Can grep show only words that match search pattern?

Question

Is there a way to make grep output "words" from files that match the search expression?

If I want to find all the instances of, say, "th" in a number of files, I can do:

grep "th" *

but the output will be something like (bold is by me);

some-text-file : the cat sat on the mat  
some-other-text-file : the quick brown fox  
yet-another-text-file : i hope this explains it thoroughly 

What I want it to output, using the same search, is:

the
the
the
this
thoroughly

Is this possible using grep? Or using another combination of tools?

Solution 1

Try grep -o:

grep -oh "\w*th\w*" *

Edit: matching from Phil's comment.

From the docs:

-h, --no-filename
    Suppress the prefixing of file names on output. This is the default
    when there is only  one  file  (or only standard input) to search.
-o, --only-matching
    Print  only  the matched (non-empty) parts of a matching line,
    with each such part on a separate output line.

Solution 2

Cross distribution safe answer (including windows minGW?)

grep -h "[[:alpha:]]*th[[:alpha:]]*" 'filename' | tr ' ' '\n' | grep -h "[[:alpha:]]*th[[:alpha:]]*"

If you're using older versions of grep (like 2.4.2) which do not include the -o option, then use the above. Else use the simpler to maintain version below.

Linux cross distribution safe answer

grep -oh "[[:alpha:]]*th[[:alpha:]]*" 'filename'

To summarize: -oh outputs the regular expression matches to the file content (and not its filename), just like how you would expect a regular expression to work in vim/etc... What word or regular expression you would be searching for then, is up to you! As long as you remain with POSIX and not perl syntax (refer below)

More from the manual for grep

-o      Print each match, but only the match, not the entire line.
-h      Never print filename headers (i.e. filenames) with output lines.
-w      The expression is searched for as a word (as if surrounded by
         `[[:<:]]' and `[[:>:]]';

---

The reason why the original answer does not work for everyone

The usage of \w varies from platform to platform, as it's an extended "perl" syntax. As such, those grep installations that are limited to work with POSIX character classes use [[:alpha:]] and not its perl equivalent of \w. See the Wikipedia page on regular expression for more

Ultimately, the POSIX answer above will be a lot more reliable regardless of platform (being the original) for grep

As for support of grep without -o option, the first grep outputs the relevant lines, the tr splits the spaces to new lines, the final grep filters only for the respective lines.

(PS: I know most platforms by now would have been patched for \w.... but there are always those that lag behind)

Credit for the "-o" workaround from @AdamRosenfield answer

Solution 3

It's more simple than you think. Try this:

egrep -wo 'th.[a-z]*' filename.txt #### (Case Sensitive)

egrep -iwo 'th.[a-z]*' filename.txt  ### (Case Insensitive)

Where,

 egrep: Grep will work with extended regular expression.
 w    : Matches only word/words instead of substring.
 o    : Display only matched pattern instead of whole line.
 i    : If u want to ignore case sensitivity.

Solution 4

You could translate spaces to newlines and then grep, e.g.:

cat * | tr ' ' '\n' | grep th

Solution 5

Just awk, no need combination of tools.

# awk '{for(i=1;i<=NF;i++){if($i~/^th/){print $i}}}' file
the
the
the
this
thoroughly

View more solutions

Comments

[hakish]

Dan Midwood solution works perfectly and deserves the credit.

[Linguist]

Is there a way one can print those matched words without changing the lines. Rather the matched string should remain in the same line?

[ghostdog74]

that won't give the correct result. also, if using Perl, no need to use grep. do everything in Perl.

[Admin]

Thanks for pointing out the error, ghostdog74. I have changed it to print all the words on the line, not just the first.

[ghostdog74]

like i said, grep is not necessary. perl -n -e'while(/(\s+th\w*)/g) {print "$1\n"}' file

[Admin]

I don't think it's important here to avoid using grep.

[ghostdog74]

no need cat. tr ' ' '\n' < file | grep th. Slow for big files.

[ghostdog74]

up to you. i am just illustrating a point. If its not necessary, don't do it. that extra "|" will cost you one process more.

[Neil Baldwin]

This didn't work. The output still contained the filename and the entire line from the file that contained the match. Anyway, one of the other solutions offered worked. Thanks for the input though.

[Adam Rosenfield]

@ghostdog74: good point, although if you have more than file, you'll need to use cat. @Neil Baldwin: are you sure you typed it in right? When there's only one input file (stdin in this case), grep doesn't print the filename.

[Neil Baldwin]

@Adam - yes, sorry Adam, it does work with one file but not multiple.

[Adam Rosenfield]

@Neil Baldwin: just list all of your files as parameters to cat, it works fine with multiple files

[Neil Baldwin]

@Adam - so where you've got 'file' in the example, I would just put 'file1 file2 file3' etc. ?

[tripleee]

That will still print the entire line containing the match. It constrains the actual match so that the no longer matches e.g. "these" or "bathe".

[ksinkar]

@user181548, The grep -o option works only for GNU grep. So if you are not using GNU grep, it might not work for you.

[Brilliand]

What about -o only working in GNU grep (as ksinkar mentioned in a comment on the accepted answer)?

[PicoCreator]

@Brilliand hmm, im having trouble finding a linux implementation that does not support '-o', i can look for a work around if i know which platform to check against.

[Bruce Peterson]

@pico The -o option is not present in the windows grep that installs with the git package (minGW?): "c:\Program Files (x86)\Git\bin\grep" --version grep (GNU grep) 2.4.2

[PicoCreator]

@BrucePeterson i have added in AdamRosenfield workaround answer for -o : Help me check if the windows git includes tr / sed and its version. So i can check if this workaround works

[Bruce Peterson]

@pico: for GIT: GNU sed version 4.2.1, tr (GNU textutils) 2.0

[Shayan]

or just grep -Eio "th[a-z]+" filename

[Carcamano]

@ghostdog74 if the slow part is because of tr, he could do grep first, so tr would be applied only to matching lines: grep th filename | tr ' ' '\n' | grep th

[tripleee]

The useless {1} quantifiers should be dropped. Or if you want to be consistent, t{1}h{1}e{1} etc.

[Professor Photon]

In Perl 5.10 or later: perl -nE '@a = /(regexp)/ig; say join "\n", @a'

[ife]

can it print with the same line?

[Collin Anderson]

-o is not valid in linux git either

[LokMac]

@A-B-B It depends if you want to display the name of the matched file or not. I'm not sure under what conditions it does and doesn't display, but I do know that when I used grep across a number of directories it did display the full file path for all matched files, whereas with -h it just displayed the matched words without any specification about which file it is. So, to match the original question, I think it is necessary in certain circumstances.

[Bishwas Mishra]

What about display of only the matched group?

[jeremysprofile]

I needed an explanation for what "\w*th\w*" * means, so I figured I'd post. \w is [_[:alnum:]], so this matches basically any "word" that contains 'th' (since \w doesn't include space). The * after the quoted section is a glob for which files (i.e., matching all files in this directory)

[tripleee]

\w is not generally portable to grep -E; for proper portability, use the POSIX character class name [[:alnum:]] instead (or [_[:alnum:]] if you really want the underscore, too; or try grep -P if your platform has that).

[tripleee]

This doesn't seem to add anything over the existing answers from 4+ years before.

[tripleee]

@CollinAnderson Your comment doesn't really make sense. GNU grep and thus pretty much every Linux box has grep -o; there is no -o option in Git itself, but many Windows victims install a gitpackage which includes many Unix utilities, including a grep implementation.

[tripleee]

@BrucePeterson If you genuinely have GNU grep 2.4.2 then it's frightfully old; the -o option was introduced in 2.5.1 sometime in 2001

[tripleee]

Maybe see also Useless use of cat?

[tripleee]

This doesn't work; it will only ever find th because you requested the shortest possible repetition of the wildcard.

[Collin Anderson]

@tripleee I meant git grep

[Ken Williams]

@tripleee - it won't have that problem, because there's a space included at the end of the regex. However, it will miss words that don't have spaces after them, e.g. at the ends of lines.

[El Ronnoco]

@A-B-B Given the desired output shown by the OP the -h is entirely necessary I would say.. ?

[Abhinandan prasad]

@tripleee I found my approach is better and simple so I posted this.

[neverMind9]

Much better than abk.

[Nathan McKaskle]

-o only gives me exactly what I searched for, I need the whole line. Wtf? It's like trying to work with a genie that takes everything too literally.

[econometrica_33]

I'm not sure why this answer is downvoted. I was using ripgrep looking for the answer to the same question and simply by adding the -o option I got exactly as desired.

[Kasthuri Shravankumar]

tac file.log | grep "In msg::" | grep -oh "templateId=.*, temp"