how to use sed, awk, or gawk to print only what is matched?

I see lots of examples and man pages on how to do things like search-and-replace using sed, awk, or gawk.

But in my case, I have a regular expression that I want to run against a text file to extract a specific value. I don't want to do search-and-replace. This is being called from bash. Let's use an example:

Example regular expression:

.*abc([0-9]+)xyz.*

Example input file:

a
b
c
abc12345xyz
a
b
c

As simple as this sounds, I cannot figure out how to call sed/awk/gawk correctly. What I was hoping to do, is from within my bash script have:

myvalue=$( sed <...something...> input.txt )

Things I've tried include:

sed -e 's/.*([0-9]).*/\\1/g' example.txt # extracts the entire input file
sed -n 's/.*([0-9]).*/\\1/g' example.txt # extracts nothing

My sed (Mac OS X) didn't work with +. I tried * instead and I added p tag for printing match:

sed -n 's/^.*abc\([0-9]*\)xyz.*$/\1/p' example.txt

For matching at least one numeric character without +, I would use:

sed -n 's/^.*abc\([0-9][0-9]*\)xyz.*$/\1/p' example.txt

You can use sed to do this

 sed -rn 's/.*abc([0-9]+)xyz.*/\1/gp'

• -n don't print the resulting line
• -r this makes it so you don't have the escape the capture group parens().
• \1 the capture group match
• /g global match
• /p print the result

I wrote a tool for myself that makes this easier

rip 'abc(\d+)xyz' '$1'

I use perl to make this easier for myself. e.g.

perl -ne 'print $1 if /.*abc([0-9]+)xyz.*/'

This runs Perl, the -n option instructs Perl to read in one line at a time from STDIN and execute the code. The -e option specifies the instruction to run.

The instruction runs a regexp on the line read, and if it matches prints out the contents of the first set of bracks ($1).

You can do this will multiple file names on the end also. e.g.

perl -ne 'print $1 if /.*abc([0-9]+)xyz.*/' example1.txt example2.txt

If your version of grep supports it you could use the -o option to print only the portion of any line that matches your regexp.

If not then here's the best sed I could come up with:

sed -e '/[0-9]/!d' -e 's/^[^0-9]*//' -e 's/[^0-9]*$//'

... which deletes/skips with no digits and, for the remaining lines, removes all leading and trailing non-digit characters. (I'm only guessing that your intention is to extract the number from each line that contains one).

The problem with something like:

sed -e 's/.*\([0-9]*\).*/&/' 

.... or

sed -e 's/.*\([0-9]*\).*/\1/'

... is that sed only supports "greedy" match ... so the first .* will match the rest of the line. Unless we can use a negated character class to achieve a non-greedy match ... or a version of sed with Perl-compatible or other extensions to its regexes, we can't extract a precise pattern match from with the pattern space (a line).

You can use awk with match() to access the captured group:

$ awk 'match($0, /abc([0-9]+)xyz/, matches) {print matches[1]}' file
12345

This tries to match the pattern abc[0-9]+xyz. If it does so, it stores its slices in the array matches, whose first item is the block [0-9]+. Since match() returns the character position, or index, of where that substring begins (1, if it starts at the beginning of string), it triggers the print action.

---

With grep you can use a look-behind and look-ahead:

$ grep -oP '(?<=abc)[0-9]+(?=xyz)' file
12345

$ grep -oP 'abc\K[0-9]+(?=xyz)' file
12345

This checks the pattern [0-9]+ when it occurs within abc and xyz and just prints the digits.

perl is the cleanest syntax, but if you don't have perl (not always there, I understand), then the only way to use gawk and components of a regex is to use the gensub feature.

gawk '/abc[0-9]+xyz/ { print gensub(/.*([0-9]+).*/,"\\1","g"); }' < file

output of the sample input file will be

12345

Note: gensub replaces the entire regex (between the //), so you need to put the .* before and after the ([0-9]+) to get rid of text before and after the number in the substitution.

If you want to select lines then strip out the bits you don't want:

egrep 'abc[0-9]+xyz' inputFile | sed -e 's/^.*abc//' -e 's/xyz.*$//'

It basically selects the lines you want with egrep and then uses sed to strip off the bits before and after the number.

You can see this in action here:

pax> echo 'a
b
c
abc12345xyz
a
b
c' | egrep 'abc[0-9]+xyz' | sed -e 's/^.*abc//' -e 's/xyz.*$//'
12345
pax> 

Update: obviously if you actual situation is more complex, the REs will need to me modified. For example if you always had a single number buried within zero or more non-numerics at the start and end:

egrep '[^0-9]*[0-9]+[^0-9]*$' inputFile | sed -e 's/^[^0-9]*//' -e 's/[^0-9]*$//'

The OP's case doesn't specify that there can be multiple matches on a single line, but for the Google traffic, I'll add an example for that too.

Since the OP's need is to extract a group from a pattern, using grep -o will require 2 passes. But, I still find this the most intuitive way to get the job done.

$ cat > example.txt <<TXT
a
b
c
abc12345xyz
a
abc23451xyz asdf abc34512xyz
c
TXT

$ cat example.txt | grep -oE 'abc([0-9]+)xyz'
abc12345xyz
abc23451xyz
abc34512xyz

$ cat example.txt | grep -oE 'abc([0-9]+)xyz' | grep -oE '[0-9]+'
12345
23451
34512

Since processor time is basically free but human readability is priceless, I tend to refactor my code based on the question, "a year from now, what am I going to think this does?" In fact, for code that I intend to share publicly or with my team, I'll even open man grep to figure out what the long options are and substitute those. Like so: grep --only-matching --extended-regexp

why even need match group

gawk/mawk/mawk2 'BEGIN{ FS="(^.*abc|xyz.*$)" } ($2 ~ /^[0-9]+$/) {print $2}'

Let FS collect away both ends of the line.

If $2, the leftover not swallowed by FS, doesn't contain non-numeric characters, that's your answer to print out.

If you're extra cautious, confirm length of $1 and $3 both being zero.

** edited answer after realizing zero length $2 will trip up my previous solution

there's a standard piece of code from awk channel called "FindAllMatches" but it's still very manual, literally, just long loops of while(), match(), substr(), more substr(), then rinse and repeat.

If you're looking for ideas on how to obtain just the matched pieces, but upon a complex regex that matches multiple times each line, or none at all, try this :

mawk/mawk2/gawk 'BEGIN { srand(); for(x = 0; x < 128; x++ ) { 

    alnumstr = sprintf("%s%c", alnumstr , x) 
 }; 
 gsub(/[^[:alnum:]_=]+|[AEIOUaeiou]+/, "", alnumstr) 
                       
                    # resulting str should be 44-chars long :
                    # all digits, non-vowels, equal sign =, and underscore _

 x = 10; do { nonceFS = nonceFS substr(alnumstr, 1 + int(44*rand()), 1)

 } while ( --x );   # you can pick any level of precision you need.
                    # 10 chars randomly among the set is approx. 54-bits 
                    #
                    # i prefer this set over all ASCII being these 
                    # just about never require escaping 
                    # feel free to skip the _ or = or r/t/b/v/f/0 if you're concerned.
                    #
                    # now you've made a random nonce that can be 
                    # inserted right in the middle of just about ANYTHING
                    # -- ASCII, Unicode, binary data -- (1) which will always fully
                    # print out, (2) has extremely low chance of actually
                    # appearing inside any real word data, and (3) even lower chance
                    # it accidentally alters the meaning of the underlying data.
                    # (so intentionally leaving them in there and 
                    # passing it along unix pipes remains quite harmless)
                    #
                    # this is essentially the lazy man's approach to making nonces
                    # that kinda-sorta have some resemblance to base64
                    # encoded, without having to write such a module (unless u have
                    # one for awk handy)


    regex1 = (..);  # build whatever regex you want here

    FS = OFS = nonceFS;

 } $0 ~ regex1 { 

    gsub(regex1, nonceFS "&" nonceFS); $0 = $0;  

                   # now you've essentially replicated what gawk patsplit( ) does,
                   # or gawk's split(..., seps) tracking 2 arrays one for the data
                   # in between, and one for the seps.
                   #
                   # via this method, that can all be done upon the entire $0,
                   # without any of the hassle (and slow downs) of 
                   # reading from associatively-hashed arrays,
                   # 
                   # simply print out all your even numbered columns
                   # those will be the parts of "just the match"

if you also run another OFS = ""; $1 = $1; , now instead of needing 4-argument split() or patsplit(), both of which being gawk specific to see what the regex seps were, now the entire $0's fields are in data1-sep1-data2-sep2-.... pattern, ..... all while $0 will look EXACTLY the same as when you first read in the line. a straight up print will be byte-for-byte identical to immediately printing upon reading.

Once i tested it to the extreme using a regex that represents valid UTF8 characters on this. Took maybe 30 seconds or so for mawk2 to process a 167MB text file with plenty of CJK unicode all over, all read in at once into $0, and crank this split logic, resulting in NF of around 175,000,000, and each field being 1-single character of either ASCII or multi-byte UTF8 Unicode.

you can do it with the shell

while read -r line
do
    case "$line" in
        *abc*[0-9]*xyz* ) 
            t="${line##abc}"
            echo "num is ${t%%xyz}";;
    esac
done <"file"

For awk. I would use the following script:

/.*abc([0-9]+)xyz.*/ {
			print $0;
			next;
			}
			{
			/* default, do nothing */
			}

gawk '/.*abc([0-9]+)xyz.*/' file