bug#22655: grep-2.21 (and git master): --null-data and ranges work in an odd way (-P works fine)

[Ulya Fokanova]

I've explored the following case:

   $ printf '12\n34\0' | LC_ALL=en_US.utf-8 grep -z '^[1-4]*$' | wc -c
   6

It's a bug (there should be no match).

This is what grep does:

 * triesto build DFA (as indfa.c)
 * fails to expand character range [1-4] because of multibyte
   localeen_US.utf-8 and gives up building DFA(marks [1-4] as BACKREF
   that suppressesall dfa.c-related code), note the difference with
   [1234] casein whichthere's no need to expand multibyte range
 * falls back to Regex (gnulib extension of regex.h)
 * Regex doesn't support '-z'semantics(the closest configuration to
   '-z' is RE_NEWLINE_ALT, which is already included in RE_SYNTAX_GREP
   set), so '\n'is treated as newline and match erroneously succeeds

I think this should be worked around in grep: before calling 're_search' 
it should split the input string by 'eolbyte'.

The bug also present with PCRE engine:

   $ printf '12\n34\0' | LC_ALL=en_US.utf-8 grep -z -P '^[1234]*$' | wc -c
   6
   $ printf '12\n34\0' | LC_ALL=en_US.utf-8 grep -z -P '^[1-4]*$' | wc -c
   6

Ulya