Boost C++ Libraries: Ticket #1861: time_duration type can not be correctly written or read of the period spans 24 hours of more (vc9 and probably others).

attachment set

ilya@… — Thu, 24 Apr 2008 20:44:14 GMT

attachment → time_facet.hpp.patch

status changed; resolution set

Andrey Semashev — Thu, 13 Nov 2008 17:49:49 GMT

status new → closed
resolution → fixed

The proposed fix breaks IO for ISO-formatted dates, so %H is still limited to be 2 chars long. The new %O placeholder was added in order to format even longer durations.

attachment set

ilya.bobir@… — Wed, 11 Mar 2009 21:10:57 GMT

attachment → date_time_long_hours_patch2.patch

status, version changed; cc set; resolution deleted

ilya.bobir@… — Wed, 11 Mar 2009 21:13:55 GMT

cc ilya.bobir@… added
status closed → reopened
version Boost Development Trunk → Boost 1.38.0
resolution fixed

I think that the default format for the time_duration should be changed to contain %O instead of %H as time_duration objects can by default contain unrestricted hours value. This change affects time_duration input and output only.

I've wrote tests in order to catch bugs with time_period objects not been capable of processing values that are longer then 24 hours.

One of the tests showed a bug in the boost/date_time/format_date_parser.hpp:var_string_to_int(). The function was able to read after the input data end in some cases.

All of these are contained in the patch (date_time_long_hours_patch2.patch):

The default time_duration format value is changed to %O:%M:%S%F instead of %H:%M:%S%F.
A number of tests added that deal with time_duration values longer then 24 hours.
Bug in the boost/date_time/format_date_parser.hpp:var_string_to_int() fixed.
Documentation changes. Default format specifications for time_duration updated. And the %H should probably be documented along with %O.

I think that 1 and 2 are definitely a continuation of the original bug thus I decided to reopen it instead of creating a new one.

The diff is against the 1.38.0 but the relevant files are identical to those in trunk at the moment.

Andrey Semashev — Sun, 31 May 2009 12:17:03 GMT

I've applied fix for var_string_to_int to trunk. When the tests pass I'll merge it to the release branch.

As for duration default format change, this is a breaking change. Although it has a valid point, I'm not sure the change is justified enough. This has to be discussed on the ML.

Andrey Semashev — Thu, 04 Jun 2009 08:42:21 GMT

The fix for var_string_to_int merged into the release branch in revision 53618 (will be released in 1.40).

Thu, 23 Jul 2009 10:02:44 GMT

Replying to andysem:

As for duration default format change, this is a breaking change. Although it has a valid point, I'm not sure the change is justified enough. This has to be discussed on the ML.

It seems that no one cares about this change, except for Rutger ter Borg (http://lists.boost.org/Archives/boost/2009/06/153112.php) who says the current behavior is broken. The only person except you and me who also replied on the thread was Stewart, Robert but he was talking about unrelated things like error handling for IO streams in general (http://lists.boost.org/Archives/boost/2009/06/152121.php).

Do you think the fix can be applied now?

Andrey Semashev — Thu, 01 Oct 2009 20:19:52 GMT

(In [56506]) Refs #1861. Changed the default format for time durations to "%-%O:%M:%S%F" instead of "%-%H:%M:%S%F".

status changed; resolution set

Andrey Semashev — Sat, 03 Oct 2009 10:04:01 GMT

status reopened → closed
resolution → fixed

(In [56545]) Fixes #1861, #2213 merged from trunk.