Re: [PATCH 1/1] hwclock.c, hwclock.8.in: new --show format

[J William Piggott <elseifthen@gmx.com>]

On 02/16/2016 05:29 AM, Karel Zak wrote:
> On Thu, Feb 11, 2016 at 07:48:40PM -0500, J William Piggott wrote:
>> -		lt = localtime(&hwctime.tv_sec);
>> -		strftime(ctime_now, sizeof(ctime_now), format, lt);
>> -		printf(_("%s and %06d microseconds\n"), ctime_now, (int)hwctime.tv_usec);
>> +		struct tm lt;
>> +		int zhour, zmin;
>> +
>> +		lt = *localtime(&hwctime.tv_sec);
>> +		zhour = - timezone / 60 / 60;
>> +		zmin = abs(timezone / 60 % 60);
>> +		printf(_("%4d-%.2d-%.2d %02d:%02d:%02d.%06d%+02d:%02d\n"),
>> +		       lt.tm_year + 1900, lt.tm_mon + 1, lt.tm_mday, lt.tm_hour,
>> +		       lt.tm_min, lt.tm_sec, (int)hwctime.tv_usec, zhour, zmin);
> 
> Some notes:
> 
> * what's wrong with strftime? 

Why use resources on strftime when 8601 prints directly?

While it can do other formatting, I see strftime's main purpose as
converting tm numbers into strings like: January, Monday, etc.

> 
> * We already use ISO time in util-linux and we use time designator 'T' (separator
>   between date and time).

I chose to use the optional space as a compromise between machine
friendly and human friendly, this is commonly done I think.

> 
> * tv_usec is "long"

Fixed: I followed the example from the recent commit c211401 (for code
consistency, which is why I originally used the int cast), and cast it
to long; although I do not understand why as it is already long. All
that needed to be done was to change the conversion length modifier and
drop the cast.

PULL:

  git@github.com:jwpi/util-linux.git 020716

for you to fetch changes up to e05ac5aae00913e7a999c96a7bb731dc1d09cc5c:

> 
> * fraction separator is decimal mark, either a comma or a dot, but
>   with a preference for a comma according to ISO 8601:2004
>   (wikipedia).
> 
>   -- we already use comma in util-linux, exception is logger where I
>   see dot :-(

I actually modeled the format after:

date --rfc-3339=ns
2016-02-16 20:13:25.601508440-05:00

3339 is a subset of 8601 that requires using a period for the radix
point. As I said previously, I chose this as a compromise between
machine and human readability. As it is output to tty by default, I
thought doing so might mitigate any push back against the format change.

I am not strongly opposed to using 'T' as the date-time delimiter or comma
as the radix point. I just think 'space' and 'period' are a commonly
used compromise.

> 
> 
> Anyway, it would be really nice to have a function for this purpose in
> lib/timeutils.c to avoid duplication and creativity :-) Something like:
> 
> 
>     strtime_iso_8601(char buf, size_t bufsz, struct tm tm, struct timeval frac, int flags);
> 
> where flags are
> 
>     enum {
>          ISO_8601_TIMEZONE   = (1 << 1)
>          ISO_8601_USEC       = (1 << 2)
>          ...
>     };
> 
> then we can use this function in lslogins, dmesg, hwclock, last,
> logger, ... etc.  I can also imagine:
> 
>     strtime_short_ctime()
>     strtime_ctime()           (ctime without \n)
> 
> Comments?

Also, perhaps all the above only using something similar to coreutil's
fprintftime, so it prints directly using less resources.

> 
>     Karel
>