During the last assignment you probably encountered a wide variety of error messages. The nature of these messages depends on both the language definition and the compiler or interpreter. You may have noticed that across languages and implementations these messages differ greatly in their usefulness and specificity. One feature common to all of the languages you used is syntax error recovery. In the simplest sense, syntax error recovery is the mechanism by which a compiler or interpreter continues to parse a program (and find more syntax errors) after it encounters an instance of invalid syntax.
Your task in this assignment is to implement a front end, in Rust, that incorporates syntax error recovery and that generates syntax trees for an extended version of the calculator language discussed in the text and in lecture. We are providing starter code with input buffering routines, an ad-hoc scanner, and the driver for a table-driven parser. The scanner implements simple recovery from lexical errors (for the base calculator language), but the parser currently panics (prints an error message and halts) if it encounters a syntax error. We are also providing a parser generator that takes a CFG as input and generates LL(1) parsing tables in the form of initialized Rust data structures. Starting from this initial code base, you must:
if
/elsif
/else
statements, and
while
/do
statements, as shown in the
grammar below.
The code we are giving you already handles three error cases:
unrecognized characters at the start of a token, anything
other than ‘=
’ following
‘:
’, or a letter immediately following an
integer constant. You will need to extend this to handle, in a
similar fashion, the two-character
‘<>
’
and
‘==
’ signs and
the fraction and exponent parts of real-number constants.
***
comments.
attributes.rs
) to
represent declarations, if
and while
statements, and trunc
and float
expressions.
These can be patterned after node types for the base calculator
language, which we are giving you.
match
statement in
actions.rs
. Their numbers (match
labels) must correspond to brace-enclosed action routine markers in
the CFG you give to the parser generator.
The code we are giving you has a few action routines already
present, to give you an idea what they look like. You will
need to implement more. (Our sample solution has a total of
just over 40 action routines, many of which are simple “copy
rules.”)
Each tree node will be represented either as a (parenthesized) tuple,
with a fixed number of fields, or as a (square-bracketed) list, with an
arbitrary number of elements.
In a tuple, you can think of the first field as a subtree parent and
the rest of the fields as its children.
As an simple example, the tree for the sum-and-average program we
used as an example in lecture (in the base calculator language)
would be output as [ (read (a)) (read (b)) (:= (sum) (+
(a) (b))) (write (sum)) (write (/ (sum) (2))) ]
.
More detail on the AST structure can be found
below.
As a debugging aid, the code we are giving you prints a trace of predictions, matches, and action routine invocations. You should disable this trace in the code you hand in.
When run, your program should read an extended calculator program from standard input and then print, to standard output, either syntax error messages or a correct syntax tree. (It’s ok for lexical errors not to inhibit construction of the AST, if the scanner’s repairs are acceptable to the parser.)
The initial source code for this assignment is available as a
“tarball” file.
You should download it on a csug
machine and unpack it
from the shell:
tar -xzf calc_parse.tgzThen
cd
into the calc_parse
directory and run
cargo initThat will initialize a standard Rust project, with dependence tracking and
git
-based version control.
In the src
subdirectory, file tables.rs
has
been automatically generated from the calc_gram.txt
file in
the root directory. When you make changes to the grammar (to add
features and/or action routines), you must re-run the parser generator:
table_gen < calc_gram.txt > src/tables.rsThe
table_gen
tool resides in ~cs254/bin,
which you should already have put on your shell's search path during
assignment 1.
If you’re curious how the parser generator works (or if you want
to download it so you can run it on your own machine—note that it’s
written in OCaml), you can find it in HERE.
The binary is really all you need, however.
If you’re a Rust enthusiast, you’re welcome to automate the
table_gen
step, so the
cargo build
command will automatically
regenerate tables.rs
whenever calc_gram.txt
has changed. If you don’t know how to do that, don’t
worry: regenerating the file explicitly is fine.
To invoke the Rust compiler, type
cargo buildThis will recompile all and only the files that have changed since your last rebuild. (For Unix hackers:
cargo
subsumes the functionality of
traditional make
files.)
You can run the executable version of your code directly:
./target/debug/calc_parseBetter, type
cargo runThat will rebuild your code if necessary and then run the executable.
As you work on the code, I strongly recommend that you make use
of the git
support established by
cargo init
.
At frequent intervals, type
git commit -am 'log message'If you add any files to the project, type
git add 'filename'before your next commit. If you run into trouble and need to back out recent changes, consult any decent
git
tutorial on the web.
If you’d like to test the behavior/output of your code, our sample
solution (without source!) can be found at
~cs254/bin/calc_parse
. It incorporates immediate
error detection.
If you find a bug in the starter code, the parser generator, or the sample solution, please email the instructor and grad TA, so we can fix it as quickly as possible.
Here is an LL(1) grammar for the extended calculator language:
SL → S SL | ε S → TP id :=
E |read
TPid
|write
E
|if
Cthen
SL ELend
|while
Cdo
SLend
TP → int
|real
| εEL → elsif
Cthen
SL EL |else
SL | εC → E RO E E → T TT TT → AO T TT | ε T → F FT FT → MO F FT | ε F → (
E)
|id
|i_lit
|r_lit
|trunc (
E)
|float
(
E)
RO → ==
|<>
|<
|>
|<=
|>=
AO → +
|-
MO → *
|/
Here the new nonterminal C represents a comparison. The new nonterminal RO is a “relational operator.”
Integer and real number constants are differentiated by the presence or absence of a decimal point:
wherei_lit
=d
+
r_lit
= (d
+.
d
* |d
*.
d
+ ) (e
(+
|-
| ε )d
+ | ε )
d
stands for any decimal digit and e
is
the actual letter e
.
Note that digits are required on both sides of the decimal point.
As explained in lecture, $$
is a special token created by
the scanner when it detects the end of the input; it
is not a part of the actual program text.
(In the starter code we have given you, it is TokTp::Stop
.)
We have left the augmenting production
P → SL $$
out of the grammar above. It will be inserted for you
automatically by the parser generator.
In the code we are giving you, the file calc_gram.txt
looks
like this:
SL -> S SL -> S -> Id Gets E {3} -> Read Id -> Write E {5} E -> T {7} TT T -> F {11} FT TT -> AO T TT -> FT -> MO F FT -> AO -> Plus -> Minus MO -> Times -> DivBy F -> Id {13} -> Num {17} -> LParen E RParen {19}The start symbol is the left-hand side of the first production. The augmenting production, as noted above, is added by the parser generator,
table_gen
.
All productions for a given nonterminal must be consecutive, with the
left-hand side elided on all but the first.
Epsilon productions simply have an empty right-hand size.
Every symbol’ name must have the form of a valid Rust
enum
constant (this facilitates generation of
tables.rs
).
Your scanner should recognize the actual characters (e.g.,
“:=
”, not “Gets
”),
however. It should return both the enum
constant and
the text of the token as part of its Token
-typed return
value.
Action routines take the form of natural numbers in braces (e.g.,
“{13}
”. When the parser comes upon one in
the process of parsing a right-hand side, it will call routine
do_action
, passing the natural number (e.g., 13) as
parameter. Code for action routines thus take the form of
match
statement arms in actions.rs
.
Action routine numbers don’t have to be consecutive, or appear in
any particular order.
The initial calc_gram.txt
file has seven routines for
illustration purposes only. You will need to replace most of
these in your code.
Action routine numbers 0 and 1, and symbol names
Goal
and Stop
, are reserved by
table_gen
for use in the augmenting production; you
can’t use these yourself.
Unless you choose to do so for extra credit, you will not
be checking semantic ruiles in the current assignment.
If you’re interested, however,
identifiers are intended to be declared before use, with an
int
or real
type specifier on
a read
or assignment statement that provides an initial
value.
The scope of each declaration extends from the declaration itself
through the end of the current statement list.
Integers and real numbers are not intended to be mixed in expressions
unless explicitly converted with
trunc
and float
.
As it turns out, if we assume that integers are unbounded, our extensions make the calculator language Turing complete (if still quite impractical). As an illustration, here is a program that calculates the first n primes:
read int n int cp := 2 while n > 0 do int found := 0 int cf1 := 2 int cf1s := cf1 * cf1 while cf1s <= cp do int cf2 := 2 int pr := cf1 * cf2 while pr <= cp do if pr == cp then found := 1 end cf2 := cf2 + 1 pr := cf1 * cf2 end cf1 := cf1 + 1 cf1s := cf1 * cf1 end if found == 0 then write cp n := n - 1 end cp := cp + 1 end
Your AST for the primes-printing program should look like this:
[ (decl (n) (int)) (read (n)) (decl (cp) (int)) (:= (cp) (2)) (while (> (n) (0)) [ (decl (found) (int)) (:= (found) (0)) (decl (cf1) (int)) (:= (cf1) (2)) (decl (cf1s) (int)) (:= (cf1s) (* (cf1) (cf1))) (while (<= (cf1s) (cp)) [ (decl (cf2) (int)) (:= (cf2) (2)) (decl (pr) (int)) (:= (pr) (* (cf1) (cf2))) (while (<= (pr) (cp)) [ (if (== (pr) (cp)) [ (:= (found) (1)) ] [ ] ) (:= (cf2) (+ (cf2) (1))) (:= (pr) (* (cf1) (cf2))) ] ) (:= (cf1) (+ (cf1) (1))) (:= (cf1s) (* (cf1) (cf1))) ] ) (if (== (found) (0)) [ (write (cp)) (:= (n) (- (n) (1))) ] [ ] ) (:= (cp) (+ (cp) (1))) ] ) ]Indentation and line breaks are shown here for clarity only, and need not be generated by your code. Otherwise, the output above is what our sample solution produces, using (extensions of) the
fmt
routines we’ve given you in attributes.rs
.
As noted near the top of these instructions, square brackets delimit
lists, which have an arbitrary number of elements. Parentheses
delimit tuples or structs, which have a fixed number of fields. An
if
node, for example, has three children: a condition, a
body for its then
clause, and a body for its
else
clause. The condition is a tuple containing a
comparison operator and its operands; the body is a list of statements
that should be executed when the comparison is true or false,
respectively. The program as a whole is likewise a statement
list.
Conditional statements with elsif
clauses should produce
the same AST as equivalent nested if
statements.
So, for example, the input if a < b then write c elsif d < e then
write f else write g end
should yield the tree
[ (if (< (a) (b)) [ (write (c)) ] [ (if (< (d) (e)) [ (write (f))
] [ (write (g)) ]) ]) ]
.
A note for those new to the Linux command line: If you paste characters into the terminal window as standard input, you have to hit control-D to indicate end-of-file before the generator will do anything. You will almost certainly want to put your sample calculator programs into text files and feed them into your code using shell indirection:
calc_parse < test_file1.txt
Students in 454 must implement immediate error detection: epsilon productions should be “undone” before recovering when the upcoming token turns out not to be in the local FOLLOW set of the top-of-stack nonterminal.
You will want to test your code on a variety of calculator programs,
both correct and incorrect.
Your turn-in should include the test programs you used, and your
README
file should explain how to run them.
(We will of course run additional tests of our own.)
Extra credit may be given to students who provide particularly well
designed test mechanisms in their submission.
Note that your code will employ both insertions and deletions:
when the match routine (function eat
in the starter
code) sees a token other than the one it expects,
it will print an appropriate error message and then return,
as if it had inserted and then matched the
expected token, leaving the remaining token stream unchanged.)
When the main loop of the parser encounters an Err
entry
in the parse (prediction) table, it will delete tokens until it finds
something in either the FIRST set or the
FOLLOW set of the top-of-stack nonterminal.
A variety of good resources for Rust are available on the web. You may wish to consult the following:
As in other assignments this semester, you may work alone or in teams of two. If you would like to work on a team but are in need of a partner, consider posting a note to the Blackboard discussion board. If you are working in a team, a reasonable division of work would be for one student to update the scanner and implement error recovery and the other to build the AST. If you’re implementing immediate error recovery, scanner updates might better be assigned to the person building the AST.
Be sure to follow all the rules on the Grading page. As in all assignments,
use the turn-in script:
~cs254/bin/TURN_IN
on the csug
machines.
Put your write-up in a
README.txt
or README.pdf
file in the directory in
which you run the script. Be sure to describe any features
of your code that the TAs might not immediately notice.
Only one turn-in of the main assignment (and only one
README
) is required per team,
but each student must complete the trivia (on Blackboard) separately.
float
is always of type
int
;
(e) the argument of trunc
is always of type
real
.
for
loops, or subroutines.
Before class on Thursday, February 13, complete the T2 trivia assignment found on Blackboard.