How to remove characters from a word if they are also in the next word (sed)? - linux

Question

I'm trying to find a way to delete all characters in the first word IF that character is in the second word. The input looks like this:
computer cost
And the result should be: "mpuer" because the c, o and t were deleted. There are multiple lines like this separated by a return, the 2 words are separated by a space.
I've been searching quite some time for the solution, but I'm really stuck. All help is appreciated.

Answer 1

This might work for you:
echo "computer cost" |
sed ':a;s/\(.\)\(.* .*\1.*\)/\2/;ta;s/ .*//'
mpuer
Explanation:
Make a label for future branch command :a;
Delete a character in the first word that matches with the same character in the second word s/\(.\)\(.* .*\1.*\)/\2/
If the substitution occurred branch to label ta
When no more substitutions delete the second word. s/ .*//
The substitution regexp can be further explained:
\(.\) matches any character in word one (later refered to as \1)
\(.* .*\1.*\) matches any characters in the remainder of a word one .* followed by a space followed by some on none characters in word two .* followed by a matching character from word one \1 followed by the remaining characters from word two .* this grouping will later be known as \2.
If the above matches replace it by \2 thus effectively deleting the matching character \1

Answer 2

This works (as does the solution by potong):
sed -e ': loop' \
-e 's/\([a-z]*\)\([a-z]\)\([a-z]*\) \([a-z]*\2[a-z]*\)/\1\3 \4/' \
-e 't loop' \
-e 's/ .*//' \
"$#"
The first line establishes a label. The third line branches to the label if there's been a successful substitute since the line was read and the last time the t was executed, so that establishes a loop while the substitute command finds something to do. The last line removes the word after the space once the loop is complete.
All eyes concentrate on the regexes, now. The key insight is that you can look for a repeat of a remembered pattern later in the string using \n where n is a digit. The first part of the regex partitions the line into 5 pieces. The first part is a (possibly empty) sequence of letters that aren't interesting; the second is a single letter that is interesting; the third is another (possibly empty) sequence of letters that aren't interesting; the fourth is the space separating the first word from the second. The final part can itself be subdivided into 3 parts, though they are all grouped together into a single capture expression. It consists of a sequence of zero or more uninteresting letters, a repeat of the interesting letter from the first word on the line (the \2), and another sequence of zero or more non-interesting letters.
The replacement string keeps the before and after parts of the first word, plus the space and the second word.
In combination, it finds each of the letters c, o and t in turn, eliminating them from the first word and leaving them alone in the second.
The conditional branching in sed is hard to use, but it can really score on occasion. When your hands are tied by the assignment like this, it makes the solution feasible.
$ al 'computer cost' 'encyclopedia brittanica' 'security privacy' |
> sed -e ': loop; s/\([a-z]*\)\([a-z]\)\([a-z]*\) \([a-z]*\2[a-z]*\)/\1\3 \4/; t loop'
mpuer
eyloped
seut
$
al simply lists its arguments one per line - hence the mnemonic Argument List:
#include <stdio.h>
int main(int argc, char **argv)
{
while (*++argv)
puts(*argv);
return 0;
}
Potong's solution is essentially equivalent to a 'Code Golf' version of mine:
sed ':a;s/\(.\)\(.* .*\1.*\)/\2/;ta;s/ .*//'
It uses the same general technique that mine does, but simplifies the regex. One simplification is the use of . (any character) in place of [a-z] (any letter). Another is to realize that the leading pattern doesn't matter; it will be left alone. The last is to group the tail of the first word with the whole of the second. In retrospect, I could (should?) have added a ^ anchor to my pattern. Potong's label is simply a.

Answer 3

Basically you do this by tr;
echo computer cost | while read x y;do echo $x | tr -d $y ; done;
if you have a file (words) like
computer cost
computer mop
Following command will do the replacement.
while read x y; do echo $x | tr -d $y ; done< words
If you want to use sed just replace tr -d $y with sed s/[$y]//g