ITEC380 Expr Trees

Expr Trees
eval/parse X basics; adding identifiers

Due 2020-Oct-31 (Sat.) 14:00~~30 (Fri.) at start of class~~. .
In addition, due Oct-28 (Wed) by the start of class: complete some initial X2 tests (and soln)

Note that X1 is “add code that is extremely similar to what already exists”, but X2 is not at all rote: it requires a solid understanding of the parse-tree representation and how the functions work.

Over the course of several homeworks, we'll implement a “tower” of languages (X0, X1, …, X6) each language incorporating more features than the previous. X0 is provided for you.

Download & get the provided X0 running.

Each time we add new syntax to our language, we must update three methods:

parse!, which turns source-code into an internal representation,
expr->string which turns internal representation back into source-code,
eval, which actually interprets the program, returning a value (a number for X0-X2, and a number-or-function in X3).

I recommend implementing and testing the methods in this order. See the provided test-cases, for examples. There might be other helper functions to implement as well.

Your solution may be in Java or in Racket (or see me, if you want to use a different language). (The exception is X1, which you must implement in both Racket and Java.)

You do not need to worry about error-checking the programs we process — we will only concern ourselves with syntactically correct Vi programs.
(As a matter of defensive programming, you might still want to add some assertions/sanity-checks in your code.)

Use the D2L discussion board for group questions and answers.

This project is based, in part, 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 scheme/racket than that book, you might find it helpful to skim it.

Deliverables:

If adding to an existing file, add a comment “;>>> X1” or “;>>> X2” next to the area you are changing; when grading I will search for “>>>”. This helps immensely, thanks.

Since X2 subsumes X1, you only need to submit X1 in one language, and X2 in the other.

For the D2L dropbox, please submit your files individually (no .zip/.jar). Either:

X2{,-test}.rkt, and java files with an X1 implementation ({Expr,ExprTest,BinOp,IfPos}.java)
X1{-test}.rkt, and java files with an X2 implementation ({Expr,ExprTest,BinOp,IfPos,Id,Let}.java)

(You should use those filenames, except that your new Java classes can be whatever (reasonable) name you chose.)

You do not need to include any provided files that are unchanged:

student-extras.rkt, scanner.rkt;
UtilIan.java, Pair.java;
{IfZero,Num,Value,Paren}.java

~~I should be able to download your files and run your code from the download-directory. Thus, you should D2L-submit any necessary helper files like scanner.rkt and UtilIan.java.~~

You should not (have your IDE) add any package statements. (I'll strip them out so that they can still compile in my own directory-structure.)

(15pts) Implement X1 in racket and in Java, both.
X1 is just like X0, but with two additional types of expressions:

Op ::= … | mod     Interpretation: “remainder”; see A.iii for details

Expr ::= … | IfPos
IfPos ::= if+ Expr thn Expr els Expr ;
Interpretation: if 1st Expr evals to a positive odd number, result is the 2nd Expr, else the 3rd

Be sure to write test cases first; To ensure everybody makes test cases to cover the basics, I've spelled out these strongly-suggested X2-initial-tests.html.

Add the mod operator.
1. update parse! (Java: Expr.parse) to handle this new operator (after writing test cases).
  (Or, if you don't need to update this function, understand why.)
2. update expr->string (Java: Expr.toString) to handle this new operator (after writing test cases).
  (Or, if you don't need to update this function, understand why.)
3. update eval to handle this new type of operator. The semantics of ~x mod y! is:
  the remainder after dividing x by y, where the result is always between 0 (inclusive) and y (exclusive)¹
  In particular, the result should never be positive if y<0. Notice that this is slightly different behavior than either Java's built-in % (which behaves differently on negative numbers), and from Racket's built-in modulo (which only accepts integers). In both racket and Java, you can calculate this as y * (x/y - ⌊x/y⌋), where ⌊r⌋ means the the floor of r.
  
  Note that you are already provided sufficient test cases for mods, in the comments of the X0 test-case files.
Add the IfPos expression.
1. update parse! (Java: Expr.parse) to handle this new type of expression (after writing test cases). For Java: see parsing note below.
2. update expr->string (Java: Expr.toString) to handle this new type of expression (after writing test cases).
3. update eval to handle this new type of expression. The semantics of if+ Expr₀ thn Expr₁ els Expr₂; is:
  first evaluate just Expr₀ to get value v₀; if v₀ is positive, then evaluate Expr₁ and return its value; otherwise evaluate Expr₂ and return that value.
  For example, if+ 3.1 thn 7.5 els 44; would evaluate to 7.5.
  (Note how the non-returned Expr never gets evaluated: you are implementing short-circuit semantics for IfPoss!)
  
  You must make your own test cases for IfPoss; include at least two as-simple-as-possible tests, and two tests with more deeply nested Exprs. I suggest including one where the IfPos is not the top-level comment (e.g., a add expression which contains a IfPos as one of its operands).

(25pts) Implement X2 in either racket or Java (your choice).
X2 adds identifiers to X1:

Expr ::= … | Id | LetExpr

LetExpr ::= let Id <- Expr in Expr²

where Id can be pretty much any string: it can be any token returned by our scanner that isn't already a reserved-token/punctuation. This simplistic ³ definition lets us keep our code easy: if we get a token (via peek/UtilIan~hasNextSplittingBy), and it's not the start of any other expression, then it's an identifier! (Assume for now that any nested letExpr expressions will use different Ids. We'll handle shadowing in X3, later.)

Update your three methods parse, toString (a.k.a. expr->string), eval.

add Ids to your data-definition (after deciding what data-type will represent them); then:
1. update expr->string (or, Expr.toString) to handle this new type of expression (after test cases)
2. update parse! (or, Expr.parse) to handle this new type of expression (after test cases)
3. update eval (or, Expr.eval) to handle this new type of expression:
  eval'ing an identifier simply throws an error.
  
  You don't need test cases for evaling Ids. Though if you want to be spiffy, you can use check-error in racket, or assertThrows in JUnit5. In JUnit4, the hack-ish approach is to put “ExpectedException.none().expect(RunTimeException.class)” on the line before the one that should trigger an error.
Next, add LetExprs to your data-definition (after deciding what data-type will represent them); then:
1. update expr->string (or, Expr.toString) to handle this new type of expression (after test cases)
2. update parse! (or, Expr.parse) to handle this new type of expression (after test cases). Since our parser splits on punctuation, < will be read as one token, and - as second (both racket and Java).
3. Think about how to update eval to handle this new type of expression. Now we get to the heart of the issue! Write test cases, after reading the rest of this bullet.
  In order to write eval, we need to define the semantics of let Id <- E₀ in E₁:
  - Evaluate E₀; let's call the result v₀.
  - Then, substitute v₀ for all occurrences of Id inside the tree E₁; name the result of the substitution E′.
  - Evaluate E′, and return that result.
  For example: let x <- 5 in ~x sub 3! ⇒ ~5 sub 3! ⇒ 2⁴. Be sure to write test cases for your substitution function before you write its code; include several trivial and easy tests, along with a couple of more complicated nestings and one deeply nested expression.
  
  Observe that when evaluating a (legal) X2 program, eval will never actually encounter an Id — that Id will have been substituted out before we ever recur down to it.
4. Now that we realize that eval will need to do substitution in a tree, and that's a smaller, simpler, self-contained task — perfect for its own helper-function substitute. This function only does substition in a tree, and does not attempt to do any evaluating.
  Go and write substitute (after test cases for it), before implementing eval for LetExprs. (And when starting substitute, start from the template for Exprs.) For the test cases, think about exactly types you'll be wanting to sent to substitute. Your simplest test-cases won't even contain a LetExpr — substitute is a function whose purpose-statement stands on its own, entirely independent of LetExprs and eval!
  Hint: Substituting a variable with a value in an syntax-tree is essentially the same as replacing every occurrence of one name with another in an anc-tree. (The only difference is that an anc-tree had only two cond-branches, while Expr has around five, though the code for most of those are very similar.)
  
  (Note: you must do substitution in the parse tree; no credit given for string-substitution ⁶.)
5. Finally, with the substitute helper written, we're ready: write eval for LetExprs.
  Hint: Your code will correspond almost word-for-word to the semantics given above.

Parsing IfPos in Java:

tl;dr: skip to the code-text below and use either (a) or (b) when writing your Java code for parsing IfPos's.

When parsing the token if+, there is a gotcha in using UtilIan#hasNextSplittingBy (and also if you were using the built-in Scanner#hasNext(String)): Those functions take a regular expression, and report whether the scanner's next token matches that regex. In all the provided code, that distinction is a non-issue here, since "if=" and "~" and "els", as regular-expressions, match those exact same things as Strings.

However, "if+", as a regular expression, matches strings like "if" and "ifffff" (the plus-sign is regexp for "one or more occurrences of the preceding"). So if you ask if a scanner's next token matches "if+", and the next token is "if" followed by a non-f, the function returns false. [However, if you call next, you *would* get back the next token!] So we want to pass a regex that matches "if+":

SOLUTION / what you need to do / tl;dr:

In Expr.java#parse: rather than:

if (UtilIan.hasNextSplittingBy( s, IfPos.START_TOKEN, PUNCTUATION )

change it to one of the following:

if (UtilIan.hasNextSplittingBy( s, "if\\\\+", PUNCTUATION ) Why two backslashes? Because we need a regexp with a \+, and to get a backslash into a string in Java I need "\\".
if (UtilIan.hasNextSplittingBy( s, Pattern.quote(IfPos.START_TOKEN), PUNCTUATION ) and also either import java.util.regex.Pattern; or just use that fully-qualified name directly when calling Pattern#quote.

Both of these make it so you're passing a regular expression that matches the three-character-string "if+". Version (a) is hackier (but shorter than) version (b), and is acceptable for this hw. Note that you can't apply this fix when inside your IfPos.java when defining START_TOKEN, because that token gets used in another place (ifPos#toString) where it needs to be a raw string, not a (string representing a) regex.

¹ Because we don't need to check for bad inputs, it's fine to have your interpreter crash if y=0. If you prefer to "control" crash — creating a meaningful error message and calling error or throw yourself — you are also welcome to do that. ↩

² For comparison, here is what comparable constructs look like in other languages:

ML-like:	let x = 2+3 in x*9 end;
lisp-like:	(let {[x (+ 2 3)]} (* x 9))
lisp-like, simplified:	(let x (+ 2 3) (* x 9))
C~-like:	using (var x = 2+3) { return x*9; }
javascript-like:	var x = 2+3; return x*9;
Java-like:	{ int x = 2+3; return x*9; }
Haskell-like:	* x 9 \n where x = + 2 3 \n

Another option for the assignment-character is “:=” (Ada,Pascal), or “←” (indicating which way the data flows), or even something like “Expr → Id { … }” (which might make CS1 students happier — the processing happens left-to-right, just like we read the statement).

Note that you can (and should) test and write a “substitute” function w/o worrying about the exact syntax of a LetExpr. Substituting one thing in a tree for another is its own independent task, de-coupled from eval’ing a local-binding statement.

The relevent xkcd.

↩

³ Note that the different scanner(tokenizer) implementations now mean that our definition of the language is vague: in a Java implementation, NaN is a Num, and in a racket implementation it's an Id. We won't use any test cases involving such subtle differences. However, note how our choices in designing a new language are being influenced by the language we're trying to easily implement it in! This stems from the fact that a primary design constraint on X2 is that implementing an intepreter for X2 doesn't get bogged down in minutae when using either Java or Racket. ↩

⁴ The notation “let x <- 5 in ~x sub 3! ⇒ ~5 sub 3! ⇒ 2”

  eval(parse!("let x <- 5 in ~x sub 3!"))
= eval(parse!("~5 sub 3!"))
= eval(parse!("2"))

Observe how we definitely don't write “"let x <- 5 in ~x sub 3!" ⇒ "~5 sub 3!" ⇒ 2” since those strings are not .equals(·) to each other, and strings are never equal to ints. More specifically: we distinguish between “⇒” (“code evaluates to”) and “=” (“equals”, just as “=” has meant since kindergarten). ↩

⁶ For example: what if a X2 programmer uses a variable named “add” or “if+” or “in” [which we might make into a keyword in the future]? While it's not advisable for somebody to do this, and perhaps our parse should disallow this, our eval shouldn't give wacky results in this situation. ↩

⁵ All our real code should work on the parse tree itself. String-substitution (like C pre-processor macros) can't be generalized to handle shadowed variables (scope) for X3, and is in general fraught with error ⁶. A local-variable construct which requires globally-unique names isn't very impressive! ↩

This page licensed CC-BY 4.0 Ian Barland
Page last generated

Please mail any suggestions
(incl. typos, broken links)
to ibarlandradford.edu

Expr Trees eval/parse X basics; adding identifiers

Parsing IfPos in Java:

SOLUTION / what you need to do / tl;dr:

Expr Trees
eval/parse X basics; adding identifiers