Parsing the regexp /\A{/ causes uses an uninitialized value because it
tries to parse it as a range quantifier, so it reads the character after
the closing curly bracket. This is using uninitialized values because
prism strings are not null terminated. This can be seen in the Valgrind
output:
==834710== Conditional jump or move depends on uninitialised value(s)
==834710== at 0x5DA010: pm_regexp_parse_range_quantifier (regexp.c:163)
==834710== by 0x5DA010: pm_regexp_parse_quantifier (regexp.c:243)
==834710== by 0x5DAD69: pm_regexp_parse_expression (regexp.c:738)
==834710== by 0x5DAD69: pm_regexp_parse_pattern (regexp.c:761)
==834710== by 0x5DAD69: pm_regexp_parse (regexp.c:773)
==834710== by 0x5A2EE7: parse_regular_expression_named_captures (prism.c:20886)
==834710== by 0x5A2EE7: parse_expression_infix (prism.c:21388)
==834710== by 0x5A5FA5: parse_expression (prism.c:21804)
==834710== by 0x5A64F3: parse_statements (prism.c:13858)
==834710== by 0x5A9730: parse_program (prism.c:22011)
==834710== by 0x576F0D: parse_input_success_p (extension.c:1062)
==834710== by 0x576F0D: parse_success_p (extension.c:1084)
This commit adds checks for the end of the string to
pm_regexp_parse_range_quantifier.
https://github.com/ruby/prism/commit/be6cbc23ef