[sferik]

On JRuby, errors raised while decoding a string had no position:

require 'json'
JSON.parse(%([1,\n"\\q"]))
# MRI:   invalid escape character in string: '\q"]' at line 2 column 2 (line: 2, column: 2)
# JRuby: invalid escape character in string: q (line: nil, column: nil)

The same went for incomplete \u escapes, broken surrogate pairs and raw control characters. StringDecoder built its own ParserError, so the parser's parseError, which adds the line, column, message suffix and json_path, was never used for these.

StringDecoder now throws a DecodeError with the message and the offset to report, and the parser turns it into a regular parse error at that offset. The offsets follow parser.c:

• errors in an escape sequence are reported at its backslash
• surrogate errors are reported at the start of the text after the previous escape (p in json_string_unescape)
• raw control characters are reported where they are

The messages keep their current wording, except that surrogate error snippets now start at the reported offset, as in C. Porting the wording belongs with the other error message differences in #1123.

Remaining differences, left for later:

• When a string has both a bad escape and a later raw control character, C reports the control character, because it scans the whole string first. Java reports whichever comes first.
• Invalid UTF-8 is still rejected by the Java parser (C and Pure parser accept invalid UTF-8 strings, the Java parser doesn't. #138).

[headius]

Quick review on mobile looks good. I will approve after a more rigorous review shortly.