Fix issue #47: syntax error points to wrong position ...

Due to the way the parser was handing nested parts of the JSON5
grammar, we would incorrectly report errors as having happened
at the start of a block of content, rather than where the error
actually happened.

A simple case of this would be reporting `"foo"bar` as happening
at column 1 rather than column 5. The bug report (#47) has a more
complicated and realistic example.

To fix this we needed to rev the underlying parser generator (glop)
to v0.6.4 and also rewrite the JSON5 grammar to use a positive
lookahead.

json5/json5.g

@@ -4,7 +4,7 @@ sp             = ws*
 
 ws             = '\u0020' | eol | comment
                | '\u0009' | '\u000B' | '\u000C' | '\u00A0' | '\uFEFF'
-               | anything:x ?( is_unicat(x, 'Zs') ) -> x
+               | ~~(anything:x ?( is_unicat(x, 'Zs') )) anything:x -> x
 
 eol            = '\u000D' '\u000A' | '\u000D' | '\u000A'
                | '\u2028' | '\u2029'

json5/parser.py

@@ -70,11 +70,13 @@ def _bind(self, rule, var):
 
     def _not(self, rule):
         p = self.pos
+        errpos = self.errpos
         rule()
         if self.failed:
             self._succeed(None, p)
         else:
             self._rewind(p)
+            self.errpos = errpos
             self._fail()
 
     def _opt(self, rule):
@@ -100,8 +102,6 @@ def _star(self, rule, vs=None):
             rule()
             if self.failed:
                 self._rewind(p)
-                if p < self.errpos:
-                    self.errpos = p
                 break
             else:
                 vs.append(self.val)
@@ -200,11 +200,21 @@ def _ws__c7_(self):
 
     def _ws__c8_(self):
         self._push('ws__c8')
-        self._seq([lambda: self._bind(self._anything_, 'x'), self._ws__c8__s1_,
+        self._seq([self._ws__c8__s0_, lambda: self._bind(self._anything_, 'x'),
                    lambda: self._succeed(self._get('x'))])
         self._pop('ws__c8')
 
-    def _ws__c8__s1_(self):
+    def _ws__c8__s0_(self):
+        self._not(lambda: self._not(self._ws__c8__s0_n_n_))
+
+    def _ws__c8__s0_n_n_(self):
+        (lambda: self._choose([self._ws__c8__s0_n_n_g__c0_]))()
+
+    def _ws__c8__s0_n_n_g__c0_(self):
+        self._seq([lambda: self._bind(self._anything_, 'x'),
+                   self._ws__c8__s0_n_n_g__c0__s1_])
+
+    def _ws__c8__s0_n_n_g__c0__s1_(self):
         v = self._is_unicat(self._get('x'), 'Zs')
         if v:
             self._succeed(v)

tests/lib_test.py

@@ -252,6 +252,21 @@ def test_whitespace(self):
         self.check(u'\u2028 1', 1)
         self.check(u'\u2029 1', 1)
 
+    def test_error_reporting(self):
+        self.check_fail('[ ,]',
+            err='<string>:1 Unexpected "," at column 3')
+
+        self.check_fail(
+            '{\n'
+            '    version: "1.0",\n'
+            '    author: "John Smith",\n'
+            '    people : [\n'
+            '        "Monty",\n'
+            '        "Python"foo\n'
+            '    ]\n'
+            '}\n',
+            err='<string>:6 Unexpected "f" at column 17')
+
 
 class TestDump(unittest.TestCase):
     def test_basic(self):