date issue with time_t value near dst switch (test, patch)

[Denys Vlasenko]

On Thursday 30 December 2010 08:57, Alexey Soloviev wrote:
> Hello,
> I haven't received any feedback to my post about date issue.
> I'd very appreciate any even negative.
> Now I've added to patch a few tests for date with time_t values.
> Also I've done thorough automated testing of patched busybox date 
> handling time_t values
>  comparing output with date from Debian Lenny's coreutils.
> Any notes, comments and questions are welcome.
> 
> Best Regards,
> Alexey.

Applied, thanks!


 
> ext Alexey Soloviev пишет:
> >
> > Hello,
> > There is an issue with date utility when using time_t values (unix time)
> > in 1 hour before autumn's DST switch off. Result time is 1hr in advance.
> > To reproduce set e.g. Helsinki timezone (e.g."export
> > TZ=Europe/Helsinki") and
> > type: "date @1288486799 ; date +%s"
> >
> > Expected outcome is
> > Sun Oct 31 03:59:59 EEST 2010
> > 1288486799
> >
> > Actual outcome is
> > Sun Oct 31 03:59:59 EET 2010
> > 1288490399
> >
> > coreutils date from Debian lenny works as expected.
> >
> > As I see, busybox date does the following
> >   1) Gets current time as time_t (unix time)
> >   2) Calls localtime_r to convert it to tm structure (human readable)
> >   3) Parses argument, which is in time_t already, converts to tm using
> > localtime
> >   4) Invalidates DST value by setting isdst=-1 and calls mktime to get
> > time_t
> > value that is then used to set system time
> >
> > The problem is without respecting dst value of time, we get into time
> > interval
> > that happens twice both before DST is switched off and after it's
> > switched off
> > and time gets 1hr back. So only dst value could tell if DST already
> > switched or
> > not. And mktime assumes that received time is after DST switch,
> > which is not true in the case.
> >
> > Suggested fix is to invalidate dst only if time is not in time_t format
> > (check for @ prefix).
> >
> > I've attached the patch and a test. The test can be run with runtest 
> > or like
> >  sh date- at -works || echo 'FAIL'
> >
> > I'd appreciate any comments on my 1st post to this maillist.
> >
> > Best Regards, Alexey.

-- 
vda