Version 20 of clock scan

Updated 2014-04-02 00:38:37 by AMG

[clock scan] scans a string and extracts a time.

See Also

Tcl Chatroom ,2013-10-19
KBK dishes some of the details of [clock scan]
Time and date validator, sbron

Documentation

official reference

Synopsis

clock scan inputString ?-option value...?

Description

Parses the time described in inputString and returns the number of seconds (since the start of the Unix epoch) that that time corresponds to. Supported -options are:

-base time
-format format
-gmt boolean
-locale localeName
-timezone zoneName

Incomplete Times

[clock scan] does what it can to make sense of incomplete times, but the results can be surprising. If possible, make sure at least the year, month, and day are part of the string to scan. If the results of scanning an incomplete or impossible date like March 0 or April 34 are surprising, carefully re-read the seven rules that are applied, in order of preference, to make sense of a time.

Removing /etc/localtime

JMN 2007-11-14:

On FreeBSD - you can configure the timezone to UTC by removing /etc/localtime. I don't know if this is actually a legitimate way to set it to UTC, but it was the only way I knew of until now, and it does seem to cause a problem with Tcl's clock scan.

% clock scan "2007-11-01" -format "%Y-%m-%d"
time value too large/small to represent
% clock scan "2007-11-01" -format "%Y-%m-%d"
1193875200

The call always fails the first time in each process, and seems to work for all calls thereafter. (including in other interps)

If you configure the timezone to UTC by instead making /etc/localtime a link to /usr/share/zoneinfo/Etc/UTC the problem doesn't occur. I have a number of systems configured with no /etc/localtime - but I guess I'll change that now in light of this.

This issue also shows up in the FAQ on http://amsn-project.net/ .

Is a missing /etc/localtime something that clock scan should be able to handle - or is it would it be considered an OS configuration issue?

DKF: Ultimately, I'd consider this to be OS misconfiguration as it feels like stating that the system has no local timezone instead of that it is UTC.


TCV 2013-02-11: providing only a year (e.g. %Y) in the format string will not produce the time for the first day of that year. Similarly, providing only the year and month (e.g. %m/%Y) in the format string will not produce the time for the first day of that month. In both of these cases, you will instead get the time corresponding to midnight on the current day. Currently in clock scan, to avoid this behaviour, day level accuracy is required; this can be done by specifying:

  • year and day-of-year
  • week and day-of-week
  • month and day-of-month

So in the examples above, just adding to the input string 01 for each of the missing fields will give you the time you're interested in (first day of the year or month, respectively).

Examples:

# Things you might think would behave differently.

% clock scan 2000 -format %Y
1360558800
% clock scan 10/1900 -format %m/%Y
1360558800

# They're actually midnight today!

% clock format [clock seconds] -format %Y-%m-%d
2013-02-11
% clock format 1360558800 -format %Y-%m-%d %H:%M:%S
2013-02-11 00:00:00

# A simple solution. That's what I really meant!

% clock scan "2000-01-01" -format "%Y-%m-%d"
946702800
% clock scan "1900-10-01" -format "%Y-%m-%d"
-2185387200

AMG: [clock scan] is behaving badly for me when processing RCS timestamps:

% clock format [clock scan "2000/11/01 06:37:54"]
Tue Aug 11 06:37:54 CST 2167

Why so far off? This date should be interpreted as 2000-Nov-01. Let's try that again:

% clock format [clock scan "2000/11/01 06:37:54" -format "%Y/%m/%d %T"]
Wed Nov 01 06:37:54 CST 2000