![]() |
![]() |
|
Due
18 (Thu) start of class19 (Fri) 23:59
We continue to build on the language implementation of F1/F2. You can implement this homework in either Java or Racket. You may use the F2-soln if you want. Link is usuable once you E2L-submit F2; just let me know if you need to bypass that.
F3 is just like F2, except we now allow one variable to shadow another.
For example, we might have a moon x on … eclipses …
which in turn contains another moon x on … eclipses …
somewhere inside of it.
In that case the inner x should shadow the outer one:
⇒
moon x on 7 eclipses moon x on 5 eclipses tilt x sum 3
⇒
moon x on 5 eclipses tilt x sum 31
⇒
tilt 5 sum 3
⇒
8.
And of course,
shadowing may occur between non-adjacent scopes:
moon x on 3 eclipses moon y on 4 eclipses moon x on 5 eclipses ….
So what does our interpreter need to do? Well, when eval does substitution in a LetExpr, we just need to be a bit cautious: If we're substituting every x in an Expr, and we encounter a moon x on Exprrhs eclipses Exprbody then we shouldn't substitute any xs inside the Exprbody. Though of course, if we encounter a moon y on Exprrhs eclipses Exprbody, then this doesn't affect our substitution.
Using our programming-language vocabulary: when substituting, only substitute “free occurrences” of an Id in E1, not any “bound occurrences”2.
As an example, for the expression
moon x on 3 eclipses (moon y on 4 eclipses tilt x sum (moon x on 5 eclipses tilt x sum y)),
we have the syntax tree drawn at
the right albeit with
sum
written as boii
,
with a dotted-arrow from each binding-occurrence to its bound occurrence(s).
This corresponds to some runnable test-cases3:
(check-expect (eval (string->expr "moon x on 3 eclipses (moon y on 4 eclipses tilt x sum (moon x on 5 eclipses tilt x sum y))")) (+ 3 (+ 5 4))) (check-expect (subst "x" 3 (string->expr "(moon y on 4 eclipses tilt x sum (moon x on 5 eclipses tilt x sum y))")) (string->expr "(moon y on 4 eclipses tilt 3 sum (moon x on 5 eclipses tilt x sum y))")) |
hint/spoiler: in moon zed on Exprinitialize eclipses Exprbody, we know thatzed can never occur free in Exprbody. We don't even need to look inside it, if we are only looking for free occurrences of (a surrounding) zed!
Update F2 to F3,
by the necessary changes to enable shadowing.
You are encouraged to build on your own previous solution,
but you may also use the
F2-soln (.rkt)
F4 adds (non-recursive) functions and function-application to our language:
Expr → … | FuncExpr | <FuncApplyExpr> ;>>>F4 FuncExpr → orbit Id aroundExpr ;>>>F4 Interpretation: function-value; the Id is the parameter, and the Expr is the function's body. FuncApplyExpr → launch Expr into Expr ;>>>F4 Interpretation: apply the function (first Expr) to the argument (2nd Expr). |
Here is
the function (λ (x) (+ (* 3 x) 1)) written in F4:
orbit x around tilt tilt x spr 3 sum 1.
And, here is the (uniterated) collatz function,
(λ(n) (if (even? n) (* n 1/2) (+ (* 3 n) 1))), written in F4:
orbit n around if tilt n fall 2 equinox tilt n spr 0.5 : tilt tilt 3 spr n sum 1 |
A FuncApplyExpr represents calling a function. Here are two expressions, both evaluating to 5·3+1 = 16:
moon tripleAndInc on orbit x around tilt tilt x spr 3 sum 1 body of `tripleAndInc` eclipses launch tripleAndInc into 5 call tripleAndInc(5)… launch orbit x around tilt tilt x spr 3 sum 1 into 5 Equivalent to above: apply a function-literal (w/o bothering to give it a name) |
orbitand
tilt. They'll just be in comments since they're not racket code. Or hey, you can put them in a string, and then pass those strings to parse as test-cases!.
Note: You won’t be able to evaluate function-applications for recursive functions yet (see F5), but we can still write the test cases! (You can comment out that one test case for now, since it’ll trigger a run-time exception otherwise.)
(define (make-adder n) (lambda (m) (+ n m))) ((make-adder 3) 4) ; evals to 7 ; Note that `(make-adder 3)` evals to `(lambda (m) (+ 3 m))` ; the `((` means we have *two* function-applications: ; we first call `make-adder` (getting back a lambda-value), ; then we call that result we got back. |
heads-up: The provided racket and java parsers each return a punctuation-character as a single token. So when parsing “<=” you'll need to pop twice to consume both punctuation-characters.
Semantics of a function-literal: Just as numbers are self-evaluating, so are FuncExprs. Evaluating (an internal representation of) a function results in that same (internal representation of the) function. We won’t actually evaluate the body until the function is applied. (This is exactly how Java, racket, python, javascript, etc. treat functions.)
Testing function-values: How can we make a test-cases, when the result isn't a number? For example, for evaluating orbit n around tilt n sum 1 or launch makeAdder into 5 (whose expected-results are both <FuncExpr> structs). You have two approaches:
(check-expect (eval (string->expr "orbit n around tilt n sum 1")) (make-my-func-expr "n" (make-binop "sum" "n" 1))) |
["orbit n around tilt n sum 1" ,(make-my-func-expr "n" (make-binop "sum" "n" 1))] |
[optional] Why the comma?: If you want to understand why it doesn't work without a comma, and what the comma does for us, you can read ../../Lectures/backquote.html (recommended, but not necessary to complete this homework).tl;dr: The test-harness's list of program+result pairs isn't created by calling list (over and over for each pair); instead it was created by using a single backquote, “`”: (define all-tests `(("tilt 2 sum 3" 5) …). This quote then distributes over all the nested open-parens to create sub-lists woo-hoo!, a handy time-saver. In fact, it also quotes the values inside those sub-lists — which didn't previously matter to us, because quoting string-literals and integer-literals has no effect ('16 is 16, and '"hello" is "hello").
However: quoting an identifier turns it into a symbol ('hello is 'hello :-). So `(3 (sqrt 16) 5) is (list 3 (list 'sqrt 16) 5), and not (list 3 4 5) like we hoped. To suppress the quoting of (sqrt 16), we precede it with a comma: `(3 ,(sqrt 16) 5) gives (list 2 3 4). (The comma is known as unquote.)
The semantics of eval’ing the function-application
launch Expr1 into Expr0
launch Expr0 into Expr1:
external resource: Another description of this algorithm is in VT's textbook's sectionA Substitution-Based Model of Evaluation.
You can submit your prolog queries inside a comment of a (racket) file, at the top of your submitted hardcopy, thanks! Only include your added rules/facts, not the rest of the provided .pl file.
Note: Put our prolog queries inside a comment of a (racket) file, at the top of your submitted hardcopy, thanks! Only include your added rules/facts, not the rest of the provided .pl file.
Prolog paths (transitive closure):
Continuing from defining compatible in the previous problem,
write friendly:
We say two super-people are friendly if
they are connected through some chain of compatible people.
For example,
bubbles
and
rafael
would be friendly if
compatible(bubbles,magikarp), compatible(magikarp,squirtle),
and
compatible(squirtle,rafael).
(That is: friendly is the reflexive, transitive closure of the compatible relation.)
Everybody, of course, is trivially friendly with themselves
(since they’re connected by a chain of 0 compatible others).
Note:It can be okay if you get infinite loops:
Since there can be cycles of friendly people (unlike our ancestor example), prolog (which uses depth-first search, rather than breadth-first), can get into infinite loops! In particular, asking friendly(bubbles,some_person_who_isnt_friendly_w_bubbles) can trigger an infinite loop. Asking friendly(bubbles,X) will give you back the same solutions repeatedly. (But asking about two people who are friendly should work just fine.) We'll just ignore such loops, for this homework.(Fwiw: the general solution would be to either (a) add an “exclusion” list of people already tried5, or (b) use datalog, a restricted version of prolog which uses BFS, and is always guaranateed to terminate (See #lang racklog, in racket), or (c) add
We're using prolog to learn about the mindset of declarative programming (and its cool pattern-matching/unification) — I'm not interested in us learning about the details of prolog's internal algorithms.
The interpreter project is based on the first chapters of Programming Languages and Interpretation, by Shriram Krishnamurthi. As a result, this homework assignment is covered by the Creative Commons Attribution-NonCommercial-ShareAlike 3.0 United States License. Although we’re using a different dialect of racket than that book, you might find it helpful to skim it.
eval(string->expr("moon x on 5 eclipses tilt x sum 3")) = eval(string->expr("tilt 5 sum 3")) = eval(string->expr("8")) |
This page licensed CC-BY 4.0 Ian Barland Page last generated | Please mail any suggestions (incl. typos, broken links) to ibarland ![]() |