[bind10-dev] failed unittests

[JINMEI Tatuya / 神明達哉]

At Mon, 05 Dec 2011 10:23:53 +0100,
Jelte Jansen <jelte at isc.org> wrote:

> It's not. Or at least, not just that.
> 
> Just looked at that machine, the problem is that the test script needs
> some sqlite3 itself, and it does import sqlite3.
> 
> That in itself would be fine, however, since ~jreed/opt/pkg is *not* in
> LD_LIBRARY_PATH, python will load the system-install libsqlite3.
> 
> That is *also* not a problem (the python wrappers work fine with the old
> version), but after we load our datasource wrappers, and the dynamic
> library loader sees that we need libsqlite3 (with the ~jreed/opt/pkg
> hint), and will reuse the one it loaded already. And then we get missing
> symbols.
> 
> So we need to either add such custom paths to the LD_LIBRARY_PATH, or
> even simpler, move the import sqlite3 line in the test down a few lines
> (so our libs are hit first)

Hmm, either option doesn't seem to be very attractive especially for
some not-so-usual situation like someone trying to use locally built
libsqlite3.  But from a quick look there are only very few cases that
require both the higher level datasrc module and the sqlite3 module,
so I guess the best possible compromise is to change the import order
for these specific cases (with a comment of why).

---
JINMEI, Tatuya