On 27/09/16 22:14, lmb@cloudflare.com wrote:

On 26 September 2016 at 20:48, Hallvard Breien Furuseth h.b.furuseth@usit.uio.no wrote:

...

I think we can skip sigwait() too since the thread will exit when done. (...)

That's what my initial patch did. At least on OS X this leads to the process receiving SIGPIPE and dying.

OK. I've put a fixed version including that comment in branch "mdb/its8504-sigpipe" @ git://git.uio.no/u/hbf/openldap.git. We can squash the commits later:

Never clear mc_error, we'd lose failure in the other thread. It's not mutex-protected, which is OK: If both threads set it, we'll get one of the errors. LMDB already expects atomic int.

And.. duh, we've forgotten mdb_copy without MDB_CP_COMPACT. I'll do it later if nobody beats me to it. Test: Comment out SIGPIPE in mdb_copy.c, then run

bash$ (sleep 1; ./mdb_copy testdb; echo exit $? >&2) | true exit 141

Exit values >= 128 are aborts from signals.