This is for a bash script to compare a local and remote mount.
Local files and directories are symlinked to remote, except for new files.
I need to replace diff because it compares the contents and becomes very slow over the internet.
I have been trying things like diff <( ls /local/ ) <( ls /remote/ ) and diff <( tree /local/ ) <( tree /remote/ ) but cannot make them work, because it they are not recursive or the symlinks get in the way.
rsync would be my go to for determining missing files, but I cannot find a way to manage exit codes and integrate into the script.
Script looks something like this:
#!/bin/bash
set -e
echo "Backup Command"
sleep 10
while :; do
echo "Testing backup"
if
diff -r /local/ /remote/ ; then
echo "diff matches!"
break
else
echo "diff didn't match, waiting for cache"
sleep 600
fi
done
echo "Finished!"
Use cmp to check if two things are the same.
Use find to get all the possible paths.
sort the paths before passing to cmp, so they are the same.
Use zero terminated strings to handle all special characters in filenames.
if cmp -s <(cd local && find -print0 | sort -z) <(cd remote && find -print0 | sort -z); then
echo "local and remote have the same directory and file structure"
else
echo "Oooh! local and remote differ. Or there was trouble."
fi
The answer was rather simple and obvious and I had tried both, but not together. To make this work, due to the symlinks, find -L is required to follow the links and then sort is necessary to make them match.
I replaced the diff command with the following:
diff <( find -L local |sort) <( find -L remote |sort)
I'm trying to use test in a make target, specifically to check for changes to certain files with git (I don't think that particular git diff-index HEAD -- command makes a difference here).
is_diff:
if [[ -n `git diff-index HEAD --` ]]; exit 1; fi
This works fine as a bash script, but as a make target it seems to be dependent on which version of make is running, which limits its utility (and makes just pointing to a shell script version better).
Is there a change that would make this a little more make-version independent?
Use [ -n "`git diff-index HEAD --`" ] instead of the bash-specific [[ -n ... ]]
was the correct answer from a comment by Renaud Pacalet.
Also as MadScientist notes, that the original question omitted then so
if [ -n "`git diff-index HEAD --`" ]; then exit 1; fi
Due to some problems with a script which commits and pushes automatically, i'd like to implement a whitelist.
The plan is, that only commits with the pattern 'foo' and 'bar' in path, are allowed.
#!/bin/sh
WHITELIST="foo bar"
WRKDIR=/home/athur/workwork/test/repo
cd $WRKDIR
git add -A
for file in `git diff --cached -p --name-status | cut -c3-`; do
if [[ "$file" == *"$WHITELIST"* ]] ; then
echo "$file is on whitelist"
else
echo "$file is not on whitelist. Commit aborted."
exit 1
fi
done
The problem is, it's always uses the 'else' clause.
I can't find the problem. Thanks
As a best-practices approach, consider:
#!/usr/bin/env bash
# ^^^^ important: [[ ]] is not guaranteed to work with bin/sh
whitelist_re='(foo|bar)'
workdir=/home/athur/workwork/test/repo
cd -- "$workdir" || exit
git add -A
while IFS= read -r filename; do
if [[ $file =~ $whitelist ]]; then
echo "$file is on whitelist" >&2
else
echo "$file is not on whitelist; commit aborted." >&2
exit 1
fi
done < <(git diff --cached --name-only)
To walk through the changes:
The shebang specifies bash as a shell, which guarantees that extensions like [[ ]] and <(...) will be available -- a guarantee not made with /bin/sh.
A while read loop is used rather than attempting to iterate over line-oriented data with for; see DontReadLinesWithFor for an explanation of the reasoning behind this change.
The whitelist is specified as an ERE-compliant regular expression, such that =~ can be used to test whether a value matches.
Instead of using git diff --cached --name-status and then using cut to remove the status data after-the-fact, we use --name-only to generate only names in the first place.
Using lowercase variable names complies with the conventions given in http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/V1_chap08.html, specifying that POSIX-defined tools will use all-caps shell and environment variable names for their own purposes, and that names with at least one lowercase character are reserved for application use. (Keep in mind that setting a shell variable overwrites any like-named environment variable, so these conventions apply even when export is not in use).
By the way, if you just wanted to find out if any non-matches exist, without knowing which files those are, you could use:
#!/bin/sh
# ^^ actually safe here, as no non-POSIX functionality is used
whitelist_re='foo|bar'
if git diff --cached --name-only | grep -qEv "$whitelist_re"; then
echo "At least one file is not on whitelist; commit aborted" >&2
exit 1
fi
Using an explicit list
The == is not symmetric in this case and ** seems to be used badly.
Try "$WHITELIST" == *"$file"*.
(Inspired by How do I check if a variable exists in a list in BASH)
Note that using your WHITELIST, only files foo and bar will be whitelisted.
Detecting a pattern
If you need to detect individual patterns, you may need to construct a function such as:
for entry in $WHITELIST ; do
if [[ "$file" =~ $entry ]] ; then
return 0
fi
done
return 1
Below is my pre-commit git hoook
#!/bin/bash
....
# if git diff -U0 "$FILE_PATH" | grep -iq 'todo'; # Double quoting $FILE_PATH doesnt' change anything
if git diff -U0 $FILE_PATH | grep -iq 'todo';
then
echo $FILE_PATH ' -> Contains TODO'
exit 1
else
echo 'nooooooooooooooooooooooooooooooooooo'
fi
I'm always getting the noooooooooooooooooooo message, however the command below, tried directly on my terminal, works well:
git diff -U0 my/file/path.php | grep -iq 'todo' && echo 'true' || echo 'false'
Output
true
UPDATE
When running bash .git/hooks/pre-commit it works, very strange!!
FYI
I don't know if it's an important information but .git/hooks/pre-commit is a symbolik link
Most likely, your pipe does not return status 0. To verify that this is the case (and not the way you write your compound statement), you could rewrite it as
git diff -U0 "$FILE_PATH" | grep -iq 'todo'
grep_status=$?
echo grep status is $grep_status
if (( grep_status == 0 ))
then
echo contains todo
else
echo no
fi
I also noticed that your code contains an unnecessary semicolon in the if line. I first thought that this semicolon might cause the weird behaviour, but at least on my bash, where I tried your code, it does not seem to do any harm. Still, I would remove it for the safe side.
How can I test if a command outputs an empty string?
Previously, the question asked how to check whether there are files in a directory. The following code achieves that, but see rsp's answer for a better solution.
Empty output
Commands don’t return values – they output them. You can capture this output by using command substitution; e.g. $(ls -A). You can test for a non-empty string in Bash like this:
if [[ $(ls -A) ]]; then
echo "there are files"
else
echo "no files found"
fi
Note that I've used -A rather than -a, since it omits the symbolic current (.) and parent (..) directory entries.
Note: As pointed out in the comments, command substitution doesn't capture trailing newlines. Therefore, if the command outputs only newlines, the substitution will capture nothing and the test will return false. While very unlikely, this is possible in the above example, since a single newline is a valid filename! More information in this answer.
Exit code
If you want to check that the command completed successfully, you can inspect $?, which contains the exit code of the last command (zero for success, non-zero for failure). For example:
files=$(ls -A)
if [[ $? != 0 ]]; then
echo "Command failed."
elif [[ $files ]]; then
echo "Files found."
else
echo "No files found."
fi
More info here.
TL;DR
if [[ $(ls -A | head -c1 | wc -c) -ne 0 ]]; then ...; fi
Thanks to netj
for a suggestion to improve my original:if [[ $(ls -A | wc -c) -ne 0 ]]; then ...; fi
This is an old question but I see at least two things that need some improvement or at least some clarification.
First problem
First problem I see is that most of the examples provided here simply don't work. They use the ls -al and ls -Al commands - both of which output non-empty strings in empty directories. Those examples always report that there are files even when there are none.
For that reason you should use just ls -A - Why would anyone want to use the -l switch which means "use a long listing format" when all you want is test if there is any output or not, anyway?
So most of the answers here are simply incorrect.
Second problem
The second problem is that while some answers work fine (those that don't use ls -al or ls -Al but ls -A instead) they all do something like this:
run a command
buffer its entire output in RAM
convert the output into a huge single-line string
compare that string to an empty string
What I would suggest doing instead would be:
run a command
count the characters in its output without storing them
or even better - count the number of maximally 1 character using head -c1(thanks to netj for posting this idea in the comments below)
compare that number with zero
So for example, instead of:
if [[ $(ls -A) ]]
I would use:
if [[ $(ls -A | wc -c) -ne 0 ]]
# or:
if [[ $(ls -A | head -c1 | wc -c) -ne 0 ]]
Instead of:
if [ -z "$(ls -lA)" ]
I would use:
if [ $(ls -lA | wc -c) -eq 0 ]
# or:
if [ $(ls -lA | head -c1 | wc -c) -eq 0 ]
and so on.
For small outputs it may not be a problem but for larger outputs the difference may be significant:
$ time [ -z "$(seq 1 10000000)" ]
real 0m2.703s
user 0m2.485s
sys 0m0.347s
Compare it with:
$ time [ $(seq 1 10000000 | wc -c) -eq 0 ]
real 0m0.128s
user 0m0.081s
sys 0m0.105s
And even better:
$ time [ $(seq 1 10000000 | head -c1 | wc -c) -eq 0 ]
real 0m0.004s
user 0m0.000s
sys 0m0.007s
Full example
Updated example from the answer by Will Vousden:
if [[ $(ls -A | wc -c) -ne 0 ]]; then
echo "there are files"
else
echo "no files found"
fi
Updated again after suggestions by netj:
if [[ $(ls -A | head -c1 | wc -c) -ne 0 ]]; then
echo "there are files"
else
echo "no files found"
fi
Additional update by jakeonfire:
grep will exit with a failure if there is no match. We can take advantage of this to simplify the syntax slightly:
if ls -A | head -c1 | grep -E '.'; then
echo "there are files"
fi
if ! ls -A | head -c1 | grep -E '.'; then
echo "no files found"
fi
Discarding whitespace
If the command that you're testing could output some whitespace that you want to treat as an empty string, then instead of:
| wc -c
you could use:
| tr -d ' \n\r\t ' | wc -c
or with head -c1:
| tr -d ' \n\r\t ' | head -c1 | wc -c
or something like that.
Summary
First, use a command that works.
Second, avoid unnecessary storing in RAM and processing of potentially huge data.
The answer didn't specify that the output is always small so a possibility of large output needs to be considered as well.
if [ -z "$(ls -lA)" ]; then
echo "no files found"
else
echo "There are files"
fi
This will run the command and check whether the returned output (string) has a zero length.
You might want to check the 'test' manual pages for other flags.
Use the "" around the argument that is being checked, otherwise empty results will result in a syntax error as there is no second argument (to check) given!
Note: that ls -la always returns . and .. so using that will not work, see ls manual pages. Furthermore, while this might seem convenient and easy, I suppose it will break easily. Writing a small script/application that returns 0 or 1 depending on the result is much more reliable!
For those who want an elegant, bash version-independent solution (in fact should work in other modern shells) and those who love to use one-liners for quick tasks. Here we go!
ls | grep . && echo 'files found' || echo 'files not found'
(note as one of the comments mentioned, ls -al and in fact, just -l and -a will all return something, so in my answer I use simple ls
Bash Reference Manual
6.4 Bash Conditional Expressions
-z string
True if the length of string is zero.
-n string
string
True if the length of string is non-zero.
You can use shorthand version:
if [[ $(ls -A) ]]; then
echo "there are files"
else
echo "no files found"
fi
As Jon Lin commented, ls -al will always output (for . and ..). You want ls -Al to avoid these two directories.
You could for example put the output of the command into a shell variable:
v=$(ls -Al)
An older, non-nestable, notation is
v=`ls -Al`
but I prefer the nestable notation $( ... )
The you can test if that variable is non empty
if [ -n "$v" ]; then
echo there are files
else
echo no files
fi
And you could combine both as if [ -n "$(ls -Al)" ]; then
Sometimes, ls may be some shell alias. You might prefer to use $(/bin/ls -Al). See ls(1) and hier(7) and environ(7) and your ~/.bashrc (if your shell is GNU bash; my interactive shell is zsh, defined in /etc/passwd - see passwd(5) and chsh(1)).
I'm guessing you want the output of the ls -al command, so in bash, you'd have something like:
LS=`ls -la`
if [ -n "$LS" ]; then
echo "there are files"
else
echo "no files found"
fi
sometimes "something" may come not to stdout but to the stderr of the testing application, so here is the fix working more universal way:
if [[ $(partprobe ${1} 2>&1 | wc -c) -ne 0 ]]; then
echo "require fixing GPT parititioning"
else
echo "no GPT fix necessary"
fi
Here's a solution for more extreme cases:
if [ `command | head -c1 | wc -c` -gt 0 ]; then ...; fi
This will work
for all Bourne shells;
if the command output is all zeroes;
efficiently regardless of output size;
however,
the command or its subprocesses will be killed once anything is output.
All the answers given so far deal with commands that terminate and output a non-empty string.
Most are broken in the following senses:
They don't deal properly with commands outputting only newlines;
starting from Bash≥4.4 most will spam standard error if the command output null bytes (as they use command substitution);
most will slurp the full output stream, so will wait until the command terminates before answering. Some commands never terminate (try, e.g., yes).
So to fix all these issues, and to answer the following question efficiently,
How can I test if a command outputs an empty string?
you can use:
if read -n1 -d '' < <(command_here); then
echo "Command outputs something"
else
echo "Command doesn't output anything"
fi
You may also add some timeout so as to test whether a command outputs a non-empty string within a given time, using read's -t option. E.g., for a 2.5 seconds timeout:
if read -t2.5 -n1 -d '' < <(command_here); then
echo "Command outputs something"
else
echo "Command doesn't output anything"
fi
Remark. If you think you need to determine whether a command outputs a non-empty string, you very likely have an XY problem.
Here's an alternative approach that writes the std-out and std-err of some command a temporary file, and then checks to see if that file is empty. A benefit of this approach is that it captures both outputs, and does not use sub-shells or pipes. These latter aspects are important because they can interfere with trapping bash exit handling (e.g. here)
tmpfile=$(mktemp)
some-command &> "$tmpfile"
if [[ $? != 0 ]]; then
echo "Command failed"
elif [[ -s "$tmpfile" ]]; then
echo "Command generated output"
else
echo "Command has no output"
fi
rm -f "$tmpfile"
Sometimes you want to save the output, if it's non-empty, to pass it to another command. If so, you could use something like
list=`grep -l "MY_DESIRED_STRING" *.log `
if [ $? -eq 0 ]
then
/bin/rm $list
fi
This way, the rm command won't hang if the list is empty.
As mentioned by tripleee in the question comments , use moreutils ifne (if input not empty).
In this case we want ifne -n which negates the test:
ls -A /tmp/empty | ifne -n command-to-run-if-empty-input
The advantage of this over many of the another answers when the output of the initial command is non-empty. ifne will start writing it to STDOUT straight away, rather than buffering the entire output then writing it later, which is important if the initial output is slowly generated or extremely long and would overflow the maximum length of a shell variable.
There are a few utils in moreutils that arguably should be in coreutils -- they're worth checking out if you spend a lot of time living in a shell.
In particular interest to the OP may be dirempty/exists tool which at the time of writing is still under consideration, and has been for some time (it could probably use a bump).