
the arguments (that's the cadr),Similarly,walk down the tree
structure and grab the body (that's the caddr) (note that we
are assuming there is only one expression in the body),In this
case,the cadr will return the list (x*) and the caddr
will return the list (plus* x* x*),Let me stress again
that we are simply getting these as tree structure; there is no
evaluation yet,Having acquired those two structures,we then
glue them together using make-compound,
Slide 15.7.11
So the evaluation reduces to this expression,This is simply
using our constructor to glue these pieces together,and in this
case that converts to this form...,
Slide 15.7.12
..,and thus we are down to a basic primitive,We are going to
construct a new list structure,with a type tag,a list of
parameters,a body and the environment in which this
lambda* expression was created,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.7.13
And that says that the evaluation of this expression returns this
list structure,That is actually quite nice! We have created a new
object in our language,we have evaluated a lambda* and
created a procedure object,When we go to do an application (as
we will see shortly) we need an expression represented as list
structure so that we can pull out the pieces,and here they are,
All of the parts we will need for an application are represented
as list structure within this list,so that we easily use those
within the evaluator,
Slide 15.7.14
Thus,this piece of list structure represents a procedure,It is our
particular way of creating a data structure for a procedure,it has
a tag identifying it as a compound procedure,a set of
parameters,and a body,all represented in the format we expect
as a parsed tree,
Slide 15.7.15
Now that we can create procedures in our interpreter,we
naturally want to give them names,and that just uses our
define* expressions,How does this work? First of all,we
will have some initial bindings in our environment,.,
Slide 15.7.16
..,as shown here in our abstract table representation for an
environment,This environment includes prior bindings for
names to numbers and for symbols to Boolean values,We
might also have some other bindings based on built in
procedures,for example,we may have installed a primitive
procedure for plus*,.,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.7.17
..,and as we saw from our earlier work,plus* would be
represented by a primitive procedure,in this case as list
structure with a tag identifying it as a primitive and a pointer to
the actual primitive code,Remember that this was installed in
the global environment when we loaded up our evaluator,
Slide 15.7.18
That is,there is a pointer from plus* in the global
environment (that table) to the actual representation of the
object,
Slide 15.7.19
So what happens when we evaluate the expression shown at the
top? Eval will dispatch this to eval-define which
will create a binding for the symbol twice* in the
environment,to the value returned by evaluating (lambda*
(x*) (plus* x* x*)),We know what that latter
evaluation will return,that is the thing we just saw and it gives
us back this structure,
Slide 15.7.20
We do need to worry about where the environment part of the
structure points to,and we know it should point to the
environment in which the evaluation was done,hence to this
frame..,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.7.21
..,and the binding of the name twice* to this value will
take place in this environment,Thus we add a pairing to the
table of the symbol twice* and a pointer to this structure,
Slide 15.7.22
So notice what has happened,We have created the ability to
create names for procedures in our environment,that point to
one of these procedure objects,Thus we can now refer to that
procedure object by name,thus building in a level of
abstraction,
Slide 15.7.23
So let's look at the use of this abstraction,by looking at the
evaluation of an expression that involves a name for one of
these procedures,We want to see how eval and apply
work together to unwrap the abstraction to more primitive
expressions,
First,let's evaluate the expression shown with respect to the
global environment,
Slide 15.7.24
Eval will deduce that this is an application,thus reducing to
this apply expression,We must evaluate the first
subexpression (which we simply grab from the tree structure)
and apply the result to the list we get by mapping eval down
the list of arguments in the initial expression,Note that all of
these evaluations take place with respect to the same
environment,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.7.25
And what does the evaluation of the pieces do? Twice* is
just a symbol,so eval dispatches to a lookup,which returns
this list structure that captures the pieces of the procedure
(label,argument list,body and inherited environment),
Evaluating the arguments simply returns the list (4) since
numbers are self-evaluating,Thus we are ready to apply this
representation of a compound procedure to a list of argument
values,
Slide 15.7.26
And the code for apply indicates that we should grab (using
list manipulation procedures) the formal parameters of the
procedure,take the list of argument values supplied,and glue
them together with respect to the original environment in which
the procedure was created,to create a new frame,That is just
some list manipulation that we haven't detailed yet,
Conceptually,we know what this should do,create a new table
(or environment) with the formal parameters bound to the
argument list,and which is scoped by the environment in which
the procedure was created,
And then,we walk down the tree representation for the
procedure object (this version of a "double bubble"),grab off the body,and we will evaluate that new expression in
this new environment,Note that this is all just list manipulation,we have not evaluated any of the pieces,we are
simply setting up to do that,
Slide 15.7.27
So extending the environment will be some kind of abstract
data type manipulation that creates a new environment,call it
E1,in which those variables have been bound,We are now
going to evaluate this new expression with respect to that
environment,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.7.28
Conceptually,we have some new table structure,as shown
here,with a new table for the binding of x* and a pointer to
the enclosing environment (GE),We are now evaluating with
respect to this environment E1,
Notice how we have reduced evaluation of one expression with
respect to some environment to evaluation of a simpler
expression with respect to a new,extended environment,There
is a basic cycle of the environment model,and more
specifically,of our evaluator,
Slide 15.7.29
Before we carry on tracing through the evaluation of this
simpler expression with respect to this new environment,let's
stop and look at an important point,Suppose we started off the
overall evaluation of (twice* 4),not in the global
environment,but in some other environment,What would
happen?
The application would have to do an evaluation of the
subexpressions,and those would also be with respect to this
new environment since that is passed down as part of the
evaluation,
Slide 15.7.30
So at this stage we would use that same new environment,
Slide 15.7.31
But when we get to the actual application,notice that here the
environment doesn't change,because the global environment
used here comes from the creation of the procedure,When we
evaluated the lambda* expression that created the
procedure object,part of that creation as a link to the
environment in which the lambda* was evaluated,And that
means that evaluating the body is going to extend this
environment,not the environment we called it in,Thus,
although we can evaluate this expression in any environment,
the environment that gets inherited by the procedure is exactly
that one in which it was created,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.7.32
So now let's finish off tracing out the evaluation of our
expression,We now have a new environment,the one we just
created,In it there is a local binding of the name x* to the
value 4,and which is scoped by the global environment,
Slide 15.7.33
Within that environment we are evaluating the body of that
procedure,(plus* x* x*),Of course,the evaluator
doesn't know that this is a body of a procedure; it has simply
unwrapped previous evaluations to the stage of evaluating this
procedure with respect to this environment,We can now
quickly step through the evolution since it is much like earlier
examples,
This reduces to an application,Getting the first value is simply
a lookup of a binding,and the map operation will simply
lookup the bindings of the other expressions,Note that the first
binding will be found in GE,while the other bindings will be
found in E1,
This reduces to the application of a primitive procedure to a set of values,This then becomes an application of a
built in operation,which just reduces to the expected value,
Slide 15.7.34
The only thing left to be done is to actually implement the
environment model,that is,dealing with frames and
environments,Here,we get to make some design choices,Up
till now we have treated it as an abstract data type,but we can
make it much more concrete,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.7.35
For the purposes of this exposition,we can just choose to
represent an environment as a list of tables,We will use the
table abstraction,as before (which we know we can implement
in terms of list structure),and an environment will just be a list
of these tables,sequenced together,
Slide 15.7.36
So,for example,given this abstract representation for the
environment as a nested chain of tables starting with Frame A
as a table,scoped by the global environment which is also a
table,we can simply represent this much more concretely as,.,
Slide 15.7.37
..,this list structure,Now an environment points to a list,the
first element of which is the first frame,the second element of
which points to the remaining list of tables,which might be
arbitrarily long,Eventually,the list will terminate in the global
environment,which will contain the bindings for all the things
that are established in the installation of the original
environment,This is one way of representing an environment
chain,which clearly illustrates the chaining together of frames
in sequence,
Slide 15.7.38
Here is the code to actually implement environments,Most of
this is bookkeeping detail,using lists,
Creating the initial global environment is simply a matter of
starting with an empty environment,and then building a new
frame that creates pairings of names for built in procedures with
the actual representations of the primitives,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.7.39
What else do we still need? Lookup needs to be a bit more
explicit,To look up a value in an environment,we start by
asking if the environment is empty,If it is,we have a problem,
and we complaing that there cannot be a binding for this
variable,Otherwise we do what we did before,the only
difference is that we do the lookup with a table-get in
the first frame of the sequence,If there is no binding there,we
recursively move on to doing the lookup in the remainder of the
environment,Thus we are just walking down the list of frames,
looking in each one for a binding of the variable,
Slide 15.7.40
The only other change we need to make is to define*,We
know this should change the first frame of the environment
chain,Thus we should do a table-put! into the first
frame in this sequence,Note that this is just list manipulation to
create a proper sequence of frames constituting an environment,
6.001 Notes,Section 15.8
Slide 15.8.1
Thus,we have implemented our interpreter,
The key thing to note is the cycle that occurs with application of
a procedure,Eval and apply form the core of the
evaluator,Eval takes an expression and an environment and
reduces this to applying an operator to a set of values,
which in turn reduces to evaluating a new expression with
respect to a new environment,They continue this cycle,
unwinding the abstractions of the procedures,until it reduces
the expression to primitive components,
In particular,notice that there are no pending operations on
either call,Eval does not have any pending operations,and
is inherently iterative,Apply similarly has no pending operations,Thus,if we are evaluating an iterative
procedure,we will get an iterative behavior,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.8.2
So it appears that we have worked up to a fairly full-fledged
evaluator,What's is still missing from this interpreter? Only a
couple of things,right now we can't evaluated a sequence of
expressions,For example the bodies of our procedures expect
only one expression,It would be fairly easy to add such a
capability into our system,What else are we missing? We only
have a couple of data types right now,So for example we might
add in strings or other data types,
Slide 15.8.3
This then leads to the major punch line of this lecture,
Everything in this lecture would still work if we just removed
the *'s from the names! So what does that mean? It says that
what we have done is describe the process of evaluation and
interpretation in Scheme,We have really built a Scheme
interpreter,Not as complete as the full-blown one,but the
things missing are pretty minor,Literally everything we've done
would still work if we removed the *'s which says we can build
a Scheme interpreter,We happen to have done it on top of
Scheme,but we could have done it on top of a more primitive
language,or we could have done it in some other language,and
this is the central point,By defining an evaluator or an interpreter we are defining the language,We are
specifying what it means to evaluate expressions in this language,what is the legal syntax and the legal semantics,
and once we have done that we have created a language that we can then use to solve other problems,This point
we will return to,but this is the "take home" message of this lecture,
