from https://www.gnu.org/software/gawk/manual/html_node/Time-Functions.html I understand that gawk has only 2 functions to work on date/time mktime and strftime .

So, I can parse any date using mktime that return a long, so I can make any math op, and so I can format the output desired with strftime

This works like a charm for any date after "1970 01 01 00 00 00"

Using awk, how can I format dates before 1970 ?

$ awk 'BEGIN{t=mktime("1970 01 01 00 00 00"); print t; print strftime("%Y-%m-%d", t) }'
10800
1970-01-01
$ awk 'BEGIN{t=mktime("1960 01 01 00 00 00"); print t; print strftime("%Y-%m-%d", t) }'
-315608400
awk: cmd. line:1: (FILENAME=- FNR=1) fatal: strftime: second argument less than 0 or too big for time_t

---

Unfortunately, as you've seen, gawk just can't do this directly. The gawk manual says:

All known POSIX-compliant systems support timestamps from 0 through 2^31 - 1, which is sufficient to represent times through 2038-01-19 03:14:07 UTC. Many systems support a wider range of timestamps, including negative timestamps that represent times before the epoch.

The manual doesn't say what strftime() does if given an out-of-range date.

But even on my system, which does behave sensibly for negative time_t values, gawk's strftime() function doesn't support them (though mktime() does), and so can't handle dates before 1970. I consider this to be a bug in gawk.

(My advice would be to use Perl instead of Awk, but that doesn't answer the question you asked.)

In principle, you could reinvent the wheel by re-implementing a function like strftime() in awk. But that would be overkill.

If your system has a working GNU coreutils date command, you can invoke it from gawk . Using your example of Jan 1, 1960:

$ cat 1960.awk
#!/usr/bin/awk -f

BEGIN {
    timestamp = mktime("1960 00 00 00 00 00")
    print "mktime() returned " timestamp

    if (0) {
        # This doesn't work
        s = strftime("%Y-%m-%d %H:%M:%S", timestamp)
        print "strftime() returned ", s
    }
    else {
        # This works
        "date '+%Y-%m-%d %H:%M:%S' -d @" timestamp | getline t
        print "The date command printed "" t """
    }
}
$ ./1960.awk
mktime() returned -318355200
The date command printed "1959-11-30 00:00:00"
$

(I gave up on figuring out the sequence of quotes and backslashes needed to do this as a one-liner from a shell prompt.)

This probably makes sense if you have a large existing awk program and you need to add this one feature to it. But if you're not stuck with doing this in awk, you might consider using something else; awk may not be the right tool for what you're trying to accomplish.

Or, if you're really ambitious, you could modify the gawk sources to handle this case correctly.

---

So, it's a bug ...

I'm using GNU awk 4.0.2, a litte look at source and seems easy to fix:

glaudiston:/sources/gawk-4.0.2$ diff builtin.c.orig builtin.c
1701,1702c1701,1702
<                       if (clock_val < 0)
<                               fatal(_("strftime: second argument less than 0 or too big for time_t"));
---
>                       // if (clock_val < 0)
>                       //      fatal(_("strftime: second argument less than 0 or too big for time_t"));
glaudiston:/sources/gawk-4.0.2$ echo "" | ./gawk '{ts="1969 12 31 23 00 00";format="%Y/%m/%d";tv=mktime(ts);print tv;print strftime(format, tv)}'
7200
1969/12/31
glaudiston:/sources/gawk-4.0.2$ echo "" | ./gawk '{ts="1960 01 01 00 00 00";format="%Y/%m/%d";tv=mktime(ts);print tv;print strftime(format, tv)}'
-315608400
1960/01/01

For my purpose it worked, but I'm not sure if it was a good idea. I'll send this to gawk maillist for approval.

Discussion started at: https://lists.gnu.org/archive/html/bug-gawk/2015-04/msg00012.html

Solution Update:

The awk dev team has fixed the bug, so just upgrade your awk to a new version:

https://lists.gnu.org/archive/html/bug-gawk/2015-04/msg00036.html

链接地址: http://www.djcxy.com/p/84374.html

上一篇: 如何在R中的函数体中设置断点

下一篇: awk与1970年以前的日期