Get yesterday's date in bash on Linux, DST-safe - linux

I have a shell script that runs on Linux and uses this call to get yesterday's date in YYYY-MM-DD format:
date -d "1 day ago" '+%Y-%m-%d'
It works most of the time, but when the script ran yesterday morning at 2013-03-11 0:35 CDT it returned "2013-03-09" instead of "2013-03-10".
Presumably daylight saving time (which started yesterday) is to blame. I'm guessing the way "1 day ago" is implemented it subtracted 24 hours, and 24 hours before 2013-03-11 0:35 CDT was 2013-03-09 23:35 CST, which led to the result of "2013-03-09".
So what's a good DST-safe way to get yesterday's date in bash on Linux?

I think this should work, irrespective of how often and when you run it ...
date -d "yesterday 13:00" '+%Y-%m-%d'

Under Mac OSX date works slightly different:
For yesterday
date -v-1d +%F
For Last week
date -v-1w +%F

This should also work, but perhaps it is too much:
date -d #$(( $(date +"%s") - 86400)) +"%Y-%m-%d"

If you are certain that the script runs in the first hours of the day, you can simply do
date -d "12 hours ago" '+%Y-%m-%d'
BTW, if the script runs daily at 00:35 (via crontab?) you should ask yourself what will happen if a DST change falls in that hour; the script could not run, or run twice in some cases. Modern implementations of cron are quite clever in this regard, though.

Here a solution that will work with Solaris and AIX as well.
Manipulating the Timezone is possible for changing the clock some hours.
Due to the daylight saving time, 24 hours ago can be today or the day before yesterday.
You are sure that yesterday is 20 or 30 hours ago. Which one? Well, the most recent one that is not today.
echo -e "$(TZ=GMT+30 date +%Y-%m-%d)\n$(TZ=GMT+20 date +%Y-%m-%d)" | grep -v $(date +%Y-%m-%d) | tail -1
The -e parameter used in the echo command is needed with bash, but will not work with ksh.
In ksh you can use the same command without the -e flag.
When your script will be used in different environments, you can start the script with #!/bin/ksh or #!/bin/bash. You could also replace the \n by a newline:
echo "$(TZ=GMT+30 date +%Y-%m-%d)
$(TZ=GMT+20 date +%Y-%m-%d)" | grep -v $(date +%Y-%m-%d) | tail -1

date -d "yesterday" '+%Y-%m-%d'
To use this later:
date=$(date -d "yesterday" '+%Y-%m-%d')

you can use
date -d "30 days ago" +"%d/%m/%Y"
to get the date from 30 days ago, similarly you can replace 30 with x amount of days

Just use date and trusty seconds:
As you rightly point out, a lot of the details about the underlying computation are hidden if you rely on English time arithmetic. E.g. -d yesterday, and -d 1 day ago will have different behaviour.
Instead, you can reliably depend on the (precisely documented) seconds since the unix epoch UTC, and bash arithmetic to obtain the moment you want:
date -d #$(( $(date +"%s") - 24*3600)) +"%Y-%m-%d"
This was pointed out in another answer. This form is more portable across platforms with different date command line flags, is language-independent (e.g. "yesterday" vs "hier" in French locale), and frankly (in the long-term) will be easier to remember, because well, you know it already. You might otherwise keep asking yourself: "Was it -d 2 hours ago or -d 2 hour ago again?" or "Is it -d yesterday or -d 1 day ago that I want?"). The only tricky bit here is the #.
Armed with bash and nothing else:
Bash solely on bash, you can also get yesterday's time, via the printf builtin:
%(datefmt)T
causes printf to output the date-time string resulting from using
datefmt as a format string for strftime(3). The corresponding argu‐
ment is an integer representing the number of seconds since the
epoch. Two special argument values may be used: -1 represents the
current time, and -2 represents the time the shell was invoked.
If no argument is specified, conversion behaves as if -1 had
been given.
This is an exception to the usual printf behavior.
So,
# inner printf gets you the current unix time in seconds
# outer printf spits it out according to the format
printf "%(%Y-%m-%d)T\n" $(( $(printf "%(%s)T" -1) - 24*3600 ))
or, equivalently with a temp variable (outer subshell optional, but keeps environment vars clean).
(
now=$(printf "%(%s)T" -1);
printf "%(%Y-%m-%d)T\n" $((now - 24*3600));
)
Note: despite the manpage stating that no argument to the %()T formatter will assume a default -1, i seem to get a 0 instead (thank you, bash manual version 4.3.48)

You can use:
date -d "yesterday 13:55" '+%Y-%m-%d'
Or whatever time you want to retrieve will retrieved by bash.
For month:
date -d "30 days ago" '+%Y-%m-%d'

As this question is tagged bash "DST safe":
And using fork to date command implie delay, there is a simple and more efficient way using pure bash built-in:
printf -v tznow '%(%z %s)T' -1
TZ=${tznow% *} printf -v yesterday '%(%Y-%m-%d)T' $(( ${tznow#* } - 86400 ))
echo $yesterday
This is a lot quicker on more system friendly than having to fork to date.
From bash version 5.0, there is a new variable $EPOCHSECONDS
printf -v tz '%(%z)T' -1
TZ=$tz printf -v yesterday '%(%Y-%m-%d)T' $(( EPOCHSECONDS - 86400 ))
echo $yesterday

Related

Find number of days from two dates using shell

I would like to find number of days from two given dates as 25 Aug 2017 and 05 Sep 2017.
My script:
start='2017-08-25';
end='2017-09-05';
ndays=(strtotime($end)- strtotime($start))/24/3600;
echo $ndays
When run this script, I am getting following error messages.
Line 3: syntax error near unexpected token `('
Desire output value:
10
No shell that I am aware of has any tools for working with dates. At the very least, you need an external tool like date to convert your dates to an intermediate form, like the number of seconds since some fixed point in time (Unix uses Jan 1, 1970), then do your calculation with those values before processing the result further.
Assuming the use of GNU date from the Linux tag, you would do something like
start='2017-08-25'
end='2017-09-05'
start_seconds=$(date +%s --date "$start" --utc)
end_seconds=$(date +%s --date "$end" --utc)
ndays=$(( (end_seconds - start_seconds) / 24 / 3600 ));
echo $ndays
Note that since most shells only support integer arithmetic, this won't give you an exact number of days.
You can use date to convert each date to epoch seconds, then use arithmetic expansion to do the subtraction and conversion back to days:
#! /bin/bash
start='2017-08-25';
end='2017-09-05';
diff=$(date -d $end +%s)-$(date -d $start +%s)
echo $(( ($diff) / 60 / 60 / 24 )) # 11

How I can get the next month in shell?

I would like to get the current and next months using shell script, I have tried this command:
$ date '+%b'
mar
$ date +"%B %Y" --date="$(date +%Y-%m-15) next month"
March 2018
But it always displays only the current month.
Could you please help me if there is something wrong with the commands.
$ date -d "next month" '+%B %Y'
April 2018
Check this post about specific caveats
Note: your command works just fine for me (archlinux, bash4, date GNU coreutils 8.29)
I wouldn't rely on date alone to do this. Instead, perform a little basic math on the month number.
this_month=$(date +%-m) # GNU extension to avoid leading 0
next_month=$(( this_month % 12 + 1 ))
next_month_name=$(date +%B --date "2018-$next_month-1")
Since you are using bash, you don't need to use date at all to get the current month; the built-in printf can call the underlying date/time routines itself, saving a fork.
$ printf -v this_month '%(%-m)T\n'
$ echo $this_month
3
What variant and version of date are you running? "Solaris 5.2" was never released, though SunOS 5.2 was a kernel in Solaris 2.2 (EOL in 1999). See the Solaris OS version history. The Solaris 10 (SunOS 5.10) man page for date does not support GNU's --date= syntax used in the question, so I'm guessing you're using some version of date from GNU coreutils.
Here's a solution using BSD date (note, this is academic in the face of BSD's date -v 1m):
date -jf %s $((1728000+$(date -jf %Y-%m-%d $(date +%Y-%m-15) +%s))) +"%B %Y"
There are three date calls here. BSD's date allows specifying a format (GNU can intuit most formats on its own). The parent call is the one that takes the final time (as seconds since the 1970 epoch), expressing it in the desired "Month Year" format. Seconds are determine by adding 20 days to the epoch time of the current month on the 15th day. Since no month has 15+20 days, this is always the following month.
Here's a direct translation of that logic to GNU date:
date +"%B %Y" --date="#$((1728000+$(date +%s --date=$(date +%Y-%m-15))))"
Here's a simpler solution using GNU date, with one fewer date call:
date +"%B %Y" --date="$(date +%Y-%m-15) 20 days"
(A bug in GNU date will give you the wrong month if you run date --date="next month" on the 31st.)
The below one works in Red Hat 4.4.7-23, Linux version 2.6.32-754.2.1.el6.x86_64.
Just use the "month" for future months and "month ago" for previous months.. Dont confuse with adding +/- signs to the number. Check out.
> date "+%B-%Y" #current month
November-2018
> date -d" 1 month" "+%B-%Y"
December-2018
> date -d" 1 month ago" "+%B-%Y"
October-2018
>
More..
> date -d" 7 month ago" "+%B-%Y"
April-2018
> date -d" 7 month " "+%B-%Y"
June-2019
>

Integer expression expected when calculating date difference

I have this script I am building and need a bit of help. I want to use as minimal lines of code as possible. However what I want the script to do is the following.
Backup a specified file such as authlog into a new file with the keywords I specify.
However. I also want it to back up the file if current backed up file is older than 2+ days ago and if it is newer. Leave it alone. if (file) is older than +2 days old delete it and replace it with an updated one.
I'm getting integer expression expected and I'm not sure how to fix it
#!/bin/bash
authlog=/home/(myhomedir)/logs/backups/authlog-${current_date}
backup="$(cat /var/log/auth.log | grep -e failed -e invalid > /home/(myhomedir)/logs/backup)"
dayold=$(date -d '1 day ago' "+%Y-%m-%d")
current_date=$(date +"%m_%d_%Y")
if [ $authlog -lt $dayold ] ;
then ${backup}
else
echo $authlog is newer than $current_date $authlog has not been updated
if [ $authlog -eq ${current_date} ]
then "$(rm {authlog})"
fi
fi
You can't use -lt to compare non-numeric strings. This is why you're getting "integer expression expected". You're trying to compare a filename to a date string.
The easiest way to compare dates is by converting them to seconds since epoch (Jan 1, 1970 00:00:00 UTC). As an example, suppose I have a file called foo.bar and I want to check if it was last modified 3 or more days ago:
filetime=$(stat -c %Y foo.bar)
three_days_ago=$(date -d "3 days ago" +%s)
if [ $filetime -lt $three_days_ago ]; then
echo "foo.bar was last modified more than 3 days ago"
fi
Also, your authlog variable does not contain a date. I'm assuming there's a missing step that you forgot.

bash - extract day of week of variable

What is the syntax to extract the day of the week from a stored date variable?
The dateinfile format is always [alphanum]_YYYYMMDD.
In this pseudocode example, trying to get dayofweek to store Saturday:
#! /bin/bash
dateinfile="P_20090530"
dayofweek="$dateinfile -u +%A"
[me#home]$ date --date=${dateinfile#?_} "+%A"
Saturday
Or, to put it as you've requested:
[me#home]$ dayofweek=$(date --date=${dateinfile#?_} "+%A")
[me#home]$ echo $dayofweek
Saturday
date -d $(echo $dateinfile | cut -f2 -d_) -u +%A
The inner expression separates the 20090530 from P_20090530, and the outer one extracts the day of week from that date
I needed the same output recently and googled upon this, but I had the same problem stating Bad Substition error.
I then read the date manual and made my version up as follows:
#! /bin/sh
dateinfile="P_20090530"
dayofweek=`date --reference $dateinfile +%A`
echo $dayofweek

Get mtime of specific file using Bash?

I am well aware of being able to do find myfile.txt -mtime +5 to check if my file is older than 5 days or not. However I would like to fetch mtime in days of myfile.txt and store it into a variable for further usage. How would I do that?
stat can give you that info:
filemtime=$(stat -c %Y myfile.txt)
%Y gives you the last modification as "seconds since The Epoch", but there are lots of other options; more info. So if the file was modified on 2011-01-22 at 15:30 GMT, the above would return a number in the region of 1295710237.
Edit: Ah, you want the time in days since it was modified. That's going to be more complicated, not least because a "day" is not a fixed period of time (some "days" have only 23 hours, others 25 — thanks to daylight savings time).
The naive version might look like this:
filemtime=$(stat -c %Y "$1")
currtime=$(date +%s)
diff=$(( (currtime - filemtime) / 86400 ))
echo $diff
...but again, that's assuming a day is always exactly 86,400 second long.
More about arithmetic in bash here.
The date utility has a convenient switch for extracting the mtime from a file, which you can then display or store using a format string.
date -r file "+%F"
# 2021-01-12
file_mtime=$(date -r file "+%F")
See man date, the output of date is controlled by a format string beginning with "+"
Useful format strings for comparing many dates might include:
"+%j": day of year
"+%s": unix epoch time
Arithmetic with dates is a bit of a pain in bash, so if you need relative time that will work in all corner cases, you may be better off with another language.
AGE=$(perl -e 'print -M $ARGV[0]' $file)
will set $AGE to the age of $file in days, as Perl's -M operator handles the stat call and the conversion to days for you.
The return value is a floating-point value (e.g., 6.62849537 days). Add an int to the expression if you need to have an integer result
AGE=$(perl -e 'print int -M $ARGV[0]' $file)
Ruby and Python also have their one-liners to stat a file and return some data, but I believe Perl has the most concise way.
I this the answer?
A=$(stat -c "%y" myfile.txt)
look at stat-help
stat --help
Usage: stat [OPTION]... FILE...
Display file or file system status.
[...]
-c --format=FORMAT use the specified FORMAT instead of the default;
output a newline after each use of FORMAT
[...]
The valid format sequences for files
[...]
%y Time of last modification, human-readable
%Y Time of last modification, seconds since Epoch
[...]

Resources