Touch a file in Linux

chraebsli@programming.dev to Programmer Humor@programming.dev – 1163 points –
94

Does anyone actually use touch for its intended purpose? Must be up there with cat.

TIL it's actually for changing timestamps.

https://www.man7.org/linux/man-pages/man1/touch.1.html

Wtf. All these years I thought 'touch' was reference to Michelangelo's Creation of Adam.

The intended use of touch is to update the timestamp right?

Yeah. It could just as well have issued a file not found error when you try to touch a nonexistent file. And we would be none the wiser about what we're missing in the world.

“Do one thing and do it very well” is the UNIX philosophy after all; if you’re 99% likely to just create that missing file after you get a file not found error, why should touch waste your time?

Because now touch does two things.

Without touch, we could "just" use the shell to create files.

: > foo.txt

Touch does one thing from a “contract” perspective:

Ensure the timestamp of is

with this logic, any command that moves, copies or opens a file should just create a new file if it doesn't exist

and now you're just creating new files without realising just because of a typo

But this directly goes against that philosophy, since now instead of changing timestamps it's also creating files

You can pass -c to not create a file, but it does go against the philosophy that it creates them by default instead of that being an option

EDIT: Looking closer into the code, it would appear to maybe be an efficiency thing based on underlying system calls

Without that check, touch just opens a file for writing, with no other filesystem check, and closes it

With that check, touch first checks if the file exists, and then if so opens the file for writing

We use it to trigger service restarts.

touch tmp/service-restart.txt

Using monit to detect the timestamp change and do the actual restart command.

This is an interesting idea to allow non-root users to restart a service. It looks like this is doable with systemd too. https://superuser.com/a/1531261

Indeed. Replacing monit with systemd for this job is still on our todo list.

what is cat's use if not seeing whats inside a file?

It is short for concatenate, which is to join things together. You can give it multiple inputs and it will output each one directly following the previous. It so happens to also work with just one input.

It is to use along with split. e.g.

  1. You take a single large file, say 16GB
  2. Use split to break it into multiple files of 4GB
  3. Now you can transfer it to a FAT32 Removable Flash Drive and transfer it to whatever other computer that doesn't have Ethernet.
  4. Here, you can use cat to combine all files into the original file. (preferably accompanied by a checksum)

Doesnt computers do this automatically if you try to copy over a file larger than its per file size limit?

I sometimes use cat to concatenate files. For example, add a header to a csv file without manually copy and paste it. It’s rare, but at least more frequent than using touch.

$ cat file1 > output_file
$ cat file2 >> output_file
$ cat file3 >> output_file

I'm sorry!

When you updated a Django server, you were supposed to touch the settings.py file so the server would know to reload your code. (I haven't used any for a long time, so I don't know if it's still the procedure.)

There are many small things that use it.

cat

Ahhhhh, fuck. I'm quite noob with linux. I got into some rabbit hole trying to read the docs. I found 2 man pages, one is cat(1) and the other cat(1p). Apparently the 1p is for POSIX.

If someone could help me understand... As far as I could understand I would normally be concerned with (1), but what would I need to be doing to be affected by (1p)?

The POSIX standard is more portable. If you are writing scripts for your system, you can use the full features in the main man pages. If you are writing code that you want to run on other Linux systems, maybe with reduced feature sets like a tiny embedded computer or alternates to gnu tools like alpine linux, or even other unixes like the BSDs, you will have a better time if you limit yourself to POSIX-compatible features and options -- any POSIX-compatible Unix-like implementation should be able to run POSIX-compliant code.

This is also why many shell scripts will call #!/bin/sh instead of #!/bin/bash -- sh is more likely to be available on tinier systems than bash.

If you are just writing scripts and commands for your own purposes, or you know they will only be used on full-feature distributions, it's often simpler and more comfortable to use all of the advanced features available on your system.

If you execute a binary without specifying the path to it, it will be searched from the $PATH environment variable, which is a list of places to look for the binary. From left to right, the first found one is returned.

You can use which cat to see what it resolves to and whereis cat to get all possible results.

If you intentionally wants to use a different binary with the same name, you can either directly use its path, or prepend its path to $PATH.

I used it recently to update the creation date of a bunch of notes. Just wanted them to display in the correct order in Obsidian. Besides that though, always just used it for file creation lol

Touch is super useful for commands that interact with a file but don't create the file by default. For example, yesterday I needed to copy a file to a remote machine accessible over ssh so I used scp (often known as "secure copy") but needed to touch the file in order to create it before scp would copy into it

I mean, timestamps aren't really all that useful. Really just if you do some stuff with makefiles but even then it's a stretch. I did once use cat for its intended purpose tho, for a report. We split up the individual chapters into their own files so we have an easier time with git stuff, made a script that had an array with the files in the order we wanted, gave it to cat and piped that into pandoc

Yes, when you are for example checking if the permissions in the directory are correct, or if you want to check if your nfs export is working. It's one of those commands that once you know it exists, you WILL find a way to use it.

i use both frequently but im also a pretty dumb user

These are some weird looking dolph--- oh

Remember to confirm consent before touching.

You can only touch in places where you have permission to touch.

Is there a command that's actually just for creating a new file?

Nope. If you open a nonexistent path and you have permissions to write to that directory, then that file is created.

I guess printf "" > file

Feels dangerous to run. What happens if the file already exists and has something important in it?

touch -a is probably better

The other command could just be printf '' >> file to not overwrite it. Or even simpler >>file and then interrupt

or :>>file then you don't need to interrupt

.“:>>” is “append null” right? Do you get a file with a single ASCII NUL or is it truly empty?

Not really. I believe : is the "true" builtin. So it's like running a program that exits with zero and writes nothing to stdout. The >> streams the empty stdout into the named file.

$ :|wc -c 0 $ touch /tmp/f; :>>/tmp/f; wc -c /tmp/f 0 /tmp/f

that's awesome, did not know about that handy operator!

Yeah!

it's basically a noop, I use it as a placeholder when I'm writing a script, since bash doesn't accept code blocks with no commands

I mean, nano filename will work, but there's no mkfile that I can find...

$>filename would also work, but it's not explicitly for creating a new file

most shells will accept outputting from a silent command to a file, e.g. :> foo.txt (where : is the posix synonym to the true command)

How often do you actually need a blank file though? Usually you'd be writing something in the file.

I'm betting that's why none ever materialized. Most tools that can manipulate a file, can also create that file first, so there's just never been a usecase.

Right-clicking the desktop to create a new txt file in Windows feels so natural, but I can't really think of any time you'd want to create a new file and do nothing with it in a CLI.

You might if some other program checks whether that file exists and behaves differently depending on that.

But even still, what's a realistic usecase that would that involve needing a blank, unmodified file in that instance?

One use case is if you're running a web server that is configured to return a "maintenance" page instead of the live site if a particular file exists. Which is actually pretty cool because then you don't have to update the config when you need to do something or let your users get a bunch of 502 errors, you just touch maintenance and you're good.

I'm way to used to doing nano file.txt that I always forget about touch.

Although most times, if I create a file, it's to put something in it

If you need multiple files for testing a script or such: touch file{1..5}.txt

I do the opposite, I forget I can just create a file with nano. I run touch then open it with nano after to edit.

I usually do open filename because I prefer GUI text editors.