Sed incorrectly matches the anchor '^' to a newline character
(ASCII 10) despite using NUL line terminators (-z).
See https://bugs.gnu.org/24615 .
Previous (incorrect) behavior:
$ printf "a\0b\0" | sed -z 'N;s/^/X/mg;' | od -An -a
X a nul b nul
With this fix:
$ printf "a\0b\0" | sed -z 'N;s/^/X/mg;' | od -An -a
X a nul X b nul
Current regex engine can not handle NUL multiline anchors.
Implement such search manually (by breaking the pattern into separate
NUL-terminated lines) and searching each one individually.
* NEWS: Mention it.
* sed/regexp.c (compile_regex_1): Don't use newline_anchor of regex, if
the buffer delimiter is not newline character.
(match_regex): Do above case line-by-line.
* testsuite/zero-anchor.good, testsuite/zero-anchor.inp,
testsuite/zero-anchor.sed: New test.
* testsuite/local.mk: Add the test.
* testsuite/Makefile.tests: Add the test.
* bootstrap.conf: Add memrchr.
* lib/.gitignore, m4/.gitignore: Ignore memrchr files.