Always copy the current line string out of the token buffer when generating an error report, rather than just passing the token buffer itself. The token buffer wasn't necessarily a well-terminated string, so displaying the contents of the string in the error report produced unexpected results.
The unicode string in the error report is owned by a JSString; this string is rooted for the (stack-based) lifetime of the error report.
Fix courtesy jband.
r=mccabe
a=beard
git-svn-id: svn://10.0.0.236/trunk@76035 18797224-902f-48f8-a5cc-f745e15eee43
e147d2735e