Re: LMDB test failures under musl on mips

[Szabolcs Nagy <nsz@port70.net>]

* Martin Lucina <martin@lucina.net> [2014-02-14 10:31:56 +0100]:
> That's what I get, and also these:
> 
> mdb.c:5176: Assertion 'IS_LEAF(mp)' failed in mdb_cursor_next()
> 
> or
> 
> mdb.c:1713: Assertion 'rc == 0' failed in mdb_page_dirty()
> 
> etc.
> 
> mtest is somewhat fickle, it uses random() to decide exactly what it's
> doing. I have a hunch that I can provoke this with a simpler test program,
> going to try that now.

i removed the srandom(time(NULL)) and disabled ASLR and it's still fickle
i haven't looked further

> 
> Do you still want those strace logs from me?
> 

no, i think strace does not help here
(at least i didnt see anything obvious)

i don't quite understand the nondeterministic behaviour

it seems to do reads/writes and mmap through two different fds to the same
underlying file, but it does fdatasync on one and O_SYNC on the other
so i think the behaviour should be deterministic
(i'd need to know more about mdb and see the mmap accesses as well
to figure out what's going on..)

> Both of the targets (ASUS RT-N66u running Tomato, TP-Link TL-WDR4300
> running OpenWRT trunk) I tried have 4k page size, so nothing out of the
> ordinary there.
> 

i tried it on a wrt160nl with old openwrt image
(Atheros AR9130 cpu)