In shell, how can I tail
the latest file created in a directory?
-
1come on closers, programmers need to tail!– amit kumarMar 8, 2010 at 14:37
-
The close is only for moving to superuser or serverfault. The question will live there, and more people that might be interested will find it.– MnementhMar 8, 2010 at 15:14
-
The real problem here is finding the most recently update file in the directory and I believe that that has already been answered (either here or on Super User, I can't recall).– dmckee --- ex-moderator kittenMar 8, 2010 at 19:17
12 Answers
tail `ls -t | head -1`
If you're worried about filenames with spaces,
tail "`ls -t | head -1`"
-
1But what happens when your latest file has spaces or special characters? Use $() instead of `` and quote your subshell to avoid this problem.– phoggMar 8, 2010 at 14:05
-
I like this. Clean and simple. As it should be.– wicMar 8, 2010 at 14:28
-
6
-
2Well, it depends on what you're doing, really. A solution that always works everywhere, for all possible filenames, is very nice, but in a constrained situation (log files, for example, with known non-weird names) it might be unnecessary.– PointyMar 8, 2010 at 14:49
-
Do not parse the output of ls! Parsing the output of ls is difficult and unreliable.
If you must do this I recommend using find. Originally I had here a simple example merely to give you the gist of the solution, but since this answer seems somewhat popular I decided to revise this to provide a version that is safe to copy/paste and use with all inputs. Are you sitting comfortably? We'll start with a oneliner that will give you the latest file in the current directory:
tail -- "$(find . -maxdepth 1 -type f -printf '%T@.%p\0' | sort -znr -t. -k1,2 | while IFS= read -r -d '' -r record ; do printf '%s' "$record" | cut -d. -f3- ; break ; done)"
Not quite a oneliner now, is it? Here it is again as a shell function and formatted for easier reading:
latest-file-in-directory () {
find "${@:-.}" -maxdepth 1 -type f -printf '%T@.%p\0' | \
sort -znr -t. -k1,2 | \
while IFS= read -r -d '' -r record ; do
printf '%s' "$record" | cut -d. -f3-
break
done
}
And now that as a oneliner:
tail -- "$(latest-file-in-directory)"
If all else fails you can include the above function in your .bashrc
and consider the problem solved, with one caveat. If you just wanted to get the job done you need not read further.
The caveat with this is that a file name ending in one or more newlines will still not be passed to tail
correctly. Working around this problem is complicated and I consider it sufficient that if such a malicious file name is encountered the relatively safe behavior of encountering a "No such file" error will occur instead of anything more dangerous.
Juicy details
For the curious this is the tedious explanation of how it works, why it's safe and why other methods probably aren't.
Danger, Will Robinson
First of all, the only byte that is safe to delimit file paths is null because it is the only byte universally forbidden in file paths on Unix systems. It is important when handling any list of file paths to only use null as a delimiter and, when handing even a single file path from one program to another, to do so in a manner which will not choke on arbitrary bytes. There are many seemingly-correct ways to solve this and other problems which fail by assuming (even accidentally) that file names will not have either new lines or spaces in them. Neither assumption is safe.
For today's purposes step one is to get a null-delimited list of files out of find. This is pretty easy if you have a find
supporting -print0
such as GNU's:
find . -print0
But this list still does not tell us which one is newest, so we need to include that information. I choose to use find's -printf
switch which lets me specify what data appears in the output. Not all versions of find
support -printf
(it is not standard) but GNU find does. If you find yourself without -printf
you will need to rely on -exec stat {} \;
at which point you must give up all hope of portability as stat
is not standard either. For now I'm going to move on assuming you have GNU tools.
find . -printf '%T@.%p\0'
Here I am asking for printf format %T@
which is the modification time in seconds since the beginning of the Unix epoch followed by a period and then followed by a number indicating fractions of a second. I add to this another period and then %p
(which is the full path to the file) before ending with a null byte.
Now I have
find . -maxdepth 1 \! -type d -printf '%T@.%p\0'
It may go without saying but for the sake of being complete -maxdepth 1
prevents find
from listing the contents of sub directories and \! -type d
skips directories which you are unlikely to want to tail
. So far I have files in the current directory with modification time information, so now I need to sort by that modification time.
Getting it in the right order
By default sort
expects its input to be newline-delimited records. If you have GNU sort
you can ask it to expect null-delimited records instead by using the -z
switch.; for standard sort
there is no solution. I am only interested in sorting by the first two numbers (seconds and fractions of a second) and don't want to sort by the actual file name so I tell sort
two things: First, that it should consider the period (.
) a field delimiter and second that it should only use the first and second fields when considering how to sort the records.
| sort -znr -t. -k1,2
First of all I am bundling three short options that take no value together; -znr
is just a concise way of saying -z -n -r
). After that -t .
(the space is optional) tells sort
the field delimiter character and -k 1,2
specifies the field numbers: first and second (sort
counts fields from one, not zero). Remember that a sample record for the current directory would look like:
1000000000.0000000000../some-file-name
This means sort
will look at first 1000000000
and then 0000000000
when ordering this record. The -n
option tells sort
to use numeric comparison when comparing these values, because both values are numbers. This may not be important since the numbers are of fixed length but it does no harm.
The other switch given to sort
is -r
for "reverse." By default the output of a numeric sort will be lowest numbers first, -r
changes it so that it lists the lowest numbers last and the highest numbers first. Since these numbers are timestamps higher will mean newer and this puts the newest record at the beginning of the list.
Just the important bits
As the list of file paths emerges from sort
it now has the desired answer we're looking for right at the top. What remains is to find a way to discard the other records and to strip the timestamp. Unfortunately even GNU head
and tail
do not accept switches to make them operate on null-delimited input. Instead I use a while loop as a kind of poor man's head
.
| while IFS= read -r -d '' record
First I unset IFS
so that the list of files is not subjected to word splitting. Next I tell read
two things: Do not interpret escape sequences in the input (-r
) and the input is delimited with a null byte (-d
); here the empty string ''
is used to indicate "no delimiter" aka delimited by null. Each record will be read in to the variable record
so that each time the while
loop iterates it has a single timestamp and a single file name. Note that -d
is a GNU extension; if you have only a standard read
this technique will not work and you have little recourse.
We know that the record
variable has three parts to it, all delimited by period characters. Using the cut
utility it is possible to extract a portion of them.
printf '%s' "$record" | cut -d. -f3-
Here the entire record is passed to printf
and from there piped to cut
; in bash you could simplify this further using a here string to cut -d. -3f- <<<"$record"
for better performance. We tell cut
two things: First with -d
that it should a specific delimiter for identifying fields (as with sort
the delimiter .
is used). Second cut
is instructed with -f
to print only values from specific fields; the field list is given as a range 3-
which indicates the value from the third field and from all following fields. This means that cut
will read and ignore everything up to and including the second .
that it finds in the record and then will print the remainder, which is the file path portion.
Having printed the newest file path there's no need to keep going: break
exits the loop without letting it move on to the second file path.
The only thing that remains is running tail
on the file path returned by this pipeline. You may have noticed in my example that I did this by enclosing the pipeline in a subshell; what you may not have noticed is that I enclosed the subshell in double quotes. This is important because at the last even with all of this effort to be safe for any file names an unquoted subshell expansion could still break things. A more detailed explanation is available if you're interested. The second important but easily-overlooked aspect to the invocation of tail
is that I provided the option --
to it before expanding the file name. This will instruct tail
that no more options are being specified and everything following is a file name, which makes it safe to handle file names that begin with -
.
-
1@AakashM: because you may get "surprising" results, e.g. if a file has "unusual" characters in its name (almost all characters are legal). Mar 8, 2010 at 13:55
-
9People who use special characters in their file names deserve everything they get :-)– paxdiabloMar 8, 2010 at 13:56
-
6Seeing paxdiablo make that remark was painful enough, but then two people voted it up! People who write buggy software intentionally deserve everything they get. Mar 8, 2010 at 14:10
-
4So the solution above doesn't work on osx due to lack of -printf option in find, but the following works only on osx due to differences in the stat command... maybe it will still help somebody
tail -f $(find . -type f -exec stat -f "%m {}" {} \;| sort -n | tail -n 1 | cut -d ' ' -f 2)
May 22, 2012 at 17:59 -
2"Unfortunately even GNU
head
andtail
do not accept switches to make them operate on null-delimited input." My replacement forhead
:… | grep -zm <number> ""
. Aug 17, 2017 at 20:29
You can use:
tail $(ls -1t | head -1)
The $()
construct starts a sub-shell which runs the command ls -1t
(listing all files in time order, one per line) and piping that through head -1
to get the first line (file).
The output of that command (the most recent file) is then passed to tail
to be processed.
Keep in mind this runs the risk of getting a directory if that's the most recent directory entry created. I've used that trick in an alias to edit the most recent log file (from a rotating set) in a directory that contained only those log files.
-
The
-1
isn't necessary,ls
does that for you when it's in a pipe. Comparels
andls|cat
, for example. Mar 8, 2010 at 14:27 -
That may be the case under Linux. In "true" Unix, processes didn't change their behaviour based on where their output was going. That would make pipeline debugging really annoying :-)– paxdiabloMar 8, 2010 at 14:35
-
Hmmm, not sure that's correct -- ISTR having to issue "ls -C" to get column-formatted output under 4.2BSD when piping the output through a filter, and I'm pretty sure ls under Solaris works the same way. What is the "one, true Unix" anyway?– TMNMar 8, 2010 at 15:14
-
-
@TMN: The one true Unix way is not to rely on ls for non-human consumers. "If the output is to a terminal, the format is implementation-defined." - this is the spec. If you want to be sure you have to say ls -1 or ls -C.– phoggMar 8, 2010 at 19:08
In zsh
:
tail *(.om[1])
See: http://zsh.sourceforge.net/Doc/Release/Expansion.html#Glob-Qualifiers, here m
denotes modification time m[Mwhms][-|+]n
, and the preceding o
means that it is sorted in one way (O
sorts it the other way). The .
means only regular files. Within the brackets [1]
picks the first item. To pick three use [1,3]
, to get the oldest use [-1]
.
It's nice short and doesn't use ls
.
-
Works great, but be careful not to surround this in quotes, because it doesn't work that way.– NoumenonMay 10, 2021 at 21:59
On POSIX systems, there is no way of getting the "last created" directory entry. Each directory entry has atime
, mtime
and ctime
, but contrary to Microsoft Windows, the ctime
doesn't mean CreationTime, but "Time of last status change".
So the best you can get is to "tail the last recently modified file", which is explained in the other answers. I would go for this command:
tail -f "$(ls -tr | sed 1q)"
Note the quotes around the ls
command. This makes the snippet work with almost all filenames.
There are probably a million ways to do this, but the way I would do it is this:
tail `ls -t | head -n 1`
The bits between the backticks (the quote like characters) are interpreted and the result returned to tail.
ls -t #gets the list of files in time order
head -n 1 # returns the first line only
-
2
A simple:
tail -f /path/to/directory/*
works just fine for me.
The problem is to get files that are generated after you started the tail command. But if you don't need that (as all the solutions above do not care for it), the asterisk is just simpler solution, IMO.
tail`ls -tr | tail -1`
Someone posted it, and then erased it for some reason, but this is the only one that works, so...
tail -f `ls -tr | tail`
-
-
1I posted this originally but I deleted it since I agree with Sorpigal that parsing output from
ls
isn't the smartest thing to do... Mar 8, 2010 at 13:56 -
I need it quick and dirty, no directories in it. So, If you'll add your answer, I will accept that one Mar 8, 2010 at 13:58
tail -f `ls -lt | grep -v ^d | head -2 | tail -1 | tr -s " " | cut -f 8 -d " "`
Explanation:
- ls -lt: List of all files and directories sorted by modification time
- grep -v ^d: exclude directories
- head -2 onwards: parsing the needed filename
-
1+1 for clever, -2 for parsing ls output, -1 for not quoting the subshell, -1 for a magic "field 8" assumption (it's not portable!) and finally -1 for too clever. Overall score: -4.– phoggMar 8, 2010 at 14:17
-
-