On Mon, Jan 9, 2017 at 9:27 PM, Assaf Gordon wrote: > Hello all, > >> On Jan 9, 2017, at 09:11, Norihiro Tanaka wrote: >> >> Sorry, I adjusted commit log, New patch does not change >> testsuite/Makefile.tests. >> <0001-tests-new-test-for-dfa-crash-bug.patch> > > > Paul, > Thank you for the quick fix. > > Norihiro, > Thank you for the dfa improvements and sed-tests. > > Using your example of: > printf '0123456789abcd\nx\n' | valgrind ./sed/sed 'N;s/0123456789abcd\n//' > I wasn't able to trigger the segfault (or even a valgrind warning) on sed-4.3 / x86_64. > > I suggest modifying the input just a bit, making it slightly more similar to the original bug report - with it I'm able to always reproduce the segfault: > printf "abcdefg abcdefg\nB\n" | valgrind ./sed/sed 'N;s/abcdefg.*\n//' > > What do you think ? > > I'm also considering duplicating the test - once with and once without valgrind. > Is this warranted or an overkill ? > > Attach patch contains updated tests (and slightly modified git-comment). > The second commit updates gnulib (comes after adding the tests just temporarily, to make testing before/after gnulib update easier). Hi Assaf, Thank you for adjusting the tests and commit log. Those look fine. The only problem is that the new newline-valgrind.sh test would fail when run against an ASAN-enabled sed. That is because valgrind just doesn't work when the binary is ASAN-enabled. So I have extemded init.cfg's require_valgrind_ function so that it also detects this case and skips the test. I will push the attached shortly, after which you are welcome to push your commits.