6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
6.001 Notes,Section 15.1
Slide 15.1.1
Our goal over the next few lectures is to build an interpreter,
which in a very basic sense is the ultimate in programming,
since doing so will allow us to define our language,
This is a somewhat surprising statement,But,in fact,as we will
see through these lectures,it really is correct,The reason it is
correct is the following,
Every expression we write in our language has a meaning
associated with it,Deducing the meaning associated with an
expression is the process of evaluation,And therefore,if we
can define the program that executes that deduction for us,our
definition of that program (i.e,our definition of the interpreter)
provides for us the exact specification of what's legal in our
language and how to deduce meanings of expressions within our language,
Our goal is to work through this understanding in stages,We will explore a series of examples,initially some very
simple examples,but culminating in a full-scale Scheme evaluator,We are going work our way up to this,letting
you see how to add the different pieces into an interpreter,
Before you proceed,however,there is a code handout that goes with this lecture,I suggest that you stop,go back to
the web page,and print out a copy of the code to have next to you as we go through the lecture,
Slide 15.1.2
First,let's set the stage for why we want to do this,why do we
want to build an interpreter? Why do we need an interpreter?
Think about what we have seen so far in this course,We have
spent a lot of time talking about,and using,the ideas of
abstraction,both procedural and data,We know that this is a
powerful idea,as it lets us bury details,and it certainly supports
the expression of ideas at an appropriate level,Said another
way,we can think about what we want to do and separate that
from the details of how to actually make it happen,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.1.3
But eventually,we need a process for unwinding all of those
abstractions,to get the value that corresponds to an expression’s
meaning,That means we need to implement semantics of
interpreting expressions,
Slide 15.1.4
Notice the words I used,I said,"we need a process for
unwinding those abstractions",If we can have such a process,
then we should be able to describe it in a procedure,that is our
language for describing processes,
In fact,you have already seen a version of this description,just
not as a procedure,What was the description?,.,the
environment model!
If you think about it,that makes sense,The environment model
just described the process for how to determine a meaning
associated with an expression,which in turn just unwrapped the
abstractions down to the primitive operations,Today,we want
to talk about how to actually build an evaluator as a procedure
rather than as that abstract environment model,
Slide 15.1.5
First,what are the stages of an interpreter? For the kind of
languages we are considering,Scheme or Lisp like languages,
typically there are five stages,There is a lexical analyzer,a
parser,an evaluator that typically works hand-in-hand with an
environment,and there is a printer,
Let's look at what goes on in each of these at a high level,
before we build them,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.1.6
To do that,let's talk about the input and output characteristics
of each of them,
By focusing on the input to each successive stage in the
interpretation,we can get a sense of what should happen at each
stage,and thus get a sense of how to build an interpreter,
Slide 15.1.7
The initial input is a string of characters,which represents the
typewritten version of the expression we want to evaluate,This
is exactly the thing that we would type in at a terminal if we
wanted to have an expression evaluated,So the initial input is a
string of characters,
Slide 15.1.8
The first step is to use a lexical analyzer to convert that string
of characters into units or words,This is shown here,where the
string gets converted into a set of words or isolated characters
like "(" and ")" and "+" and numbers,Thus the input to the next
stage is an ordered sequence of these units or words,
Slide 15.1.9
The second stage then parses those words into a structure that
we can use for evaluation,In particular,we convert the linear
sequence of words into a tree structure,We are using pairs here
for convenience but that is not required,We could use any other
representation of trees as well,
As we do this,we are going convert the self-evaluating
expressions into their internal values,So notice the form we get
for the next stage,it's a tree structure,and hanging off of the
leaves of the tree are numbers,symbols,or other objects that
are represented as basic words,This is the input to the next
stage,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.1.10
Now comes the heart of the interpreter,We want to take that
tree structure and deduce its value,First,notice the form of tree
structure,We will talk about this in detail later,but you can
already see how the parser has converted things into a tree,
Every time we see an "(",indicating the beginning of a new
combination,we have created a new list structure,If the parser
is already inside a list structure,it drops down a level,so that
we build up a tree where each horizontal slice through the tree
corresponds to some combination,Now what is the evaluator
supposed to do? It wants to take that tree structure,plus an
environment,and interpret it,And what does that mean? Think
of the environment as a way of associating names with more primitive values,It acts much like a dictionary,The
evaluator will use a set of rules to walk through this tree,looking up values associated with symbols from the
environment,i.e,from that dictionary,and then using the rules to reduce complex expressions to simpler things,
culminating in some simple value that we will return,
Slide 15.1.11
That value becomes input to the final stage,The printer simply
converts things back into the appropriate form for display on
the monitor,and then,.,
Slide 15.1.12
..,that just gets displayed to the user,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.1.13
So here is a summary of that process in words,
Slide 15.1.14
Our goal is to implement an interpreter,Actually,that's not
quite right,Our goal is for you to understand what goes into an
interpreter,which we will explore by implementing one,Since
the key part of an interpreter,the crucial part,is the evaluator,
we are going to concentrate almost exclusively on that,We are
going to use Scheme for all the rest of the pieces,that is,we
will use Scheme's lexical analyzer and parser,and Scheme's
printer,rather than building them from scratch,This means of
course that we will need to create an evaluator for a language
that looks a lot like Scheme in the sense of having a tree
structure as the output of its parser and a set of rules for
manipulating that tree structure,as the way of dealing with the actual evaluation,It also says that the syntax of the
language we are going to use to demonstrate an interpreter will need to look at lot like Scheme,in terms of things
like using parentheses to delimit expressions and other related issues,
We say this because we don't want you to get confused between what is going in Scheme and the general ideas of
building an evaluator and interpreter,Our goal is to build an interpreter,especially the evaluator part,and let you
see how that occurs and use that to explore the idea of how these things implement the rules for a language,We are
going to build our own simple language and its evaluator,For convenience,we are going to call this language,
Scheme*,It has a lot of the characteristics of Scheme,but we will use the convention that a * will be placed at the
end of every expression in our language,to distinguish it from the corresponding Scheme expression,
We'll start with a simple evaluator and work our way up,The first simple evaluator will be one that handles simple
arithmetic expressions,
6.001 Notes,Section 15.2
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.2.1
Our plan is to start by building an evaluator that handles
arithmetic expressions,and in fact we will restrict ourselves just
arithmetic expressions of two or fewer arguments,We would
like to be able to evaluate things like the example shown on the
slide,adding 24 to whatever we get by adding 5 and 6,
Notice the * at the end of the symbol plus to indicate that
this is something that we will build within our language,
Slide 15.2.2
And here is some code that captures how we will evaluate
expressions of this form,This is identical to the code listed in
the separate code handout,and I suggest you have that page
handy as we go through this development,
Slide 15.2.3
Notice what we are doing here,We are using our knowledge of
Scheme to describe the process of evaluating expressions in this
new language,We are writing,in Scheme,the description of
that process,
Okay,what do we need? We have a procedure for evaluating
expressions in our new language,called eval,Notice its
form,It has a way of dealing with the base case,which is an
expression that just consists of a number,And to do that it uses
type checking,
Then,we have a way of dealing with the compound case,Here,
it uses type checking to see if we have a sum and notice how
this works,It uses the keyword of the expression to determine the type of that expression,If the expression is a
sum then we will just add,using the primitive operation of addition,the values of the subexpressions,But a key
point arises here! To get those values we need to evaluate each subexpression as well,since we don't know at this
stage if they are just numbers or are themselves compound expressions,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.2.4
Let's look at this in more detail,First,let's look at the input to
this evaluation process,Remember that our expression,which
we typed in,is converted into list structure,as a tree of symbols
and numbers,It looks like what is shown on the slide,and this
is what gets handed to the evaluator as a representation of our
example expression,So let's treat this as if this exact tree
structure were passed in to eval,
Slide 15.2.5
And what does eval do with this input? Check the code on
the handout,Eval grabs the list and tests its tag,That means
it first checks to see if this whole thing is a number,Since it is
not,it takes the first element of this list structure and checks to
see if it is the special symbol plus*,
Slide 15.2.6
Having done that it dispatches on type to the right procedure to
handle this kind of expression,Having determine that it is a
sum by checking the tag,it sends it off to eval-sum,and
this is (for now at least) just a normal procedure application,
We apply the procedure to the expression,
Slide 15.2.7
So now eval has reduced this to applying eval-sum to
the tree structure shown,Notice what the body of eval
sum does,It walks down the tree,grabbing out the two
subexpressions,that is the first and second components of this
sum,Eval-sum then converts this into adding,using the
built-in primitive,whatever I get by evaluating the first
subexpression and whatever I get by evaluating the second
subexpression,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.2.8
That leads to this form,Notice what we have done,we have
reduced evaluation of one expression into some simple
primitive operations on evaluation of other subexpressions,And
what has that done? It has literally just walked down the tree
structure that represents this expression,pulling out the right
pieces,It has used the tag to tell us who to send the expression
to,and then it has simply grabbed cars and cdrs of the list
structure,and handed them off to new evaluations,Now at this
stage,evaluating the first subexpression (eval 24) is
easy,We see from our code that it will use type checking to
determine this is a number and simply return that expression,
And that’s nice! This is just pointing to the number 24 so the
number 24 gets returned,What about the other piece?
Slide 15.2.9
Well this just looks like the kind of expression we started with,
We are evaluating some list structure that happens to represent
a sum,It's got a tag at the front to say it is one of these
plus* expressions,so we can do exactly the same thing,
The evaluation of this expression will unwrap into an eval
sum of the same list structure,and that will reduce to a
primitive application of + to whatever I get by evaluating the
subexpressions,and that I get by walking down the tree,
grabbing the right pieces,applying eval and getting back the
numbers,
Slide 15.2.10
And now we see that we have unwrapped this abstraction down
to some primitive operations,primitive application of addition
to some simple expressions,in this case just numbers,And of
course this will finally reduce to the answer we expect,
However,a key thing to note is how this simple evaluator has
taken in a tree structure representing an expression and has
unwrapped it into successive evaluations until it reduces to a set
of applications of primitive built-in procedures to primitive
values,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.2.11
Since this is important,these stages of eval unwrapping into
simpler and simpler things,and the dispatching on type to the
correct procedure,let's look at this one more time,In this case,
let's focus on how eval unwinds the abstraction,and what
values are returned at each stage of the evaluation,As before,
you may find it convenient to have a copy of the code in front
of you as we go through this examination,
Slide 15.2.12
So we start with eval of this full expression,We've put a '
in front of the expression to show that we want list structure
equivalent to this expression,
Thus we start with an eval of this expression,
Slide 15.2.13
Eval first checks the type of this expression,deduces that it
is not a number,but is a sum (because of the type tag),so this
expression gets dispatched to eval-sum,Eval sends the
expression to the procedure that is exactly set up to deal with
this particular form of list structure,
Slide 15.2.14
Now,eval-sum says,"go ahead and add whatever I get by
evaling each of the pieces",We haven't actually specified
in what order to do the subpieces,but for convenience assume
that it is done from left to right,So we now need to trace down
the tree,and get (eval 24),
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.2.15
Eval once again checks the type of this expression,deduces
it is a number,and just returns that expression,literally a
pointer to that thing which is an internal representation for the
number 24,
Slide 15.2.16
Next,eval has to evaluate the second subexpression,so this
is the eval of the expression shown at top right,As before,
we are going to dispatch on type,i.e,check to see what kind of
"beast" this is,deduce that it is a "sum" and therefore pass this
on to the right procedure to handle sums,
Slide 15.2.17
Once more,eval-sum will reduce to applying the addition
operation to whatever it gets by evaluating the subpieces,Thus,
we need to extract the subexpressions and once again apply
eval to them,Notice the nice recursive unwinding that is
going on here,
Slide 15.2.18
Well this just unwraps one more time,Again,we will apply +
to whatever we get by evaluating the two pieces,and eval in
both cases just dispatches on type,determines the expression is
a number and returns the expression as the value,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.2.19
Notice where we are at this stage,We have unwrapped this
compound expression into a nested sequence of operations of
primitive things to primitive values,At this stage we can gather
up the things we have left to do,We have some deferred
operations,for example in the topmost eval-sum we can
now add 5 and 6 to get 11,and so on,reducing all the deferred
primitive operations down to a single value,
Slide 15.2.20
Thus,we have built a simple evaluator that handles sums of no
more than two arguments,Here are some key points to notice
from this exercise,since our goal is to understand the process of
evaluation,
First,eval does type tag checking,It dispatches based on
type,much like we saw earlier in the course,
Second,numbers are just numbers,so there is nothing really to
do,
Third,complex expressions nicely get recursively evaluated in
pieces,Eval unwraps a complex expression into an
evaluation of the simpler parts,plus a deferred operation to
gather the values back together,Numbers just get handled as numbers,And eventually we reduce this whole thing
down to a set of primitive operations on primitive values,
6.001 Notes,Section 15.3
Slide 15.3.1
Okay,now let's build on this basic system,Suppose we want to
give names to things,For example,suppose we want to have
the behavior shown here,in which we can store intermediate
results as named values,and then just use those names
anywhere that we would want to use the actual expression and
its resulting value,This is the kind of behavior we saw earlier in
the term in Scheme,how would we add that behavior to the
evaluator we are building for simple arithmetic expressions?
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.3.2
Of course,the first thing we realize is that this means we will
need a way of storing bindings between names and values,We
can certainly imagine getting the value for something,but now
we have to have a way of storing the name and value together,
Define* has to have some way of gluing pieces together,
So we need to add this capability to our evaluator,
Slide 15.3.3
This is now the second page of your code handout,Don't be
intimidated by this code,as we have highlighted the things we
have changed from the first evaluator,shown here in bold face
font,
So what have we added? First,we need another type checker,
something that checks whether the expression is a define*
expression,here called define?,We have also added two
new pieces to the evaluator,
Slide 15.3.4
Note that in this version of eval we have a way of creating
names for values,and a way of getting back the value
associated with a name,Thus,we have two new dispatches in
our evaluator,something that checks to see if the expression is
a symbol,in which case we will lookup its value; and
something that checks to see if the expression is a definition
(checked using the special tag define*),in which case we
dispatch to something the evaluates these special expressions,
Slide 15.3.5
Before we look at the procedures that will handle lookup
and eval-define,let's first think about what we need,
We will need a way of gluing things together,and we know
how to do that,Let’s just assume a data abstraction,called a
table,It has a constructor,make-table,It has a way of
getting things out of a table,thus given a table and symbol,
table-get gives us back either a nil to indicate no
binding for that symbol was present,or the actual binding,We
have a way of putting things into the table,table-put!,
and we have binding-value which when given a binding,
returns the value part of that pairing,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
What is the point of this? We can simply assume that this table abstraction exists,and then we can build an
environment,Let's define environment in our underlying Scheme to be a table,and then we can create
procedures for lookup and eval-define,
Slide 15.3.6
Looking up the value of a symbol is simply a matter of
manipulating the table,We find the binding in the table for this
symbol,and then return that value part of the binding,This is
simply manipulating the environment and we can abstract that
away,
Slide 15.3.7
The real thing we have added is a way of dealing with a new
kind of expression,something built by define* that is
creating a binding of a name and a value,So what should this
do?
This procedure says to walk the tree structure to get out the
name (remember this is just walking the tree structure,there is
no evaluation going on here),Then,evaluate the expression
that will provide the value of the binding,Notice the use of
eval which recursively returns to the top level,and evaluates
this expression using the same rules! Once I get a value,I stick
it into the table that represents the environment,paired with the
name in a binding,
Slide 15.3.8
Since we have added a new component to our evaluator,let's
again look at what happens if we evaluate these two expressions
in our evaluator,especially watching to see what is returned
each time we recursively call eval in this process,So keep
track of the argument and return values of eval as we trace
through this process,using the code handout to keep track of
this process,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.3.9
So let's use our extended evaluator,by evaluating this
define* expression,Remember that this expression is just
represented as tree structure,the ' is just used here to remind
that the parser will have reduced this input expression to the
equivalent tree structure,So what does eval do with this
expression? It is a dispatch,so it will deduce that this is a
define* expression,and will thus send the expression to a
handler,That procedure will decide to eval the value
subexpression,
Slide 15.3.10
Well that is a lot like what we saw before! We are now doing an
eval on this list structure,And we have literally walked
down the list structure for the first expression to grab off this
piece and now evaluate it,This just recursively determines the
type of expression (a sum) and thus passes it off to eval
sum which evaluates each of the subexpressions in turn,then
applies the primitive + operation to return a value (9),So the
value returned by evaluating this subexpression is just 9,
Slide 15.3.11
And then what does eval-define say to do (remember
this is where we were when we went off to evaluate the
subexpression),It simply takes the name (or symbol) x*,
grabbing that off of the tree structure,it takes the returned
value,and it binds them together in the table somewhere,
Slide 15.3.12
And having done its work,it just returns the symbol
undefined to tell us it has completed the evaluation of
this definition,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.3.13
So now let's evaluate an expression that includes one of these
names,such as the expression shown,As before,this expression
is converted to tree structure,which is then evaluated,The tag
checking dispatches this to eval-sum,That procedure says
to add together (using +) whatever I get by recursively
evaluating the subpieces,
First is (eval x*),Notice in line 9 of evaluator that
because this part of the tree structure is a symbol,we do a
lookup,That is,we use the table abstraction to get the
binding associated with this symbol,and returns the value
associated with that binding in the table,in this case 9,The rest
of the evaluation proceeds as in previous cases,
Slide 15.3.14
So in fact this was fairly straightforward,As long as we have
the abstraction of the table,we see how we can add bindings to
it,through definitions,Notice also how just walking the tree
structure of the expression allows us to grab off symbols
without evaluating them,in this case,getting x* so that we
could bind it together with a value in the table,Other than that,
all the other things still hold for our evaluator,we can simply
now give names to values,
Slide 15.3.15
Having now extended our evaluator by adding in this new
capability,let's step back and extract some key messages,First
of all,notice that we added two new dispatch clauses to our
evaluator,One of them just dispatches to eval-define
using standard type checking of tags,The other clause,
however,relies on using symbol?,the underlying Scheme
function,to check for a name,Why is this reasonable?
Note that we are relying on the underlying Scheme reader to
convert things into a parse tree,and in particular,that reader
will convert sequences of characters like x * into symbols in
the parse tree,which then gets passed on to eval,Thus,
eval will get a structure that can be handled by symbol?,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.3.16
Note the second thing we added,a table,Or rather,we added an
abstract data type of a table to represent our environments,Note
that we could have used any implementation of a table,We are
not relying in any way on the specifics on a table
implementation,and that is exactly the point,As long as we
have a way of creating tables,we can build an evaluator that
simply manipulates them,
Slide 15.3.17
A key thing to note as well is that we have now added a special
form to our little evaluator! If you think carefully about it,you
will note that prior to this we were simply evaluating normal
combinations,which for this language were sums of two
arguments,However,eval-define does something
different,It recursively evaluates only the second
subexpression! In other words,eval-define takes the
first subtree of the parse tree passed in,without evaluation,and
treats it as a symbol,It only does evaluation on the second
subtree,
6.001 Notes,Section 15.4
Slide 15.4.1
What else can we add to our evaluator? To this point,we have
no way of making decisions,We can't branch to do different
things depending on a value,So let's extend our evaluator to
handle this,i.e,let's extend our calculator to handle conditions
and ifs,Statements such as the example shown,in which if
something is true we want to do one thing,otherwise we want
to do something else,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.4.2
We can already see from our example that we are going to need
to add two kinds of things,First,we will need procedures that
return Boolean values,Thus,greater* should evaluate a
pair of expressions,then return true if the value of the first
expression is larger than the value of the second expression,Of
course,we can easily imaging adding other such expression to
our language,so long as we have some expressions that
evaluate expressions and return Boolean values,
The second thing we will need is an operation that lets us
control the branching of our code,We know that this should be
a special form in which we should first evaluate the first
subexpression,check its value,and then based on that value
either evaluate and return the second subexpression or evaluate and return the third subexpression,This means we
need to add an if* expression to our language,
Slide 15.4.3
As before,there is a fair amount of code here,but we have
highlighted the changes from the earlier version,First,we need
two new ways of dispatching on greater* and if*
expressions,just using type checking as before,This means we
need ways of checking the type of such expressions,and we
need to add dispatch clauses to the cond expression inside
eval,one for each new type of expression,This is just like
before,
Slide 15.4.4
First,greater* expressions,Remember that exp will be
a pointer to some tree structure that has been created by the
parser,The first subexpression in that tree structure will be the
symbol greater*,The second subexpression will be some
other expression,as will the third subexpression,Our plan is to
get the values of the second and third subexpressions,We will
use tree operations to extract the right pieces,the recursively
evaluate them,Once we have the values of these
subexpressions,we can use the underlying Scheme primitive
operation to compare the numerical values and return a Boolean
answer,
This may seem a little odd; why not use Scheme's operation directly? Remember,our goal is to understand how to
connect the evaluator to built-in primitives,and to allow for alternative ways of executing operations,So remember
what we do,we walk the tree,grab the pieces,recursively evaluate them,then pass the pieces on to primitive
procedures to complete the reduction to a value,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.4.5
What about if* expressions? What happens when we
dispatch to eval-if?
Once more,remember that what is passed in to this procedure is
tree structure,created by the parser,This tree structure will
several pieces,the tag if*; then tree structure representing the
predicate of the expression; then tree structure representing the
consequent of the expression; then tree structure representing
the alternative of the expression,Notice how in our procedure
we use list manipulation to pull off each piece of the tree,and
then we evaluate the first one,
Once we have the value of test,we can then use the
underlying primitive mechanism for branching,cond,to decide that if test is true,then and only then,
take the consequent and evaluate it,If test is false,then and only then,take the alternative and evaluate it,
Notice the order in which things are done,This is a special form in which we first evaluate the predicate,
Depending on that value,we either evaluate the consequent or the alternative,and return that value as the value of
the overall expression,
Slide 15.4.6
So having added these two new expressions to our evaluator,
let's look at some examples of evaluation using them,as shown
on the slide,tracing through the process of recursively using
eval,
Slide 15.4.7
We can assume that we have done the evaluation of
(define* y* 9),since we know what that does,Now
let's look at evaluating the if* expression,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.4.8
The first thing eval does is check this expression,Is it a
number? Is it a sum? Is it a symbol? Is it a define? Is it a
greater? Is it an if? In other words,it checks the tag on the
expression,eventually determining that this is an if*
expression,Thus it dispatches the expression to eval-if,If
you look at the code handout,you will see that eval-if
first walks down the tree structure and pulls out the predicate
expression,It then evaluates that piece,as shown,
Slide 15.4.9
Depending on what value is returned by evaluating that first
piece,the evaluator will either walk the tree structure to get out
the consequent piece or the alternative piece,and it will
evaluate that piece,Thus,this special form will first evaluate
one piece,and then branch to an evaluation of one of the other
two pieces,but only one of them,
Slide 15.4.10
Then we can just step through the stages,The first stage
evaluates the full expression,This dispatches to eval-if,
which evaluates the first subexpression in the tree structure,
That does a second dispatch to the procedure that handles
greater* expressions,Here we walk further down the tree
to get the subexpressions to this procedure,and recursively
evaluate them,Having returned values (one by lookup,the
other as a number),we actually apply the primitive comparison
operator,and since the returned value is true we then extract
the appropriate piece of the tree (the consequent) and we
evaluate that piece,This just becomes a dispatch like the
previous cases,and you can see the recursive calls to eval used to get the values of the subexpressions before
the application of the primitive procedure,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.4.11
Once more,let's step back and extract some messages from this
exercise,
Note that eval-greater is really just like eval
sum,Both recursively call eval on the arguments,then
apply a primitive procedure to the resulting values,
Slide 15.4.12
On the other hand,our conditional,eval-if,behaved
differently,It is a special form,Rather than evaluating all of its
arguments,it only evaluates some of them,and in a particular
order,as shown on the slide,
6.001 Notes,Section 15.5
Slide 15.5.1
In the last example,when we introduced a new kind of
expression into our evaluator,we had to add a new dispatch into
eval,We created a new type checker for the expression,and
then we added a new clause to the cond to dispatch off to a
procedure to handle this kind of expression,
That's okay with a small number of things,but we would like to
add lots of operators,for multiplication,exponentiation,less
than,and so on,How can we add lots of operators to our
evaluator but still keep eval compact,and more importantly
easy for us to extend later?
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.5.2
We have already hinted at how to do this,We have noted that
operations like plus* and greater* are quite similar,
They evaluate all their arguments,and then they perform a
particular operation on the resulting values,That sounds like a
common pattern,so we would like to capture that pattern and
take advantage of that abstraction,
Slide 15.5.3
In particular,we will call this standard pattern an application,
It is an application of a particular operator to a set of arguments,
We can therefore implement a single case in our evaluator for
all applications,just one way of dispatching,That will allow us
to easily extend other operations to our evaluator,by simply
creating the operation and having the same dispatch handle the
application,
Slide 15.5.4
Here is our plan for accomplishing this,We will now first
evaluate the first subexpression of an application,Note that in
the previous cases we did not do this,we used the name to
dispatch in eval to the right primitive procedure,Now we
will evaluate the first subexpression to deduce the operation to
apply,But that means we have to have a way of getting access
to the appropriate operation,so we will put a name in the
environment for each operation that we want to treat as an
application,The value of that name will be a procedure and
then we can simply apply the procedure to the values of the
other subexpressions,How can we make this extension to our
evaluator?
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.5.5
As with the previous cases,we suggest that you look at a
printout of the code as we walk through this exposition,As
before,we will try highlighting the changes we make to our
existing evaluator,as we evolve to handle new kinds of
expressions,
First,we need a way of detecting an application,and we are
going to simply rely on the expression being a pair,that is,if it
is a combination,then it is an application,But if we are going to
assume that,then in our dispatch cases inside eval,
applications had best be things we do after we check for the
special kinds of things,And thus we can see a form evolving,In
eval we first check for the primitives (numbers and
symbols),and then we consider compound expressions and check for any that are special forms (things that do not
obey the normal rules for evaluation,such as define* and if*),Finally,we check for applications,that is,
we make sure we do have a compound expression,in which case we treat it as an application,Notice that we can
now drop the special checks for greater* and similar operations,since they will now be caught by the
application case,
Notice what we do if we have an application,Remember that the input is one of those tree structures representing
the expression,We grab the first element of the tree and evaluate it,That should give us a procedure,We then
walk down all the remaining elements of the tree and we will map eval down that list,We are just treating
map as a primitive here,so that it applies eval to each element of that list and create a new list of the resulting
values,
Then we will apply that procedure value to that list of arguments,
Slide 15.5.6
We know that recursively eval should do the right thing,If
our expression is an application of a named procedure to some
arguments,it will lookup the value of the name,which
presumably is attached to some procedure in the environment,
Similarly,mapping eval down the list of other expressions
will return a list of values,Now,how do we do an application?
Conceptually,apply says,check to see if the operator,the first
subexpression,is a primitive,that is,one of the things built into
my simple calculator (e.g,greater*,plus*),If it is,then get the
corresponding primitive and apply it to the arguments,that is
just do the normal Scheme thing to this primitive application,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.5.7
So to use this,I simply need to create primitive operators,and I
can do that with an abstract data type that stores away
underlying procedures (basic things we are inheriting from
Scheme) into a structure,We will make a primitive object by
taking an underlying procedure,tagging it,so that I can check
tags to find primitives and get out the actual procedure from the
data abstraction,
Slide 15.5.8
Remember that we have an environment,a little table
abstraction that contains bindings for things,so initially,before
I start using my evaluator,I can establish in that environment
some built in primitives,For example,I can put in a built in
primitive for plus* and one for greater*,and one for
the symbol true*,which is bound to the machine
representation for the Boolean value of true,
This should start looking familiar to you,Note that we are
setting up an environment that contains bindings of built in
names to built in primitive procedures,We also have for those
procedures a way of tagging them to tell that they are
procedures we have created,Then our evaluator is going to be able to use those built in primitives to execute the
application of a procedure to some arguments,
Slide 15.5.9
Okay,let's check it out,Let's evaluate the three shown
expressions in order in our little evaluator,This will check both
application and special forms in our new evaluator,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.5.10
First,what does our environment look like? It is just an abstract
data type of a table,containing names and values,This will
include a binding of z* to 9 which we know comes from
evaluating the define* expression,We also have a binding
for true*,since that was one of the built in definitions,We
also established bindings for names to representations of some
primitive procedures,
Slide 15.5.11
So let's step through the stages of evaluation of a simple
expression,As before,keep the code handy to follow along,We
start with the evaluation of the tree structure associated with
(plus* 9 6),the ' is just use to remind us that this is
tree structure along which we are walking,Eval steps
through a big dispatch checking the type of this expression
against each of its cases,Having decided it is not one of the
explicit things about which it knows,it checks that it is a
compound procedure,and since this is,eval assumes we
have an application,
Slide 15.5.12
So eval unwinds this into an application of whatever we get
by evaluating the first part of the tree to the list of the
evaluation of the remaining parts of the tree,Notice how we are
just walking through the tree,grabbing the pieces,and
recursively evaluating each one in turn,
Slide 15.5.13
So what happens with the eval of the symbol plus*?
Again,eval walks through each of its cases,checking the
type of this expression against those cases,It decides this is a
symbol,so it just does a look up in the environment,This just
returns the representation associated with that name,in this case
a tagged list,as shown,We now have the value of the operator,
We can move on to mapping eval down the list of other
pieces,which we know turns into a list of applying eval to
each piece in turn,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.5.14
And of course we have seen what this does,For each
expression,eval checks the type of the expression,decides it
is a number,and just returns that expression as the value,This
results in a list of the values 9 and 6,
Notice the form we now have,an application of a representation
of a primitive procedure to a list of values,So we can now do
the apply,
Slide 15.5.15
Apply first checks the argument to make sure it is a
primitive,Since this one does have the right tag,it can reduce
this to the underlying Scheme application of the procedure
associated with addition to the values,And that just reduces to
..,
Slide 15.5.16
...this case,We just have a Scheme application of one of its
primitive procedures to a list of values,Notice what we have
done,We have reduced evaluation of an expression to an
application of a procedure to a set of values,where we have
recursively evaluated each subexpression to get that procedure
and list of arguments,If the arguments were themselves
combinations,we would have done the same process
recursively on each of them,continually unwinding the
evaluation down to an application,in this case of addition to
some values,.,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.5.17
..,returning the value we expect,The key thing to see here is
the evolution of evaluation of a combination reducing to
application of a procedure to a set of values,
Slide 15.5.18
That example demonstrates that our normal procedure
applications now work,Let's make sure that special forms still
work,by considering the example of an if* expression,
Of course we get the behavior we expect,Given the tree
structure associated with this expression,eval checks each
case in its set of possibilities,checking the type of expression
until it deduces that this is an if* expression,In this case,it
dispatches off to a procedure designed to handle such
expressions,Note that it reaches this case before it gets to
normal applications,
Slide 15.5.19
That reduces to using the special procedure eval-if on
this expression,and check the code to remind yourself what
eval-if does,
Slide 15.5.20
Eval-if first evaluates the predicate,It takes the
expression,walks down the tree structure,pulls off the first
subexpression,and evaluates it,We will temporarily give the
result the name test,Key point is that we only evaluate this first
subexpression,before we look at anything else,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.5.21
Eval on this expression again walks through its cases,
decides this is a symbol,and does a look up on it,It goes into
the environment,finds the binding of this symbol in that table
abstraction,and returns the binding,in this case the underlying
representation for the Boolean value of true,
Slide 15.5.22
Having now obtained the value for this lookup,we can now
proceed to the conditional expression,Thus,based on the value,
it walks through the rest of the tree structure representing the
expression,and grabs the appropriate piece,
Slide 15.5.23
and this reduces to simply evaluating this piece of the
expression tree,and will return the value of this expression as
the value of the overall if* expression,Because this is a
number,this is self-evaluating and returns,.,
Slide 15.5.24
...just the expression itself,
Key thing to notice is how eval of an if* has walked
through the tree structure representing the expression,in a
particular order,and it has reduced the evaluation of that
expression to the evaluation of one of the subexpressions,
Also notice how we did not walk through all of the tree
evaluating all of the pieces,We only evaluated "on demand",
which is very different than an application,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.5.25
In fact,notice that apply is never called here,We only do
evaluation of pieces of the expression,Of course if one of those
pieces had been an application we would have used apply
but in terms of the parts of if* we never used apply,
If* is a special form that handles evaluation of subpieces in a
particular order,but never relies on an application of a
procedure since it is not a procedure application,
Slide 15.5.26
Once more,let's step back from our example and extract some
key observations,We have now extended our evaluator to
include applications,which means we can create other kinds of
expressions,But to do this,we must have applications be the
last thing considered as an option in eval,We are not
checking tags for applications,we are simply relying on the fact
that is something is not one of the known special forms,it must
be an application,
Also notice the form of eval,in which we first check for
primitives,then for special forms,then for applications,
Slide 15.5.27
And as we just said,we never used apply in the last
example,Applications evaluate all of their arguments before
proceeding,Special forms handle things in a different order,
exactly to control when arguments are evaluated,
So now we have seen a basic way of structuring an evaluator,
We have primitive expressions,primitive means of
combination,primitive conditionals,and primitive means of
application,which enable creation of other expressions,Next
time,we will look at further extending our evaluator,
6.001 Notes,Section 15.1
Slide 15.1.1
Our goal over the next few lectures is to build an interpreter,
which in a very basic sense is the ultimate in programming,
since doing so will allow us to define our language,
This is a somewhat surprising statement,But,in fact,as we will
see through these lectures,it really is correct,The reason it is
correct is the following,
Every expression we write in our language has a meaning
associated with it,Deducing the meaning associated with an
expression is the process of evaluation,And therefore,if we
can define the program that executes that deduction for us,our
definition of that program (i.e,our definition of the interpreter)
provides for us the exact specification of what's legal in our
language and how to deduce meanings of expressions within our language,
Our goal is to work through this understanding in stages,We will explore a series of examples,initially some very
simple examples,but culminating in a full-scale Scheme evaluator,We are going work our way up to this,letting
you see how to add the different pieces into an interpreter,
Before you proceed,however,there is a code handout that goes with this lecture,I suggest that you stop,go back to
the web page,and print out a copy of the code to have next to you as we go through the lecture,
Slide 15.1.2
First,let's set the stage for why we want to do this,why do we
want to build an interpreter? Why do we need an interpreter?
Think about what we have seen so far in this course,We have
spent a lot of time talking about,and using,the ideas of
abstraction,both procedural and data,We know that this is a
powerful idea,as it lets us bury details,and it certainly supports
the expression of ideas at an appropriate level,Said another
way,we can think about what we want to do and separate that
from the details of how to actually make it happen,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.1.3
But eventually,we need a process for unwinding all of those
abstractions,to get the value that corresponds to an expression’s
meaning,That means we need to implement semantics of
interpreting expressions,
Slide 15.1.4
Notice the words I used,I said,"we need a process for
unwinding those abstractions",If we can have such a process,
then we should be able to describe it in a procedure,that is our
language for describing processes,
In fact,you have already seen a version of this description,just
not as a procedure,What was the description?,.,the
environment model!
If you think about it,that makes sense,The environment model
just described the process for how to determine a meaning
associated with an expression,which in turn just unwrapped the
abstractions down to the primitive operations,Today,we want
to talk about how to actually build an evaluator as a procedure
rather than as that abstract environment model,
Slide 15.1.5
First,what are the stages of an interpreter? For the kind of
languages we are considering,Scheme or Lisp like languages,
typically there are five stages,There is a lexical analyzer,a
parser,an evaluator that typically works hand-in-hand with an
environment,and there is a printer,
Let's look at what goes on in each of these at a high level,
before we build them,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.1.6
To do that,let's talk about the input and output characteristics
of each of them,
By focusing on the input to each successive stage in the
interpretation,we can get a sense of what should happen at each
stage,and thus get a sense of how to build an interpreter,
Slide 15.1.7
The initial input is a string of characters,which represents the
typewritten version of the expression we want to evaluate,This
is exactly the thing that we would type in at a terminal if we
wanted to have an expression evaluated,So the initial input is a
string of characters,
Slide 15.1.8
The first step is to use a lexical analyzer to convert that string
of characters into units or words,This is shown here,where the
string gets converted into a set of words or isolated characters
like "(" and ")" and "+" and numbers,Thus the input to the next
stage is an ordered sequence of these units or words,
Slide 15.1.9
The second stage then parses those words into a structure that
we can use for evaluation,In particular,we convert the linear
sequence of words into a tree structure,We are using pairs here
for convenience but that is not required,We could use any other
representation of trees as well,
As we do this,we are going convert the self-evaluating
expressions into their internal values,So notice the form we get
for the next stage,it's a tree structure,and hanging off of the
leaves of the tree are numbers,symbols,or other objects that
are represented as basic words,This is the input to the next
stage,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.1.10
Now comes the heart of the interpreter,We want to take that
tree structure and deduce its value,First,notice the form of tree
structure,We will talk about this in detail later,but you can
already see how the parser has converted things into a tree,
Every time we see an "(",indicating the beginning of a new
combination,we have created a new list structure,If the parser
is already inside a list structure,it drops down a level,so that
we build up a tree where each horizontal slice through the tree
corresponds to some combination,Now what is the evaluator
supposed to do? It wants to take that tree structure,plus an
environment,and interpret it,And what does that mean? Think
of the environment as a way of associating names with more primitive values,It acts much like a dictionary,The
evaluator will use a set of rules to walk through this tree,looking up values associated with symbols from the
environment,i.e,from that dictionary,and then using the rules to reduce complex expressions to simpler things,
culminating in some simple value that we will return,
Slide 15.1.11
That value becomes input to the final stage,The printer simply
converts things back into the appropriate form for display on
the monitor,and then,.,
Slide 15.1.12
..,that just gets displayed to the user,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.1.13
So here is a summary of that process in words,
Slide 15.1.14
Our goal is to implement an interpreter,Actually,that's not
quite right,Our goal is for you to understand what goes into an
interpreter,which we will explore by implementing one,Since
the key part of an interpreter,the crucial part,is the evaluator,
we are going to concentrate almost exclusively on that,We are
going to use Scheme for all the rest of the pieces,that is,we
will use Scheme's lexical analyzer and parser,and Scheme's
printer,rather than building them from scratch,This means of
course that we will need to create an evaluator for a language
that looks a lot like Scheme in the sense of having a tree
structure as the output of its parser and a set of rules for
manipulating that tree structure,as the way of dealing with the actual evaluation,It also says that the syntax of the
language we are going to use to demonstrate an interpreter will need to look at lot like Scheme,in terms of things
like using parentheses to delimit expressions and other related issues,
We say this because we don't want you to get confused between what is going in Scheme and the general ideas of
building an evaluator and interpreter,Our goal is to build an interpreter,especially the evaluator part,and let you
see how that occurs and use that to explore the idea of how these things implement the rules for a language,We are
going to build our own simple language and its evaluator,For convenience,we are going to call this language,
Scheme*,It has a lot of the characteristics of Scheme,but we will use the convention that a * will be placed at the
end of every expression in our language,to distinguish it from the corresponding Scheme expression,
We'll start with a simple evaluator and work our way up,The first simple evaluator will be one that handles simple
arithmetic expressions,
6.001 Notes,Section 15.2
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.2.1
Our plan is to start by building an evaluator that handles
arithmetic expressions,and in fact we will restrict ourselves just
arithmetic expressions of two or fewer arguments,We would
like to be able to evaluate things like the example shown on the
slide,adding 24 to whatever we get by adding 5 and 6,
Notice the * at the end of the symbol plus to indicate that
this is something that we will build within our language,
Slide 15.2.2
And here is some code that captures how we will evaluate
expressions of this form,This is identical to the code listed in
the separate code handout,and I suggest you have that page
handy as we go through this development,
Slide 15.2.3
Notice what we are doing here,We are using our knowledge of
Scheme to describe the process of evaluating expressions in this
new language,We are writing,in Scheme,the description of
that process,
Okay,what do we need? We have a procedure for evaluating
expressions in our new language,called eval,Notice its
form,It has a way of dealing with the base case,which is an
expression that just consists of a number,And to do that it uses
type checking,
Then,we have a way of dealing with the compound case,Here,
it uses type checking to see if we have a sum and notice how
this works,It uses the keyword of the expression to determine the type of that expression,If the expression is a
sum then we will just add,using the primitive operation of addition,the values of the subexpressions,But a key
point arises here! To get those values we need to evaluate each subexpression as well,since we don't know at this
stage if they are just numbers or are themselves compound expressions,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.2.4
Let's look at this in more detail,First,let's look at the input to
this evaluation process,Remember that our expression,which
we typed in,is converted into list structure,as a tree of symbols
and numbers,It looks like what is shown on the slide,and this
is what gets handed to the evaluator as a representation of our
example expression,So let's treat this as if this exact tree
structure were passed in to eval,
Slide 15.2.5
And what does eval do with this input? Check the code on
the handout,Eval grabs the list and tests its tag,That means
it first checks to see if this whole thing is a number,Since it is
not,it takes the first element of this list structure and checks to
see if it is the special symbol plus*,
Slide 15.2.6
Having done that it dispatches on type to the right procedure to
handle this kind of expression,Having determine that it is a
sum by checking the tag,it sends it off to eval-sum,and
this is (for now at least) just a normal procedure application,
We apply the procedure to the expression,
Slide 15.2.7
So now eval has reduced this to applying eval-sum to
the tree structure shown,Notice what the body of eval
sum does,It walks down the tree,grabbing out the two
subexpressions,that is the first and second components of this
sum,Eval-sum then converts this into adding,using the
built-in primitive,whatever I get by evaluating the first
subexpression and whatever I get by evaluating the second
subexpression,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.2.8
That leads to this form,Notice what we have done,we have
reduced evaluation of one expression into some simple
primitive operations on evaluation of other subexpressions,And
what has that done? It has literally just walked down the tree
structure that represents this expression,pulling out the right
pieces,It has used the tag to tell us who to send the expression
to,and then it has simply grabbed cars and cdrs of the list
structure,and handed them off to new evaluations,Now at this
stage,evaluating the first subexpression (eval 24) is
easy,We see from our code that it will use type checking to
determine this is a number and simply return that expression,
And that’s nice! This is just pointing to the number 24 so the
number 24 gets returned,What about the other piece?
Slide 15.2.9
Well this just looks like the kind of expression we started with,
We are evaluating some list structure that happens to represent
a sum,It's got a tag at the front to say it is one of these
plus* expressions,so we can do exactly the same thing,
The evaluation of this expression will unwrap into an eval
sum of the same list structure,and that will reduce to a
primitive application of + to whatever I get by evaluating the
subexpressions,and that I get by walking down the tree,
grabbing the right pieces,applying eval and getting back the
numbers,
Slide 15.2.10
And now we see that we have unwrapped this abstraction down
to some primitive operations,primitive application of addition
to some simple expressions,in this case just numbers,And of
course this will finally reduce to the answer we expect,
However,a key thing to note is how this simple evaluator has
taken in a tree structure representing an expression and has
unwrapped it into successive evaluations until it reduces to a set
of applications of primitive built-in procedures to primitive
values,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.2.11
Since this is important,these stages of eval unwrapping into
simpler and simpler things,and the dispatching on type to the
correct procedure,let's look at this one more time,In this case,
let's focus on how eval unwinds the abstraction,and what
values are returned at each stage of the evaluation,As before,
you may find it convenient to have a copy of the code in front
of you as we go through this examination,
Slide 15.2.12
So we start with eval of this full expression,We've put a '
in front of the expression to show that we want list structure
equivalent to this expression,
Thus we start with an eval of this expression,
Slide 15.2.13
Eval first checks the type of this expression,deduces that it
is not a number,but is a sum (because of the type tag),so this
expression gets dispatched to eval-sum,Eval sends the
expression to the procedure that is exactly set up to deal with
this particular form of list structure,
Slide 15.2.14
Now,eval-sum says,"go ahead and add whatever I get by
evaling each of the pieces",We haven't actually specified
in what order to do the subpieces,but for convenience assume
that it is done from left to right,So we now need to trace down
the tree,and get (eval 24),
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.2.15
Eval once again checks the type of this expression,deduces
it is a number,and just returns that expression,literally a
pointer to that thing which is an internal representation for the
number 24,
Slide 15.2.16
Next,eval has to evaluate the second subexpression,so this
is the eval of the expression shown at top right,As before,
we are going to dispatch on type,i.e,check to see what kind of
"beast" this is,deduce that it is a "sum" and therefore pass this
on to the right procedure to handle sums,
Slide 15.2.17
Once more,eval-sum will reduce to applying the addition
operation to whatever it gets by evaluating the subpieces,Thus,
we need to extract the subexpressions and once again apply
eval to them,Notice the nice recursive unwinding that is
going on here,
Slide 15.2.18
Well this just unwraps one more time,Again,we will apply +
to whatever we get by evaluating the two pieces,and eval in
both cases just dispatches on type,determines the expression is
a number and returns the expression as the value,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.2.19
Notice where we are at this stage,We have unwrapped this
compound expression into a nested sequence of operations of
primitive things to primitive values,At this stage we can gather
up the things we have left to do,We have some deferred
operations,for example in the topmost eval-sum we can
now add 5 and 6 to get 11,and so on,reducing all the deferred
primitive operations down to a single value,
Slide 15.2.20
Thus,we have built a simple evaluator that handles sums of no
more than two arguments,Here are some key points to notice
from this exercise,since our goal is to understand the process of
evaluation,
First,eval does type tag checking,It dispatches based on
type,much like we saw earlier in the course,
Second,numbers are just numbers,so there is nothing really to
do,
Third,complex expressions nicely get recursively evaluated in
pieces,Eval unwraps a complex expression into an
evaluation of the simpler parts,plus a deferred operation to
gather the values back together,Numbers just get handled as numbers,And eventually we reduce this whole thing
down to a set of primitive operations on primitive values,
6.001 Notes,Section 15.3
Slide 15.3.1
Okay,now let's build on this basic system,Suppose we want to
give names to things,For example,suppose we want to have
the behavior shown here,in which we can store intermediate
results as named values,and then just use those names
anywhere that we would want to use the actual expression and
its resulting value,This is the kind of behavior we saw earlier in
the term in Scheme,how would we add that behavior to the
evaluator we are building for simple arithmetic expressions?
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.3.2
Of course,the first thing we realize is that this means we will
need a way of storing bindings between names and values,We
can certainly imagine getting the value for something,but now
we have to have a way of storing the name and value together,
Define* has to have some way of gluing pieces together,
So we need to add this capability to our evaluator,
Slide 15.3.3
This is now the second page of your code handout,Don't be
intimidated by this code,as we have highlighted the things we
have changed from the first evaluator,shown here in bold face
font,
So what have we added? First,we need another type checker,
something that checks whether the expression is a define*
expression,here called define?,We have also added two
new pieces to the evaluator,
Slide 15.3.4
Note that in this version of eval we have a way of creating
names for values,and a way of getting back the value
associated with a name,Thus,we have two new dispatches in
our evaluator,something that checks to see if the expression is
a symbol,in which case we will lookup its value; and
something that checks to see if the expression is a definition
(checked using the special tag define*),in which case we
dispatch to something the evaluates these special expressions,
Slide 15.3.5
Before we look at the procedures that will handle lookup
and eval-define,let's first think about what we need,
We will need a way of gluing things together,and we know
how to do that,Let’s just assume a data abstraction,called a
table,It has a constructor,make-table,It has a way of
getting things out of a table,thus given a table and symbol,
table-get gives us back either a nil to indicate no
binding for that symbol was present,or the actual binding,We
have a way of putting things into the table,table-put!,
and we have binding-value which when given a binding,
returns the value part of that pairing,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
What is the point of this? We can simply assume that this table abstraction exists,and then we can build an
environment,Let's define environment in our underlying Scheme to be a table,and then we can create
procedures for lookup and eval-define,
Slide 15.3.6
Looking up the value of a symbol is simply a matter of
manipulating the table,We find the binding in the table for this
symbol,and then return that value part of the binding,This is
simply manipulating the environment and we can abstract that
away,
Slide 15.3.7
The real thing we have added is a way of dealing with a new
kind of expression,something built by define* that is
creating a binding of a name and a value,So what should this
do?
This procedure says to walk the tree structure to get out the
name (remember this is just walking the tree structure,there is
no evaluation going on here),Then,evaluate the expression
that will provide the value of the binding,Notice the use of
eval which recursively returns to the top level,and evaluates
this expression using the same rules! Once I get a value,I stick
it into the table that represents the environment,paired with the
name in a binding,
Slide 15.3.8
Since we have added a new component to our evaluator,let's
again look at what happens if we evaluate these two expressions
in our evaluator,especially watching to see what is returned
each time we recursively call eval in this process,So keep
track of the argument and return values of eval as we trace
through this process,using the code handout to keep track of
this process,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.3.9
So let's use our extended evaluator,by evaluating this
define* expression,Remember that this expression is just
represented as tree structure,the ' is just used here to remind
that the parser will have reduced this input expression to the
equivalent tree structure,So what does eval do with this
expression? It is a dispatch,so it will deduce that this is a
define* expression,and will thus send the expression to a
handler,That procedure will decide to eval the value
subexpression,
Slide 15.3.10
Well that is a lot like what we saw before! We are now doing an
eval on this list structure,And we have literally walked
down the list structure for the first expression to grab off this
piece and now evaluate it,This just recursively determines the
type of expression (a sum) and thus passes it off to eval
sum which evaluates each of the subexpressions in turn,then
applies the primitive + operation to return a value (9),So the
value returned by evaluating this subexpression is just 9,
Slide 15.3.11
And then what does eval-define say to do (remember
this is where we were when we went off to evaluate the
subexpression),It simply takes the name (or symbol) x*,
grabbing that off of the tree structure,it takes the returned
value,and it binds them together in the table somewhere,
Slide 15.3.12
And having done its work,it just returns the symbol
undefined to tell us it has completed the evaluation of
this definition,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.3.13
So now let's evaluate an expression that includes one of these
names,such as the expression shown,As before,this expression
is converted to tree structure,which is then evaluated,The tag
checking dispatches this to eval-sum,That procedure says
to add together (using +) whatever I get by recursively
evaluating the subpieces,
First is (eval x*),Notice in line 9 of evaluator that
because this part of the tree structure is a symbol,we do a
lookup,That is,we use the table abstraction to get the
binding associated with this symbol,and returns the value
associated with that binding in the table,in this case 9,The rest
of the evaluation proceeds as in previous cases,
Slide 15.3.14
So in fact this was fairly straightforward,As long as we have
the abstraction of the table,we see how we can add bindings to
it,through definitions,Notice also how just walking the tree
structure of the expression allows us to grab off symbols
without evaluating them,in this case,getting x* so that we
could bind it together with a value in the table,Other than that,
all the other things still hold for our evaluator,we can simply
now give names to values,
Slide 15.3.15
Having now extended our evaluator by adding in this new
capability,let's step back and extract some key messages,First
of all,notice that we added two new dispatch clauses to our
evaluator,One of them just dispatches to eval-define
using standard type checking of tags,The other clause,
however,relies on using symbol?,the underlying Scheme
function,to check for a name,Why is this reasonable?
Note that we are relying on the underlying Scheme reader to
convert things into a parse tree,and in particular,that reader
will convert sequences of characters like x * into symbols in
the parse tree,which then gets passed on to eval,Thus,
eval will get a structure that can be handled by symbol?,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.3.16
Note the second thing we added,a table,Or rather,we added an
abstract data type of a table to represent our environments,Note
that we could have used any implementation of a table,We are
not relying in any way on the specifics on a table
implementation,and that is exactly the point,As long as we
have a way of creating tables,we can build an evaluator that
simply manipulates them,
Slide 15.3.17
A key thing to note as well is that we have now added a special
form to our little evaluator! If you think carefully about it,you
will note that prior to this we were simply evaluating normal
combinations,which for this language were sums of two
arguments,However,eval-define does something
different,It recursively evaluates only the second
subexpression! In other words,eval-define takes the
first subtree of the parse tree passed in,without evaluation,and
treats it as a symbol,It only does evaluation on the second
subtree,
6.001 Notes,Section 15.4
Slide 15.4.1
What else can we add to our evaluator? To this point,we have
no way of making decisions,We can't branch to do different
things depending on a value,So let's extend our evaluator to
handle this,i.e,let's extend our calculator to handle conditions
and ifs,Statements such as the example shown,in which if
something is true we want to do one thing,otherwise we want
to do something else,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.4.2
We can already see from our example that we are going to need
to add two kinds of things,First,we will need procedures that
return Boolean values,Thus,greater* should evaluate a
pair of expressions,then return true if the value of the first
expression is larger than the value of the second expression,Of
course,we can easily imaging adding other such expression to
our language,so long as we have some expressions that
evaluate expressions and return Boolean values,
The second thing we will need is an operation that lets us
control the branching of our code,We know that this should be
a special form in which we should first evaluate the first
subexpression,check its value,and then based on that value
either evaluate and return the second subexpression or evaluate and return the third subexpression,This means we
need to add an if* expression to our language,
Slide 15.4.3
As before,there is a fair amount of code here,but we have
highlighted the changes from the earlier version,First,we need
two new ways of dispatching on greater* and if*
expressions,just using type checking as before,This means we
need ways of checking the type of such expressions,and we
need to add dispatch clauses to the cond expression inside
eval,one for each new type of expression,This is just like
before,
Slide 15.4.4
First,greater* expressions,Remember that exp will be
a pointer to some tree structure that has been created by the
parser,The first subexpression in that tree structure will be the
symbol greater*,The second subexpression will be some
other expression,as will the third subexpression,Our plan is to
get the values of the second and third subexpressions,We will
use tree operations to extract the right pieces,the recursively
evaluate them,Once we have the values of these
subexpressions,we can use the underlying Scheme primitive
operation to compare the numerical values and return a Boolean
answer,
This may seem a little odd; why not use Scheme's operation directly? Remember,our goal is to understand how to
connect the evaluator to built-in primitives,and to allow for alternative ways of executing operations,So remember
what we do,we walk the tree,grab the pieces,recursively evaluate them,then pass the pieces on to primitive
procedures to complete the reduction to a value,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.4.5
What about if* expressions? What happens when we
dispatch to eval-if?
Once more,remember that what is passed in to this procedure is
tree structure,created by the parser,This tree structure will
several pieces,the tag if*; then tree structure representing the
predicate of the expression; then tree structure representing the
consequent of the expression; then tree structure representing
the alternative of the expression,Notice how in our procedure
we use list manipulation to pull off each piece of the tree,and
then we evaluate the first one,
Once we have the value of test,we can then use the
underlying primitive mechanism for branching,cond,to decide that if test is true,then and only then,
take the consequent and evaluate it,If test is false,then and only then,take the alternative and evaluate it,
Notice the order in which things are done,This is a special form in which we first evaluate the predicate,
Depending on that value,we either evaluate the consequent or the alternative,and return that value as the value of
the overall expression,
Slide 15.4.6
So having added these two new expressions to our evaluator,
let's look at some examples of evaluation using them,as shown
on the slide,tracing through the process of recursively using
eval,
Slide 15.4.7
We can assume that we have done the evaluation of
(define* y* 9),since we know what that does,Now
let's look at evaluating the if* expression,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.4.8
The first thing eval does is check this expression,Is it a
number? Is it a sum? Is it a symbol? Is it a define? Is it a
greater? Is it an if? In other words,it checks the tag on the
expression,eventually determining that this is an if*
expression,Thus it dispatches the expression to eval-if,If
you look at the code handout,you will see that eval-if
first walks down the tree structure and pulls out the predicate
expression,It then evaluates that piece,as shown,
Slide 15.4.9
Depending on what value is returned by evaluating that first
piece,the evaluator will either walk the tree structure to get out
the consequent piece or the alternative piece,and it will
evaluate that piece,Thus,this special form will first evaluate
one piece,and then branch to an evaluation of one of the other
two pieces,but only one of them,
Slide 15.4.10
Then we can just step through the stages,The first stage
evaluates the full expression,This dispatches to eval-if,
which evaluates the first subexpression in the tree structure,
That does a second dispatch to the procedure that handles
greater* expressions,Here we walk further down the tree
to get the subexpressions to this procedure,and recursively
evaluate them,Having returned values (one by lookup,the
other as a number),we actually apply the primitive comparison
operator,and since the returned value is true we then extract
the appropriate piece of the tree (the consequent) and we
evaluate that piece,This just becomes a dispatch like the
previous cases,and you can see the recursive calls to eval used to get the values of the subexpressions before
the application of the primitive procedure,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.4.11
Once more,let's step back and extract some messages from this
exercise,
Note that eval-greater is really just like eval
sum,Both recursively call eval on the arguments,then
apply a primitive procedure to the resulting values,
Slide 15.4.12
On the other hand,our conditional,eval-if,behaved
differently,It is a special form,Rather than evaluating all of its
arguments,it only evaluates some of them,and in a particular
order,as shown on the slide,
6.001 Notes,Section 15.5
Slide 15.5.1
In the last example,when we introduced a new kind of
expression into our evaluator,we had to add a new dispatch into
eval,We created a new type checker for the expression,and
then we added a new clause to the cond to dispatch off to a
procedure to handle this kind of expression,
That's okay with a small number of things,but we would like to
add lots of operators,for multiplication,exponentiation,less
than,and so on,How can we add lots of operators to our
evaluator but still keep eval compact,and more importantly
easy for us to extend later?
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.5.2
We have already hinted at how to do this,We have noted that
operations like plus* and greater* are quite similar,
They evaluate all their arguments,and then they perform a
particular operation on the resulting values,That sounds like a
common pattern,so we would like to capture that pattern and
take advantage of that abstraction,
Slide 15.5.3
In particular,we will call this standard pattern an application,
It is an application of a particular operator to a set of arguments,
We can therefore implement a single case in our evaluator for
all applications,just one way of dispatching,That will allow us
to easily extend other operations to our evaluator,by simply
creating the operation and having the same dispatch handle the
application,
Slide 15.5.4
Here is our plan for accomplishing this,We will now first
evaluate the first subexpression of an application,Note that in
the previous cases we did not do this,we used the name to
dispatch in eval to the right primitive procedure,Now we
will evaluate the first subexpression to deduce the operation to
apply,But that means we have to have a way of getting access
to the appropriate operation,so we will put a name in the
environment for each operation that we want to treat as an
application,The value of that name will be a procedure and
then we can simply apply the procedure to the values of the
other subexpressions,How can we make this extension to our
evaluator?
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.5.5
As with the previous cases,we suggest that you look at a
printout of the code as we walk through this exposition,As
before,we will try highlighting the changes we make to our
existing evaluator,as we evolve to handle new kinds of
expressions,
First,we need a way of detecting an application,and we are
going to simply rely on the expression being a pair,that is,if it
is a combination,then it is an application,But if we are going to
assume that,then in our dispatch cases inside eval,
applications had best be things we do after we check for the
special kinds of things,And thus we can see a form evolving,In
eval we first check for the primitives (numbers and
symbols),and then we consider compound expressions and check for any that are special forms (things that do not
obey the normal rules for evaluation,such as define* and if*),Finally,we check for applications,that is,
we make sure we do have a compound expression,in which case we treat it as an application,Notice that we can
now drop the special checks for greater* and similar operations,since they will now be caught by the
application case,
Notice what we do if we have an application,Remember that the input is one of those tree structures representing
the expression,We grab the first element of the tree and evaluate it,That should give us a procedure,We then
walk down all the remaining elements of the tree and we will map eval down that list,We are just treating
map as a primitive here,so that it applies eval to each element of that list and create a new list of the resulting
values,
Then we will apply that procedure value to that list of arguments,
Slide 15.5.6
We know that recursively eval should do the right thing,If
our expression is an application of a named procedure to some
arguments,it will lookup the value of the name,which
presumably is attached to some procedure in the environment,
Similarly,mapping eval down the list of other expressions
will return a list of values,Now,how do we do an application?
Conceptually,apply says,check to see if the operator,the first
subexpression,is a primitive,that is,one of the things built into
my simple calculator (e.g,greater*,plus*),If it is,then get the
corresponding primitive and apply it to the arguments,that is
just do the normal Scheme thing to this primitive application,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.5.7
So to use this,I simply need to create primitive operators,and I
can do that with an abstract data type that stores away
underlying procedures (basic things we are inheriting from
Scheme) into a structure,We will make a primitive object by
taking an underlying procedure,tagging it,so that I can check
tags to find primitives and get out the actual procedure from the
data abstraction,
Slide 15.5.8
Remember that we have an environment,a little table
abstraction that contains bindings for things,so initially,before
I start using my evaluator,I can establish in that environment
some built in primitives,For example,I can put in a built in
primitive for plus* and one for greater*,and one for
the symbol true*,which is bound to the machine
representation for the Boolean value of true,
This should start looking familiar to you,Note that we are
setting up an environment that contains bindings of built in
names to built in primitive procedures,We also have for those
procedures a way of tagging them to tell that they are
procedures we have created,Then our evaluator is going to be able to use those built in primitives to execute the
application of a procedure to some arguments,
Slide 15.5.9
Okay,let's check it out,Let's evaluate the three shown
expressions in order in our little evaluator,This will check both
application and special forms in our new evaluator,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.5.10
First,what does our environment look like? It is just an abstract
data type of a table,containing names and values,This will
include a binding of z* to 9 which we know comes from
evaluating the define* expression,We also have a binding
for true*,since that was one of the built in definitions,We
also established bindings for names to representations of some
primitive procedures,
Slide 15.5.11
So let's step through the stages of evaluation of a simple
expression,As before,keep the code handy to follow along,We
start with the evaluation of the tree structure associated with
(plus* 9 6),the ' is just use to remind us that this is
tree structure along which we are walking,Eval steps
through a big dispatch checking the type of this expression
against each of its cases,Having decided it is not one of the
explicit things about which it knows,it checks that it is a
compound procedure,and since this is,eval assumes we
have an application,
Slide 15.5.12
So eval unwinds this into an application of whatever we get
by evaluating the first part of the tree to the list of the
evaluation of the remaining parts of the tree,Notice how we are
just walking through the tree,grabbing the pieces,and
recursively evaluating each one in turn,
Slide 15.5.13
So what happens with the eval of the symbol plus*?
Again,eval walks through each of its cases,checking the
type of this expression against those cases,It decides this is a
symbol,so it just does a look up in the environment,This just
returns the representation associated with that name,in this case
a tagged list,as shown,We now have the value of the operator,
We can move on to mapping eval down the list of other
pieces,which we know turns into a list of applying eval to
each piece in turn,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.5.14
And of course we have seen what this does,For each
expression,eval checks the type of the expression,decides it
is a number,and just returns that expression as the value,This
results in a list of the values 9 and 6,
Notice the form we now have,an application of a representation
of a primitive procedure to a list of values,So we can now do
the apply,
Slide 15.5.15
Apply first checks the argument to make sure it is a
primitive,Since this one does have the right tag,it can reduce
this to the underlying Scheme application of the procedure
associated with addition to the values,And that just reduces to
..,
Slide 15.5.16
...this case,We just have a Scheme application of one of its
primitive procedures to a list of values,Notice what we have
done,We have reduced evaluation of an expression to an
application of a procedure to a set of values,where we have
recursively evaluated each subexpression to get that procedure
and list of arguments,If the arguments were themselves
combinations,we would have done the same process
recursively on each of them,continually unwinding the
evaluation down to an application,in this case of addition to
some values,.,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.5.17
..,returning the value we expect,The key thing to see here is
the evolution of evaluation of a combination reducing to
application of a procedure to a set of values,
Slide 15.5.18
That example demonstrates that our normal procedure
applications now work,Let's make sure that special forms still
work,by considering the example of an if* expression,
Of course we get the behavior we expect,Given the tree
structure associated with this expression,eval checks each
case in its set of possibilities,checking the type of expression
until it deduces that this is an if* expression,In this case,it
dispatches off to a procedure designed to handle such
expressions,Note that it reaches this case before it gets to
normal applications,
Slide 15.5.19
That reduces to using the special procedure eval-if on
this expression,and check the code to remind yourself what
eval-if does,
Slide 15.5.20
Eval-if first evaluates the predicate,It takes the
expression,walks down the tree structure,pulls off the first
subexpression,and evaluates it,We will temporarily give the
result the name test,Key point is that we only evaluate this first
subexpression,before we look at anything else,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.5.21
Eval on this expression again walks through its cases,
decides this is a symbol,and does a look up on it,It goes into
the environment,finds the binding of this symbol in that table
abstraction,and returns the binding,in this case the underlying
representation for the Boolean value of true,
Slide 15.5.22
Having now obtained the value for this lookup,we can now
proceed to the conditional expression,Thus,based on the value,
it walks through the rest of the tree structure representing the
expression,and grabs the appropriate piece,
Slide 15.5.23
and this reduces to simply evaluating this piece of the
expression tree,and will return the value of this expression as
the value of the overall if* expression,Because this is a
number,this is self-evaluating and returns,.,
Slide 15.5.24
...just the expression itself,
Key thing to notice is how eval of an if* has walked
through the tree structure representing the expression,in a
particular order,and it has reduced the evaluation of that
expression to the evaluation of one of the subexpressions,
Also notice how we did not walk through all of the tree
evaluating all of the pieces,We only evaluated "on demand",
which is very different than an application,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 15.5.25
In fact,notice that apply is never called here,We only do
evaluation of pieces of the expression,Of course if one of those
pieces had been an application we would have used apply
but in terms of the parts of if* we never used apply,
If* is a special form that handles evaluation of subpieces in a
particular order,but never relies on an application of a
procedure since it is not a procedure application,
Slide 15.5.26
Once more,let's step back from our example and extract some
key observations,We have now extended our evaluator to
include applications,which means we can create other kinds of
expressions,But to do this,we must have applications be the
last thing considered as an option in eval,We are not
checking tags for applications,we are simply relying on the fact
that is something is not one of the known special forms,it must
be an application,
Also notice the form of eval,in which we first check for
primitives,then for special forms,then for applications,
Slide 15.5.27
And as we just said,we never used apply in the last
example,Applications evaluate all of their arguments before
proceeding,Special forms handle things in a different order,
exactly to control when arguments are evaluated,
So now we have seen a basic way of structuring an evaluator,
We have primitive expressions,primitive means of
combination,primitive conditionals,and primitive means of
application,which enable creation of other expressions,Next
time,we will look at further extending our evaluator,
