[Python-Dev] Epoch and Platform

[Curt Hagenlocher]

I don't *feel* anxious, but my doctor *has* been trying to persuade me
to switch to decaf...

There's no real urgency.  The reason this came up is because I just
implemented zlib, which automatically enabled the gzip unit tests.
The gzip tests are failing because the current timestamp can't be
written as a 32-bit value.  In order to checkin my changes, I can't
have any failing tests -- so my choices are

1) change the IronPython epoch so that the timestamp works for gzip and tarlib
2) change gzip and tarlib to work with a "less standard" epoch, or
3) disable the failing test(s)

...and I'd rather not resort to #3, if possible.

On Tue, Jun 17, 2008 at 10:03 AM, Guido van Rossum <guido at python.org> wrote:
> Can you explain why you are so anxious to get this resolved (apart
> from the beer :-) ?
>
> On Tue, Jun 17, 2008 at 9:26 AM, Curt Hagenlocher <curt at hagenlocher.org> wrote:
>> Any chance of an Official Pronouncement on this topic?  It would help
>> us greatly -- even if only to figure out who'll be paying for the next
>> round of beer.
>>
>> On Mon, Jun 16, 2008 at 4:38 PM, Guido van Rossum <guido at python.org> wrote:
>>> ISTR that we force the epoch to be 1970 on all major platforms -- or
>>> perhaps it happens to be 1970 even on Windows when using MS's C
>>> runtime.
>>>
>>> On Mon, Jun 16, 2008 at 4:08 PM, Curt Hagenlocher <curt at hagenlocher.org> wrote:
>>>> The documentation for the time module says that "the epoch is the point
>>>> where the time starts. On January 1st of that year, at 0 hours, the ``time
>>>> since the epoch'' is zero. For Unix, the epoch is 1970. To find out what the
>>>> epoch is, look at gmtime(0)."  This confirms that the epoch is
>>>> platform-specific.  As such, the only legal uses of the timestamp should be
>>>>
>>>> 1) comparing with another timestamp to determine elapsed time in seconds
>>>> 2) passing to another standard Python library function which expects a
>>>> timestamp
>>>> 3) as a source of randomness.
>>>>
>>>> However, the following files in the standard library have hardcoded the
>>>> assumption that the Python epoch will always be the same as the Unix epoch:
>>>> In gzip.py, method GzipFile._write_gzip_header
>>>> In tarfile.py, method _Stream._init_write_gz
>>>> In uuid.py, function uuid1
>>>>
>>>> Additionally, the following files in the standard library have hardcoded the
>>>> assumption that the Python epoch will cause timestamps to fall within the
>>>> range of a 32-bit unsigned integer value:
>>>> In imputil.py, function _compile
>>>> In py_compile.py, function compile
>>>>
>>>> So there's some kind of a potential discrepancy here, albeit a minor one.
>>>> This discrepancy can be resolved in one of at least three ways:
>>>>
>>>> 1) The documentation for the time module is wrong, and the epoch for Python
>>>> (at least versions 2.x) should be the Unix epoch.
>>>> 2) These library functions are slightly wrong and should be modified by
>>>> subtracing an "epoch offset" before doing other calculations. This offset
>>>> can be calculated as "time.mktime((1970, 1, 1, 0, 0, 0, 3, 1, 0)) -
>>>> time.timezone".
>>>> 3) These library files should be considered part of the platform-specific
>>>> implementation, and an alternate platform should provide its own version of
>>>> these files if necessary.
>>>>
>>>> Any thoughts on this?
>>>>
>>>> From the perspective of implementing IronPython, I'd prefer that the answer
>>>> is 1 or 2 -- but mainly I just want to be as compatible with "the spec" as
>>>> possible, while respecting CLR-specific norms for functionality which is
>>>> left up to individual implementations.
>>>>
>>>> --
>>>> Curt Hagenlocher
>>>> curt at hagenlocher.org
>>>> _______________________________________________
>>>> Python-Dev mailing list
>>>> Python-Dev at python.org
>>>> http://mail.python.org/mailman/listinfo/python-dev
>>>> Unsubscribe:
>>>> http://mail.python.org/mailman/options/python-dev/guido%40python.org
>>>>
>>>>
>>>
>>>
>>>
>>> --
>>> --Guido van Rossum (home page: http://www.python.org/~guido/)
>>>
>>
>
>
>
> --
> --Guido van Rossum (home page: http://www.python.org/~guido/)
>