Chromium Code Reviews| Index: utils/peg/pegparser_test.dart |
| diff --git a/utils/peg/pegparser_test.dart b/utils/peg/pegparser_test.dart |
| new file mode 100644 |
| index 0000000000000000000000000000000000000000..a5da0cafa3b0df14b26b9b2b104bc56493c8b059 |
| --- /dev/null |
| +++ b/utils/peg/pegparser_test.dart |
| @@ -0,0 +1,343 @@ |
| +#!/usr/bin/env ./dart_bin |
| +#import('pegparser.dart'); |
| + |
| +testParens() { |
|
Jennifer Messerly
2011/10/28 01:21:10
tests, awesome :)
|
| + Grammar g = new Grammar(); |
| + Symbol a = g['A']; |
| + |
| + a.def = ['(', MANY(a, min:0), ')', (a) => a]; |
| + |
| + check(g, a, "", null); |
| + check(g, a, "()", '[]'); |
| + check(g, a, "(()())", '[[],[]]'); |
| + check(g, a, "(()((()))())", '[[],[[[]]],[]]'); |
| +} |
| + |
| +testBlockComment() { |
| + |
| + // Block comment in whitespace. |
| + |
| + Grammar g = new Grammar(); |
| + Symbol blockComment = g['blockComment']; |
| + |
| + blockComment.def = |
| + ['/*', |
| + MANY(OR([blockComment, |
| + [NOT('*/'), CHAR()], |
| + [END, ERROR('EOF in block comment')] |
| + ]), |
| + min: 0), |
| + '*/']; |
| + print(blockComment); |
| + |
| + var a = MANY(TEXT('x')); |
| + |
| + g.whitespace = OR([g.whitespace, blockComment]); |
| + |
| + check(g, a, "x /**/ x", '[x,x]'); |
| + check(g, a, "x /*/**/*/ x", '[x,x]'); |
| + check(g, a, "x /*/***/ x", 'EOF in block comment'); |
| + check(g, a, "x /*/*/x**/**/ x", '[x,x]'); |
| + |
| + check(g, a, @""" |
| +/* Comment */ |
| +/* Following comment with /* nested comment*/ */ |
| +x |
| +/* x in comment */ |
| +x /* outside comment */ |
| +""", |
| + '[x,x]'); |
| +} |
| + |
| +testTEXT() { |
| + Grammar g = new Grammar(); |
| + |
| + // TEXT grabs the parsed text, |
| + check(g, TEXT(LEX(MANY(OR(['1','a'])))), ' 1a1 ', '1a1'); |
| + |
| + // Without the lexical context, TEXT will grab intervening whitespace. |
| + check(g, TEXT(MANY(OR(['1','a']))), ' 1a1 ', '1a1'); |
| + check(g, TEXT(MANY(OR(['1','a']))), ' 1 a 1 ', '1 a 1'); |
| + |
| + // Custom processing of the TEXT substring. |
| + var binaryNumber = |
| + TEXT(LEX(MANY(OR(['0','1']))), |
| + (str, start, end) { |
| + var r = 0; |
| + var zero = '0'.charCodeAt(0); |
| + for (int i = start; i < end; i++) |
| + r = r * 2 + (str.charCodeAt(i) - zero); |
| + return r; |
| + }); |
| + |
| + check(g, binaryNumber, ' 10101 ', 21); |
| + check(g, binaryNumber, '1010111', 87); |
| + check(g, binaryNumber, '1010 111', null); |
| +} |
| + |
| +testOR() { |
| + // OR matches the first match. |
| + Grammar g = new Grammar(); |
| + check(g, OR([['a', NOT(END), () => 1], |
| + ['a', () => 2], |
| + ['a', () => 3]]), |
| + 'a', 2); |
| +} |
| + |
| +testCODE() { |
| + Grammar g = new Grammar(); |
| + var a = TEXT(LEX('thing', MANY(CHAR('bcd')))); |
| + |
| + check(g, a, 'bbb', 'bbb'); |
| + check(g, a, 'ccc', 'ccc'); |
| + check(g, a, 'ddd', 'ddd'); |
| + check(g, a, 'bad', null); // a is outside range. |
| + check(g, a, 'bed', null); // e is outside range. |
| +} |
| + |
| +testC() { |
| + // Curried tree builders. |
| + binary(operation) => (second) => (first) => [operation, first, second]; |
| + unary(operation) => () => (first) => [operation, first]; |
| + reform(a, fns) { |
| + var r = a; |
| + for (var fn in fns) |
| + r = fn(r); |
| + return r; |
| + } |
| + |
| + Grammar g = new Grammar(); |
| + |
| + Symbol expression = g['expression']; |
| + Symbol postfix_e = g['postfix_e']; |
| + Symbol unary_e = g['unary_e']; |
| + Symbol cast_e = g['cast_e']; |
| + Symbol mult_e = g['mult_e']; |
| + Symbol add_e = g['add_e']; |
| + Symbol shift_e = g['shift_e']; |
| + Symbol relational_e = g['relational_e']; |
| + Symbol equality_e = g['equality_e']; |
| + Symbol cond_e = g['cond_e']; |
| + Symbol assignment_e = g['assignment_e']; |
| + |
| + // Lexical elements. |
| + var idStartChar = CHAR( |
| + @"ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz"); |
| + var idNextChar = CHAR( |
| + @"ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789$_"); |
| + |
| + var id = TEXT(LEX('identifier', [idStartChar, MANY(idNextChar, min: 0)])); |
| + |
| + var lit = TEXT(LEX('literal', MANY(CHAR('0123456789')))); |
| + |
| + |
| + var type_name = id; |
| + |
| + |
| + // Expression grammar. |
| + var primary_e = OR([id, |
| + lit, |
| + ['(', expression, ')', (e) => e] |
| + ]); |
| + |
| + var postfixes = OR([['(', MANY(assignment_e, ',', 0), ')', binary('apply')], |
| + ['++', unary('postinc')], |
| + ['--', unary('postdec')], |
| + ['.', id, binary('field')], |
| + ['->', id, binary('ptr')], |
| + ]); |
| + |
| + postfix_e.def = [primary_e, MANY(postfixes, min:0), reform]; |
| + |
| + |
| + var unary_op = OR([['&', () => 'address'], |
| + ['*', () => 'indir'], |
| + ['!', () => 'not'], |
| + ['~', () => 'not'], |
| + ['-', () => 'negate'], |
| + ['+', () => 'uplus'], |
| + ]); |
| + var sizeof = LEX('sizeof', ['sizeof', NOT(idNextChar)]); |
| + |
| + Symbol unary_e_plain = g['unary_e_plain']; |
| + unary_e_plain.def = |
| + OR([ ['++', unary_e, (e) => ['preinc', e]], |
| + ['--', unary_e, (e) => ['predec', e]], |
| + [unary_op, cast_e, (o, e) => [o, e]], |
| + [sizeof, unary_e, (e) => ['sizeof-expr', e]], |
| + [sizeof, '(', type_name , ')', (t) => ['sizeof-type', t]], |
| + postfix_e |
| + ]); |
| + |
| + unary_e.def = MEMO(unary_e_plain); |
| + //unary_e.def = unary_e_plain; |
| + |
| + cast_e.def = OR([ ['(', type_name, ')', cast_e, (t, e) => ['cast', t, e]], |
| + unary_e, |
| + ]); |
| + |
| + var mult_ops = OR([['*', cast_e, binary('mult')], |
| + ['/', cast_e, binary('div')], |
| + ['%', cast_e, binary('rem')], |
| + ]); |
| + mult_e.def = [cast_e, MANY(mult_ops, min:0), reform]; |
| + |
| + var add_ops = OR([['+', mult_e, binary('add')], |
| + ['-', mult_e, binary('sub')], |
| + ]); |
| + add_e.def = [mult_e, MANY(add_ops, min:0), reform]; |
| + |
| + var shift_ops = OR([['>>', add_e, binary('shl')], |
| + ['<<', add_e, binary('shr')], |
| + ]); |
| + shift_e.def = [add_e, MANY(shift_ops, min:0), reform]; |
| + |
| + var relational_ops = OR([['<=', shift_e, binary('le')], |
| + ['>=', shift_e, binary('ge')], |
| + ['<', shift_e, binary('lt')], |
| + ['>', shift_e, binary('gt')], |
| + ]); |
| + relational_e.def = [shift_e, MANY(relational_ops, min:0), reform]; |
| + |
| + |
| + var equality_ops = OR([['==', shift_e, binary('eq')], |
| + ['!=', shift_e, binary('ne')], |
| + ]); |
| + equality_e.def = [relational_e, MANY(equality_ops, min:0), reform]; |
| + |
| + |
| + var bit_and_op = LEX('&', ['&', NOT('&')]); // Don't see '&&' and '&', '&' |
| + var bit_or_op = LEX('|', ['|', NOT('|')]); |
| + |
| + var and_e = [equality_e, MANY([bit_and_op, equality_e, binary('bitand')], min:0), reform]; |
| + var xor_e = [and_e, MANY(['^', and_e, binary('bitxor')], min:0), reform]; |
| + var or_e = [xor_e, MANY([bit_or_op, xor_e, binary('bitor')], min:0), reform]; |
| + |
| + var log_and_e = [or_e, MANY(['&&', or_e, binary('and')], min:0), reform]; |
| + |
| + var log_or_e = [log_and_e, MANY(['||', log_and_e, binary('or')], min:0), reform]; |
| + |
| + //cond_e.def = OR([ [log_or_e, '?', expression, ':', cond_e, |
| + // (p,a,b) => ['cond', p, a, b]], |
| + // log_or_e]); |
| + // Alternate version avoids reparsing log_or_e. |
| + cond_e.def = [log_or_e, MAYBE(['?', expression, ':', cond_e]), |
| + (p, r) => r == null || r == false ? p : ['cond', p, r[0], r[1]]]; |
| + |
| + var assign_op = OR([['*=', () => 'mulassign'], |
| + ['=', () => 'assign']]); |
| + |
| + // TODO: Figure out how not to re-parse a unary_e. |
| + // Order matters - cond_e can't go first since cond_e will succeed on, e.g. 'a'. |
| + assignment_e.def = OR([[unary_e, assign_op, assignment_e, |
| + (u, op, a) => [op, u, a]], |
| + cond_e]); |
| + |
| + expression.def = [assignment_e, |
| + MANY([',', assignment_e, binary('comma')], min:0), |
| + reform]; |
| + |
| + show(g, expression, 'a'); |
| + check(g, expression, 'a', 'a'); |
| + check(g, expression, '(a)', 'a'); |
| + check(g, expression, ' ( ( a ) ) ', 'a'); |
| + |
| + check(g, expression, 'a(~1,2)', '[apply,a,[[not,1],2]]'); |
| + check(g, expression, 'a(1)(x,2)', '[apply,[apply,a,[1]],[x,2]]'); |
| + check(g, expression, 'a(1,2())', '[apply,a,[1,[apply,2,[]]]]'); |
| + |
| + check(g, expression, '++a++', '[preinc,[postinc,a]]'); |
| + check(g, expression, 'a++++b', null); |
| + check(g, expression, 'a++ ++b', null); |
| + check(g, expression, 'a+ +++b', '[add,a,[preinc,[uplus,b]]]'); |
| + check(g, expression, 'a+ + ++b', '[add,a,[uplus,[preinc,b]]]'); |
| + check(g, expression, 'a+ + + +b', '[add,a,[uplus,[uplus,[uplus,b]]]]'); |
| + check(g, expression, 'a+ ++ +b', '[add,a,[preinc,[uplus,b]]]'); |
| + check(g, expression, 'a++ + +b', '[add,[postinc,a],[uplus,b]]'); |
| + check(g, expression, 'a+++ +b', '[add,[postinc,a],[uplus,b]]'); |
| + |
| + check(g, expression, '((T)f)(x)', '[apply,[cast,T,f],[x]]'); |
| + check(g, expression, '(T)f(x)', '[cast,T,[apply,f,[x]]]'); |
| + |
| + check(g, expression, 'a++*++b', '[mult,[postinc,a],[preinc,b]]'); |
| + |
| + check(g, expression, 'a<<1>>++b', '[shl,[shr,a,1],[preinc,b]]'); |
| + |
| + check(g, expression, 'a<1&&b', '[and,[lt,a,1],b]'); |
| + |
| + check(g, expression, 'a<1 & &b', '[bitand,[lt,a,1],[address,b]]'); |
| + check(g, expression, |
| + 'a ? b ? c : d : e ? f : g', |
| + '[cond,a,[cond,b,c,d],[cond,e,f,g]]'); |
| + |
| + check(g, expression, 'a,b,c', '[comma,[comma,a,b],c]'); |
| + check(g, expression, 'a=1,b,c', '[comma,[comma,[assign,a,1],b],c]'); |
| + |
| + check(g, expression, |
| + '((((((((((((a))))))))))))=1,b,c', '[comma,[comma,[assign,a,1],b],c]'); |
| + |
| + check(g, expression, 'sizeof a', '[sizeof-expr,a]'); |
| + check(g, expression, 'sizeofa', 'sizeofa'); |
| + check(g, expression, 'sizeof (a)', '[sizeof-expr,a]'); |
| +} |
| + |
| + |
| +show(grammar, rule, input) { |
| + print('show: "$input"'); |
| + var ast; |
| + try { |
| + ast = grammar.parse(rule, input); |
| + } catch (var exception) { |
| + if (exception is ParseError) |
| + ast = exception; |
| + else |
| + throw; |
| + } |
| + print('${printList(ast)}'); |
| +} |
| + |
| +void check(grammar, rule, input, expected) { |
| + // If [expected] is String then the result is coerced to string. |
| + // If [expected] is !String, the result is compared directly. |
| + print('check: "$input"'); |
| + var ast; |
| + try { |
| + ast = grammar.parse(rule, input); |
| + } catch (var exception) { |
| + ast = exception; |
| + } |
| + |
| + var formatted = ast; |
| + if (expected is String) |
| + formatted = printList(ast); |
| + |
| + Expect.equals(expected, formatted, "parse: $input"); |
| +} |
| + |
| +// Prints the list in [1,2,3] notation, including nested lists. |
| +void printList(item) { |
| + if (item is List) { |
| + StringBuffer sb = new StringBuffer(); |
| + sb.add('['); |
| + var sep = ''; |
| + for (var x in item) { |
| + sb.add(sep); |
| + sb.add(printList(x)); |
| + sep = ','; |
| + } |
| + sb.add(']'); |
| + return sb.toString(); |
| + } |
| + if (item == null) |
| + return 'null'; |
| + return item.toString(); |
| +} |
| + |
| +main() { |
| + testCODE(); |
| + testParens(); |
| + testOR(); |
| + testTEXT(); |
| + testBlockComment(); |
| + testC(); |
| +} |