grep pattern from file, print the pattern instead matched string

Question

I want to grep with patterns from file containing regex. When the pattern matches, it prints the matched stringa but not the pattern. How can I get the pattern instead matched strings?

pattern.txt

Apple (Ball|chocolate|fall) Donut
donut (apple|ball) Chocolate
Donut Gorilla Chocolate
Chocolate (English|Fall) apple gorilla
gorilla chocolate (apple|ball)
(ball|donut) apple

strings.txt

apple ball Donut
donut ball chocolate
donut Ball Chocolate
apple donut
chocolate ball Apple

This is grep command

grep -Eix -f pattern.txt strings.txt

This command prints matched strings from strings.txt

apple ball Donut
donut ball chocolate
donut Ball Chocolate

But I want to find which patterns were used to match from pattern.txt

Apple (Ball|chocolate|fall) Donut
donut (apple|ball) Chocolate

The pattern.txt can be lower cases, upper cases, line with regex and without, free numbers of words and regex elements. There is no other kind of regex than brackets and pipe.

I don't want to use loop to read pattern.txt each line to grep as it's slow. Is there way to print which pattern or line number of pattern file in grep command? or any other command than grep can do the job not too slow?

Answer 1

Using grep I have no idea but with GNU awk:

$ awk '
BEGIN { IGNORECASE = 1 }      # for case insensitivity
NR==FNR {                     # process pattern file
    a[$0]                     # hash the entries to a
    next                      # process next line
}
{                             # process strings file
    for(i in a)               # loop all pattern file entries
        if($0 ~ "^" i "$") {  # if there is a match (see comments)
            print i           # output the matching pattern file entry
            # delete a[i]     # uncomment to delete matched patterns from a
            # next            # uncomment to end searching after first match
        }
}' pattern strings

outputs:

D (A|B) C

For each line in strings script will loop every pattern line to see if there are more than one match. There is only one match due to case-sensitivity. You can battle that, for example, using GNU awk's IGNORECASE.

Also, if you want each matched one pattern file entry to be outputed once, you could delete them from a after first match: add delete a[i] after the print. That might give you some performance advantage also.

Answer 2

EDIT: Since OP changed the Input_file(s) so adding solutions as per changed Input_file(s) too now.

awk '
FNR==NR{
   a[toupper($1),toupper($NF)]
   b[toupper($2)]
   next
}
{
   val=toupper($2)
   gsub(/\)|\(|\|/," ",val)
   num=split(val,array," ")
   for(i=1;i<=num;i++){
      if(array[i] in b){
        flag=1
        break
      }
   }
}
flag && ((toupper($1),toupper($NF)) in a){
  print;
  flag=""
}' string pattern

Output will be as follows.

Apple (Ball|chocolate|fall) Donut
donut (apple|ball) Chocolate

---

---

Solution 1st: Adding a generic solution where let's say your Input_file named pattern have more than 2 values on 2nd field eg--> (B|C|D|E) then following may help you here.

awk '
FNR==NR{
   a[$1,$NF]
   b[toupper($2)]
   next
}
{
   val=$2
   gsub(/\)|\(|\|/," ",val)
   num=split(val,array," ")
   for(i=1;i<=num;i++){
      if(array[i] in b){
        flag=1
        break
      }
   }
}
flag && (($1,$NF) in a)
{
  flag=""
}' string pattern

---

---

Solution 2nd: Could you please try following. But strictly considering that your Input_file(s) are same pattern as per shown samples only(where I am considering that your Input_file named pattern will have only 2 values in 2nd field of it)

awk '
FNR==NR{
  a[$1,$NF]
  b[toupper($2)]
  next
}
{
  val=$2
  gsub(/\)|\(|\|/," ",val)
  split(val,array," ")
}
((array[1] in b) || (array[2] in b)) && (($1,$NF) in a)
' string pattern

Output will be as follows.

A (B|C) D
D (A|B) C

Answer 3

You could try with bash built-ins:

$ cat foo.sh
#!/usr/bin/env bash

# case insensitive
shopt -s nocasematch

# associative array of patterns
declare -A patterns=()
while read -r p; do
    patterns["$p"]=1
done < pattern.txt

# read strings, test remaining patterns,
# if match print pattern and remove it from array    
while read -r s; do
    for p in "${!patterns[@]}"; do
        if [[ $s =~ ^$p$ ]]; then
            printf "%s\n" "$p"
            unset patterns["$p"]
        fi
    done
done < strings.txt
$ ./foo.sh
Apple (Ball|chocolate|fall) Donut
donut (apple|ball) Chocolate

Not sure about the performance but as there are no child processes, it should be much faster than invoking grep for each pattern.

Of course, if you have millions of patterns, storing them in an associative array could exhaust your available memory.

Answer 4

Maybe switch the paradigm?

while read pat
do grep -Eix "$pat" strings.txt >"$pat" &
done <patterns.txt

That's going to make ugly filenames, but you'd have clear lists per set. You could scrub the filenames first if you prefer. Maybe (assuming the patterns resolve to uniqueness this easily...)

while read pat
do grep -Eix "$pat" strings.txt >"${pat//[^A-Z]/}" &
done <patterns.txt

It ought to be reasonably quick, and is relatively simple to implement. Hope that helps.