16690 breaks `make check` [WAS Re: AUDIT: r16690 - gnucash/trunk/src/engine - Daniel Harding's update to Afghani currency. closes #504257]

[Derek Atkins]

Andrew Sackville-West <andrew at swclan.homelinux.org> writes:

>> But interestingly, Derek reverted 16619 and that causes make check to
>> succeed. Seems odd to me that adjusting a currency would cause 16619
>> to cause a test to break. So maybe there is more going on here than
>> meets the eye. 
>
> Somehow, the two lines in 16690 cause make check to fails. If you
> delete either line (or jsut comment the out), it passes. Leave them
> both in, it fails. I can't see what the heck would cause it.
>
> And I still believe that it should be two currencies for reasons
> stated below.  
>
> I am baffled, so someone please advise. Should I revert 16690
> altogether? just delete the line for the old currency so `make check`
> passes or is something else going on here that this changeset is
> somehow exposing. The fact that rolling back 16619 makes me think so. 
>
> Also `make check` fails later in sx, but I don't have a clue about
> that.

It's a bug in the tests..  The tests use an RNG with a fixed seed to
generate random transactions, and the additional currency changed the
results of the quazi-random selection.  That coupled with the "fixed"
lot code was causing an overflow situation.

I've fixed this in trunk..  But if we're going to backport the
afghani change then we'll need to also backport my test harness
changes from r16716.

Thank you for helping me test this last night, Andrew.  Tracking
it down to that one changeset helped me figure out what was going
on.

-derek
-- 
       Derek Atkins, SB '93 MIT EE, SM '95 MIT Media Laboratory
       Member, MIT Student Information Processing Board  (SIPB)
       URL: http://web.mit.edu/warlord/    PP-ASEL-IA     N1NWH
       warlord at MIT.EDU                        PGP key available