Florent Becker wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> Hi Ganesh
>>>>
>>> The metadata for the recorded (as opposed to .dpatch-applied) patch
> ^----s/recorded/converted/
>>> are in whatever timezone the user is in. So diff complains that
>>> 12:34:56 BST and 13:34:56 EST are different. One side effect of
>>> convert is that all dates in patches get their dates rewritten in
>>> the time-zone of the converter. I think it's largely ok, so i've
>>> changed the test to ignore that. The cleanest way would be to make
>>> all times UTC, but I haven't found an easy way to do that.
>>
>> Actually, I would call this a bug in convert; it's destroying
>> information without good cause. For now your fix to the test makes
>> sense, though. Luckily we won't need it on the all tests that
>> compare darcs send output.
>>
> Maybe I got the timezone conversion wrong, but it's not actually
> destroying information, just presenting it differently (the date of
> the patch is not changed, just the way it is stored). I don't see
> where the "timezone where the patch was recorded" bit is actually
> used, so I don't see any information we're actually destroying. Note
> that apply -> send is also not a round-trip if you care about that
> point, but let's not bikeshed about that.
Oh, I see. I misunderstood completely; the patches are always stored in
a repo with the same time and convert preserves that, but send formats
it according to the local time when providing the human-readable text at
the top. That's fine, but it does also mean that the other tests I
recently added are broken too, I'll fix them later.
Ganesh
|