Shell – Why two almost idetical grep commands return different output: w/o and with filename

globgrepregular expressionshell

I have 2 almost identical greps:

[Alex@localhost tmp]$ grep /bin/bash /etc/passwd 
root:x:0:0:root:/root:/bin/bash
AlexL:x:500:500::/home/AlexL:/bin/bash
user1:x:501:501:user1 12345:/home/user1:/bin/bash

vs.

[AlexL@localhost tmp]$ grep /bin/*sh /etc/passwd    
/etc/passwd:root:x:0:0:root:/root:/bin/bash
/etc/passwd:AlexL:x:500:500::/home/AlexL:/bin/bash
/etc/passwd:user1:x:501:501:user1 12345:/home/user1:/bin/bash

In 2nd query I got filename prefix for every line match.
I know that to get the same result I need to put -h option in 2nd grep, but my question is:

Why do they return different output?
Should I add something more?

The task consist of retrieving from /etc/passwd real users (w/o daemons and system users).

Used: CentOS 6.4, grep gnu 2.6.3 version

Best Answer

Your shell expands /bin/*sh. So, what you are really doing is

grep /bin/bash /bin/dash /bin/rbash /bin/rzsh /bin/sh /bin/zsh /etc/passwd

That is, search for the string /bin/bash in the files /bin/dash, /bin/rbash, /bin/rzsh, /bin/sh, /bin/zsh and /etc/passwd.

(Compare with the output of echo /bin/*sh /etc/passwd.)

Since there are several files to search in, grep reports which one it found the string in.

What you want is to quote your search term, so it isn't expanded by the shell, and to use a proper regular expression:

grep '/bin/.*sh' /etc/passwd

Related Solutions

How to combine grep with multiple arguments and different output switches

In bash, ksh or zsh:

sort -gum <(grep -nA5 onestring foo.txt) <(grep -nB5 otherstring foo.txt)
# Sort by general numbers, make output unique and merge sorted files,
# where files are expanded as a result of shell's command expansion,
# FIFOs/FDs that gives the command's output

This takes O(n) time, given that grep outputs already sorted things. If process substitution is unavailable, either create temp files manually or use the O(n lgn) ( grep -nA5 onestring foo.txt; grep -nB5 otherstring foo.txt ) | sort -gu.

With grep -H we need to sort it in a more verbose way (thanks to cas):

# FIXME: I need to figure out how to deal with : in filenames then.
# Use : as separator, the first field using the default alphabetical sort, and
# 2nd field using general number sort.
sort -t: -f1,2g -um <(grep -nA5 onestring foo.txt bar.txt) <(grep -nB5 otherstring foo.txt bar.txt)

Why does grep sometimes return directories with two slashes

When you run the command:

grep -r <pattern> <directory>

it prints all the filenames as <directory>/<filename>. If you put a slash at the end of <directory>, it will be included in that, so you end up with a double slash.

The code could have been smart enough to notice when the original directory name ends with / and omit it when printing filenames, but they didn't bother. I think the only special case it makes is for the root directory, e.g.

grep -r pattern /

will print /foo/bar/baz rather than //foo/bar/baz. This is because POSIX specifies that a sequence of slashes is treated equivalently to a single slash except that two slashes at the beginning have implementation-dependent meaning (this is because some network file system mechanisms made use of syntax like //server/pathname). See How does Linux handle multiple consecutive path separators (/home////username///file)?

Best Answer

Related Solutions

How to combine grep with multiple arguments and different output switches

Why does grep sometimes return directories with two slashes

Related Question