On 31 Aug 2016, at 19:03, Doug Barton <do...@dougbarton.us> wrote: > > August 31, 2016 12:37 AM, "Bernard Spil" <br...@freebsd.org> wrote: >> On 2016-08-30 17:45, Doug Barton wrote: >> >>> Since the last commit to the mariadb10* ports the resulting binaries >>> dump core on 9.3-RELEASE-p43, even for a simple --help option. I am >>> assuming that is not the intended result .... >>> Doug >> >> Hi Doug, >> >> Nope, that is not intended. Assume this is the 10.0.27 update? I did build >> them on 9.3 but have not >> checked the resulting binaries... Will need some time to investigate! >> >> https://brnrd.eu/poudriere/build.html?mastername=93amd64-svn&build=2016-08-29_20h14m28s > > Thanks Bernard. > > It turns out that the actual problem was binutils. I backed up several > revisions of mariadb, past the previous working version, and had the same > symptoms. Then I started on the deps that had changed at the same time, and > fortunately I guessed right on the first one. > > Backing up to -r{2016-08-25} on binutils got mariadb working with the latest > version.
Bisecting showed this was introduced in binutils commit 26e3a0c [1], which turns on .init_array/.fini_array support by default. It makes most C++ executables crash at startup, on 9.x, since support for this was not fully merged back to stable/9. I think Baptiste is working on a fix. It is probably best to configure binutils on 9.x with --disable-initfini-array. -Dimitry [1] https://sourceware.org/git/gitweb.cgi?p=binutils-gdb.git;a=commit;h=26e3a0c9ba4a8376fdf9f898637919d144d8b1d8
signature.asc
Description: Message signed with OpenPGP using GPGMail
