[MERGE REQUEST] revision accessor test case and timestamp fix

[Jamie Wilkinson]

This one time, at band camp, Martin Pool wrote:
>On 28 Mar 2006, Jamie Wilkinson <jaq at spacepants.org> wrote:
>> I posted this to the list a few months ago, it must've gotten lost :)  Still
>> looking for review and merge.
>
>Sorry it got lost. 

No worries, it was good to get back in the internals and work out how the
new API worked.

>> revno: 1601
>> committer: Jamie Wilkinson <jaq at spacepants.org>
>> branch nick: bzr.jaq
>> timestamp: Mon 2006-03-27 23:52:15 +1100
>> message:
>>   don't cast commit timestamp to long, which loses precision
>> ------------------------------------------------------------
>> revno: 1600
>> committer: Jamie Wilkinson <jaq at spacepants.org>
>> branch nick: bzr.jaq
>> timestamp: Mon 2006-03-27 23:44:47 +1100
>> message:
>>   test that revision accessors act idempotently
>
>If we don't cast it to long, then we need to commit to keep all of the
>precision in all storage formats.  In particular if the times are
>represented as decimals we need to be quite careful about how they're
>written in and out.  I'm a little surprised that this test actually
>passes at the moment - perhaps the default %f format for doubles writes
>enough digits to make them precise?  At any rate it should be
>specifically tested -- this test may fail intermittently if it happens
>to run at a time that's not correctly round-tripped.
>
>I don't mind removing the cast if it will work properly.

I'm in favour of either casting all the time or not at all, I don't have a
preference.  The fix in my branch was the least impact.